A Review of Novel Approaches to Indoor Localization Based on Wireless Capsule Endoscopy

A Review of Novel Approaches to Indoor Localization Based on Wireless Capsule Endoscopy | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article A Review of Novel Approaches to Indoor Localization Based on Wireless Capsule Endoscopy Zeinab Javid, Michel Kadoch This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8896919/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 4 You are reading this latest preprint version Abstract Wireless Capsule Endoscopy (WCE) has emerged as a minimally invasive diagnostic modality for gastrointestinal disorders, offering distinct advantages over conventional endoscopy. Despite clinical adoption, current WCE systems remain limited in energy efficiency, data transmission, lesion-detection accuracy, and reliable localization. These constraints hinder its full potential as a smart, autonomous diagnostic platform. This review synthesizes recent advancements at the intersection of artificial intelligence (AI), the Internet of Things (IoT), and compressed sensing (CS) as key enablers for next-generation WCE. AI-driven methods, particularly convolutional neural networks, have demonstrated near-human accuracy in lesion detection while significantly reducing diagnostic review times. IoT-enabled frameworks enhance connectivity, support remote monitoring, and integrate capsule data into broader digital health infrastructures. Meanwhile, compressed sensing techniques and application-specific integrated circuits (ASICs) improve transmission efficiency and extend battery life without compromising image quality. The paper further examines ongoing challenges in localization, where visual odometry, magnetic tracking, and hybrid multisensory fusion continue to evolve but remain short of clinical reliability. By consolidating these perspectives, the review highlights how cross-disciplinary integration is reshaping WCE from a passive imaging tool into an intelligent, multifunctional platform. However, significant translational gaps remain between experimental prototypes and clinical practice, underscoring the need for multidisciplinary collaboration. The convergence of AI, IoT, and CS not only addresses current bottlenecks but also paves the way for capsule systems capable of autonomous navigation, advanced diagnostics, and therapeutic intervention. Wireless Capsule Endoscopy Artificial Intelligence Internet of Things Compressed Sensing Indoor Localization Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 1. Introduction Gastrointestinal (GI) disorders represent a significant global health burden, with the Global Burden of Disease Study reporting that in 2019 alone, digestive diseases caused over 1.3 million deaths and accounted for more than 245 million incident cases worldwide. Beyond mortality, these conditions are also among the leading contributors to Disability-Adjusted Life Years (DALYs), imposing substantial healthcare costs and reducing workforce productivity [1]. These alarming figures underscore the urgent need for non-invasive, efficient, and patient-friendly diagnostic solutions. Capsule Endoscopy (CE) has transformed small bowel imaging by providing a wireless, less invasive alternative to conventional endoscopic procedures. The potential of CE is further amplified through integration with advanced technologies such as Artificial Intelligence (AI), the Internet of Things (IoT), and Compressed Sensing (CS), which collectively enhance diagnostic accuracy, connectivity, and energy efficiency. The market growth trend is shown in Fig. 1 , based on Grand View Research’s report [2]. Wireless Capsule Endoscopy (WCE) is specifically designed to evaluate Small Bowel (SB) disorders that are often inaccessible with conventional endoscopy [3], [4]. Its ease of use—requiring neither auxiliary tools nor sedation—minimizes patient discomfort and enables visualization of otherwise unreachable GI segments [5]. Recent advances in capsule design, such as multi-camera systems that provide 360° imaging, have expanded diagnostic coverage [3], [4]. In parallel with hardware innovations, AI-driven tools such as Express View and Suspected Blood Indicator are increasingly integrated into wireless capsule endoscopy systems [6], [7]. A single capsule can generate more than 60,000 images during transit, and convolutional neural networks have achieved detection sensitivities approaching 99.9%, reducing physician review times from nearly 100 minutes to just a few minutes [8]. Despite these advances, a persistent unmet need is highlighted in recent reviews: clinicians still face 30–120 minutes of video analysis per patient, with fatigue and diagnostic variability reducing efficiency [9], [10]. In addition, Ultra-Wide-Band (UWB) communication has been investigated to support high-volume data transmission at rates exceeding 100 Mb/s while improving energy efficiency [7]. Managing the large volume of imaging data remains a challenge in WCE, requiring efficient compression algorithms that reduce data volume without degrading image quality [11–13]. Designers must balance image resolution and frame rate with limited hardware resources and battery capacity[14], [15]. Application-Specific Integrated Circuits ASICs) have emerged as a key solution, enabling optimized image processing, compression, and signal transmission at minimal energy cost. Such developments are critical to improving capsule autonomy while maintaining diagnostic accuracy [16]. Artificial Intelligence (AI) has further advanced automated lesion detection in WCE. CNN-based models have demonstrated high sensitivity in identifying abnormalities while significantly reducing physicians' time burden [8]. Automated bleeding detection has been a particular focus, as rapid identification is crucial to prevent adverse outcomes [7]. Nonetheless, AI adoption in clinical practice remains limited by professional resistance, the lack of standardized diagnostic protocols, and insufficient validation across diverse populations [8], [17]. These limitations reflect the broader challenge that, while AI excels in sensitivity, it still falls short of replacing expert decision-making in real-world workflows [9], [10]. Accurate localization of the capsule is another critical requirement for clinical translation. Visual Odometry methods estimate position by tracking feature-point motion across consecutive frames [18], whereas magnetic-based localization uses miniature internal magnets that are tracked by external sensors [7]. These approaches support motion trajectory reconstruction and even three-dimensional mapping of the gastrointestinal tract, with reported localization errors as low as 3.5 mm [19]. However, state-of-the-art reviews emphasize that no existing technique can consistently guarantee sub-5 mm positional error or < 5° orientation error in vivo [20], while technical analyses highlight that hybrid multi-sensor fusion offers a promising solution, though it remains largely experimental. Hybrid systems that combine visual, magnetic, and Radio-Frequency (RF) data, along with multi-sensor fusion, have therefore become a primary research direction [19]. The integration of capsule-based sensors has further expanded diagnostic and monitoring capabilities. These devices incorporate sensing elements, microcontrollers, RF transmission units, and power sources, enabling real-time physiological monitoring and communication [21]. Some capsules can transmit data directly to smartphones via Bluetooth, while advanced RF modules ensure reliable connectivity with external receivers [16]. By leveraging mobile cloud computing and 5G technologies, WCE platforms are poised to support remote monitoring and personalized healthcare. To address energy limitations, research has focused on wireless power transfer—particularly near-field systems—that deliver contactless energy to capsules, as well as self-powered designs that harvest chemical or mechanical energy [21]. In addition, edible electronic components fabricated from biocompatible materials provide safe, degradable options for in-body applications [7]. Despite rapid progress, miniaturization, energy efficiency, and biocompatibility remain persistent challenges. Many laboratory prototypes still struggle to integrate all essential functionalities into a safe, swallowable capsule, limiting their translation to large-scale in vivo use [21], [22]. Addressing these constraints will be essential to enable multifunctional capsules that combine imaging, sensing, actuation, and communication within clinically viable dimensions. AI and IoT are expected to play a central role in the next generation of capsule robots. Although AI has improved diagnostic accuracy, it cannot yet fully replace expert endoscopists in patient-specific decision-making [18]. Clinical translation is further challenged by high costs, professional resistance, and limited availability of standardized diagnostic protocols [8], [17]. The absence of large, high-quality datasets, coupled with the “black box” nature of many algorithms, also hampers trust and adoption. Nonetheless, capsule robots highlight the potential of AI to enhance precision, personalization, and real-time navigation [5], [21]. Ethical considerations such as accountability, equitable access, and overreliance on automated tools must also be addressed [17]. To complement clinical expertise effectively, future research should prioritize hardware-integrated, energy-efficient AI systems capable of reliable real-world performance [22], [23]. Another major barrier is the enormous volume of video data generated during WCE procedures. Reviewing such data is time-intensive, and accurate localization after ingestion remains technically challenging. While tools such as bleeding detection and depth estimation can provide guidance, most deep learning models are too computationally heavy to be implemented directly within capsules. Recent optimizations, however, have enabled compact models capable of onboard bleeding segmentation and depth estimation, marking a significant step toward real-time in situ analysis [24]. Hybrid localization strategies that fuse complementary sensing technologies are also under investigation to achieve high-accuracy, real-time tracking, thereby supporting both diagnostic precision and emerging telemedicine applications [19], [20]. Electromagnetic safety is another fundamental consideration. Parameters such as Specific Absorption Rate (SAR) and current density are strictly regulated to ensure safe operation. Recent innovations, including biodegradable “transient electronics,” demonstrate the potential of devices that safely dissolve under physiological conditions [21]. Future capsule robots will likely incorporate these materials alongside advanced energy management strategies to balance inference accuracy, memory capacity, computational speed, and energy efficiency—challenges that must be resolved to achieve reliable, real-time performance [7]. Smart capsules are rapidly emerging as a next-generation diagnostic and monitoring platform for gastrointestinal health [22]. These devices are expected to play a central role in intelligent healthcare systems by improving resource efficiency, reducing hospitalizations, and empowering patients to manage their health actively [16], [17]. The SmartPill® system exemplifies this progress by measuring pH, temperature, and pressure during GI transit [25]. Beyond this, next-generation capsules are being equipped with sensors to monitor diverse physiological parameters, enabling more comprehensive assessment of the gut [21], [22]. By combining AI, IoT, and biomedical engineering, smart capsules hold promise for individualized treatments, real-time disease tracking, and timely therapeutic interventions [26]. Emerging robotic platforms, such as those designed for microbiota sampling, illustrate new directions with strong potential to transform diagnostics, monitoring, and therapy [27], [7]. Previous reviews have often focused on imaging hardware or AI developments in isolation [9], [10]. To the best of our knowledge, few have systematically integrated perspectives from AI, IoT, and CS. This review uniquely bridges these domains, highlighting how their convergence accelerates the development of smart capsule platforms and paves the way for multifunctional robotic capsules with diagnostic and therapeutic potential [27]. By mapping these cross-disciplinary advances, the present review aims to serve as both a comprehensive reference for researchers and a roadmap for future innovations in autonomous, intelligent capsule endoscopy. Specifically, this review highlights the synergistic role of AI, IoT, and compressed sensing in advancing smart capsule endoscopy, with a focus on data transmission, lesion detection, and localization. The paper is organized as follows: presents the review methodology, including literature selection, screening, and synthesis procedures. reports the results of the structured literature analysis and categorizes the reviewed studies. discusses the key methodologies, mathematical frameworks, and analytical tools employed across AI, IoT, and CS in WCE research. outlines the fundamentals and clinical challenges of wireless capsule endoscopy. examines AI-based automated lesion detection and intelligent diagnostic models. reviews the role of IoT in connectivity, telemetry, and remote monitoring architectures. presents compressed sensing approaches for efficient data transmission and localization. introduces the objectives and framework of the comparative study. offers a critical discussion of current limitations and translational gaps. highlights future research opportunities and open challenges. concludes the paper with a synthesis of findings and a summary of core enabling technologies. 2. Review Methodology (Methods / Experimental) This review was conducted using a structured literature-selection and synthesis workflow to identify recent and influential contributions related to wireless capsule endoscopy (WCE), with an emphasis on indoor localization and enabling technologies such as artificial intelligence (AI), the Internet of Things (IoT), and compressed sensing (CS). Search Strategy and Sources: Relevant studies were identified through searches in major scientific databases commonly used in engineering and biomedical research (e.g., IEEE Xplore, PubMed, Scopus, and Google Scholar). Search queries combined terms associated with capsule endoscopy and localization (e.g., wireless capsule endoscopy, capsule localization, indoor localization, magnetic tracking, visual odometry, RF-based localization), together with enabling-technology keywords (e.g., deep learning, CNN, IoT, telemetry, edge computing, compressed sensing, sparse reconstruction). Reference lists of highly relevant papers and recent surveys were also screened to capture additional key sources (snowballing). Inclusion and Exclusion Criteria: Articles were included if they (i) addressed WCE localization and/or closely related sensing and tracking approaches, or (ii) contributed enabling methods that directly impact WCE performance (e.g., AI-based lesion detection and triage, IoT connectivity and remote monitoring, CS-based data reduction and transmission efficiency). Priority was given to peer-reviewed journal and conference publications in English. Studies that were out of scope (e.g., unrelated endoscopic modalities, non-capsule localization settings without transferable methodology, or papers lacking sufficient technical description) were excluded. Screening and Synthesis: Titles and abstracts were initially screened to exclude clearly irrelevant records, followed by full-text evaluation to confirm eligibility. The initial search yielded more than 200 publications. After duplicate removal and relevance screening, a subset was retained for in-depth qualitative synthesis based on technical rigor and thematic relevance. The selected studies were subsequently organized into five thematic categories aligned with the structure of this review: (1) WCE fundamentals and system constraints, (2) AI-driven lesion detection and intelligent processing, (3) IoT-enabled connectivity and remote monitoring architectures, (4) compressed sensing-based acquisition, compression, and localization efficiency, and (5) comparative integration strategies and open research challenges. These five thematic categories were later synthesized into broader technological domains for high-level trend analysis. The evidence was synthesized using a narrative analytical approach supported by comparative tables, with particular attention to methodological assumptions, implementation constraints (e.g., energy consumption, bandwidth limitations, and miniaturization), and reported performance characteristics where available. 3. Results of Literature Analysis The structured search and screening process resulted in the identification of a substantial body of literature addressing wireless capsule endoscopy (WCE) localization and enabling technologies. After removing duplicates and out-of-scope studies, for high-level analytical synthesis, the selected publications were further grouped into three principal technological domains: (i) AI-based lesion detection and intelligent interpretation, (ii) IoT-enabled connectivity and system-level integration, and (iii) compressed sensing (CS)-based data reduction and efficient localization frameworks. Trend Analysis: A clear temporal progression was observed in the literature. Early studies primarily focused on hardware-based tracking mechanisms such as magnetic field localization and RF triangulation. Over time, the research focus expanded toward software-driven intelligence, including convolutional neural networks (CNNs), hybrid AI-detection models, and real-time anomaly classification. In recent years, increasing attention has been directed toward integrated system architectures combining AI inference, IoT telemetry, and energy-aware data compression techniques. Technology Distribution: Among the reviewed studies, AI-based detection approaches represent the most rapidly growing category, particularly in automated lesion detection and real-time classification tasks. IoT-related contributions focus on system connectivity, telemetry reliability, and low-power communication strategies. CS-based approaches primarily address bandwidth limitations, sparse signal reconstruction, and efficient transmission under strict energy constraints. Reported Performance Trends: Across AI-driven lesion detection studies, deep learning architectures are generally associated with improved classification performance compared to earlier feature-engineering-based approaches. Nevertheless, practical deployment remains constrained by real-time processing requirements, model interpretability concerns, and dataset heterogeneity. IoT-oriented frameworks contribute to enhanced system connectivity and remote monitoring capabilities, while continuing to face limitations related to power efficiency and secure data transmission. Similarly, compressed sensing strategies support transmission load reduction and bandwidth optimization; however, reconstruction fidelity and robustness under noisy or variable clinical conditions remain active areas of investigation. Identified Gaps: Despite significant progress, few studies demonstrate fully integrated end-to-end smart capsule architectures combining AI, IoT, and CS in a clinically validated environment. Most contributions remain domain-specific, indicating a translational gap between laboratory validation and real-world deployment. These results provide the foundation for the structured technical discussion presented in the following sections. 4. Methodologies, Mathematical Models, and Tools 4.1 Methodological Taxonomy in Smart Wireless Capsule Endoscopy Research on smart wireless capsule endoscopy (WCE) has evolved along several distinct methodological paradigms, reflecting different assumptions about data availability, system observability, and computational constraints. Existing approaches can be broadly categorized into model-based, data-driven, and hybrid methodologies, each addressing specific challenges in localization, perception, and decision support for WCE systems. Model-based methodologies rely on explicit physical or mathematical representations of the underlying system dynamics and sensing mechanisms. In the context of WCE localization, magnetic-field–based approaches exemplify this paradigm by formulating the capsule-positioning problem as a parameter-estimation (inverse) problem grounded in electromagnetic field models. Studies based on magnetic dipole formulations, gradient tensor analysis, or normalized source strength exploit known physical laws to infer capsule position and orientation from sensor measurements, often employing optimization or inversion techniques to estimate unknown states [18–20], [28]. Similarly, vision-based and kinematic localization methods use geometric constraints, motion models, or state-space representations to estimate capsule trajectories, typically under observability assumptions and bounded noise [19], [20]. These model-based approaches offer interpretability and theoretical guarantees but may be sensitive to modelling inaccuracies, environmental disturbances, and incomplete prior knowledge. In contrast, data-driven methodologies treat WCE tasks as pattern recognition or function approximation problems, leveraging large volumes of labelled or unlabelled data to learn mappings directly from observations to outputs. Recent advances in artificial intelligence have accelerated the adoption of deep-learning–based pipelines for lesion detection, frame classification, and auxiliary localization cues. Convolutional neural networks and related architectures have been widely applied to endoscopic image streams to reduce diagnostic workload and improve detection accuracy by learning discriminative features without explicit physical modelling [29–34]. These approaches demonstrate strong performance in high-dimensional perceptual tasks and are particularly effective when physical models are difficult to define. However, their dependence on extensive training data, limited interpretability, and sensitivity to domain shifts remain important methodological considerations. Bridging these two paradigms, hybrid methodologies integrate model-based reasoning with data-driven learning to exploit the complementary strengths of both approaches. Representative studies combine physics-informed models with neural networks, for example, by applying learning-based super-resolution techniques to enhance sensor data before model inversion, or by constraining optimization with data-driven initial estimates [35–37]. In magnetic localization, hybrid strategies have been proposed in which deep neural networks improve spatial resolution or estimate coarse target parameters, followed by deterministic or trust-region optimization to refine capsule position and magnetic moment estimates [37]. Such hybrid formulations reduce the dimensionality of the solution space, mitigate convergence issues, and improve robustness under real-time constraints, making them particularly attractive for clinical WCE applications. Overall, the methodological landscape of smart WCE reflects a progression from purely physics-based formulations toward learning-centric and hybrid frameworks. While model-based approaches emphasize interpretability and physical consistency, data-driven methods prioritize scalability and perceptual accuracy. Hybrid methodologies are increasingly emerging as a unifying strategy, enabling robust performance by embedding learned representations within structured mathematical models. This taxonomy provides a conceptual foundation for analyzing subsequent developments in mathematical modelling and computational toolchains for smart WCE systems. 4.2 Mathematical Modelling Frameworks Mathematical modelling plays a central role in smart wireless capsule endoscopy (WCE) by providing a principled foundation for localization, signal reconstruction, image understanding, and decision-making under severe physical and computational constraints. Across the literature, the dominant modelling frameworks can be grouped into state-space estimation models, inverse and sparse reconstruction models, and learning-based function approximation models, each reflecting different assumptions about system observability, noise statistics, and data availability. State-space and estimation-based models are widely used for capsule localization and motion inference, in which the capsule pose is treated as a latent state that evolves over time. In magnetic and sensor-based localization, the relationship between the capsule state and measured magnetic field intensities is captured by nonlinear measurement equations derived from dipole or multipole field models [18–20], [28], [38]. These formulations naturally lead to recursive estimation frameworks in which capsule position and orientation are inferred by minimizing prediction–measurement mismatches under assumed noise models. Although explicit filters are not always stated, the underlying mathematical structure corresponds to nonlinear state estimation problems, where identifiability and convergence depend on excitation conditions, sensor geometry, and modelling fidelity [19], [20]. Such models emphasize physical consistency and enable uncertainty-aware inference but may degrade when the assumed field or motion models are violated in complex anatomical environments. A second class of formulations arises from inverse problems and sparse reconstruction theory and is particularly relevant to signal acquisition, compression, and recovery under bandwidth and power constraints. Compressive sensing–based approaches model the acquired measurements as underdetermined linear systems in which the underlying signal admits a sparse representation in a suitable dictionary or frame [39–41]. Within this framework, signal recovery is cast as an optimization problem—often ℓ₁-regularized or constrained—whose solvability relies on sparsity assumptions and measurement incoherence. Extensions of modulo and nonlinear sampling further interpret signal recovery as an ill-posed inverse problem, in which uniqueness and stability depend on sampling rates, folding mechanisms, and prior knowledge of signal structure [42]. These mathematical models are attractive for WCE systems because they explicitly encode acquisition constraints while providing theoretical recovery guarantees, albeit at the cost of increased computational complexity. In parallel, learning-based mathematical models reformulate WCE tasks as nonlinear function approximation problems. Deep neural networks, particularly convolutional architectures, are used to approximate mappings from high-dimensional image or signal spaces to diagnostic labels, saliency maps, or auxiliary localization cues [30],[43],[44],[32]. From a modelling perspective, these approaches replace explicit physical equations with parameterized nonlinear operators optimized through empirical risk minimization. Techniques such as transfer learning and feature reuse can be interpreted as imposing implicit priors on the learned function space, mitigating data scarcity and stabilizing training dynamics [40]. While these models often achieve superior performance in perceptual tasks, their mathematical guarantees are largely empirical, and interpretability remains limited compared to physics-based formulations. Recent studies increasingly explore hybrid modelling frameworks that integrate learning-based components into structured mathematical models. Examples include combining sparse reconstruction principles with neural network–assisted estimation or using learned features to initialize or constrain optimization-based solvers [37]. In such cases, learning components reduce model mismatch or dimensionality, while the underlying optimization or estimation framework preserves physical interpretability and robustness. These hybrid formulations highlight a broader trend toward unifying data-driven adaptability with mathematically grounded inference in smart WCE systems. Overall, the mathematical modelling landscape in WCE reflects a balance between rigour and flexibility. State-space and inverse problem formulations provide interpretability and theoretical structure, whereas learning-based models offer scalability and performance in complex perceptual domains. Hybrid frameworks are a promising approach that leverages complementary strengths to address the inherent challenges of localization, signal recovery, and diagnosis in resource-constrained ingestible platforms. 4.3 Computational and Hardware Tools The tight coupling between computational tools and hardware platforms fundamentally constrains the practical deployment of smart wireless capsule endoscopy (WCE) systems. Unlike conventional medical imaging pipelines that rely on off-device processing, WCE requires embedded, energy-aware, and real-time computational infrastructure that operates under severe size, power, and thermal constraints. Consequently, existing studies adopt a co-design philosophy in which algorithms, software frameworks, and hardware architectures are jointly optimized. From a computational perspective, software toolchains for WCE can be broadly divided into offline training environments and on-device inference pipelines. Deep learning–based diagnostic and perception models are typically developed and trained using high-level frameworks such as TensorFlow or PyTorch, enabling rapid prototyping, transfer learning, and large-scale optimization on external computing resources [29],[44],[32]. Once trained, these models are converted into lightweight representations suitable for embedded execution, often through model compression, quantization, or architectural simplification. In edge-AI capsule systems, such as those enabling real-time lesion detection and adaptive task execution, inference pipelines are tightly optimized to meet strict latency and memory budgets while maintaining clinically acceptable accuracy [32]. Complementary to learning-based pipelines, signal processing and reconstruction tasks rely on numerical optimization and linear algebra tools that are frequently implemented in environments such as MATLAB or Python-based scientific libraries [45], [46]. These tools support compressed-sensing–based recovery, sparse optimization, and the simulation of acquisition constraints before hardware deployment. Although such environments are rarely embedded directly in the capsule, they play a critical role in validating reconstruction fidelity, tuning regularization parameters, and assessing robustness under realistic noise and sampling conditions. At the hardware level, system-on-chip (SoC) platforms form the backbone of modern smart WCE devices. Recent designs integrate microcontrollers, digital signal processors, and dedicated neural processing units to support on-capsule intelligence while minimizing energy consumption [32],[47]. Application-specific integrated circuits (ASICs) and field-programmable gate arrays (FPGAs) are increasingly used to accelerate convolutional operations and feature-extraction tasks, enabling real-time processing under tight power envelopes [48],[49]. These hardware accelerators enable critical computations to be executed locally, reducing reliance on continuous wireless transmission and thereby extending operational lifetime. Wireless communication modules are another essential hardware component, bridging embedded computation with external receivers. Technologies such as Bluetooth Low Energy (BLE), ultra-wideband (UWB), and custom RF telemetry links are used to transmit compressed data streams, intermediate inference results, or event-driven alerts [47],[49]. The choice of communication protocol directly influences computational design decisions, as bandwidth limitations motivate on-device preprocessing and selective data transmission rather than raw video streaming. Beyond active electronic components, passive and semi-passive sensing hardware also plays a significant role in WCE ecosystems. Inductor–capacitor (LC)–based sensor architectures enable battery-free or low-power sensing through resonant coupling, enabling continuous physiological monitoring without dedicated power sources [45]. These architectures reduce computational overhead by encoding sensed information as frequency or phase shifts, which can be decoded externally with minimal on-capsule processing. Overall, the computational and hardware tools employed in smart WCE systems reflect a strong emphasis on edge intelligence, hardware–software co-design, and energy efficiency. Rather than treating computation as an isolated layer, existing approaches tightly integrate algorithmic complexity with hardware capabilities and communication constraints. This integrated toolchain provides the technological foundation for advanced modelling, perception, and diagnostic functions in next-generation ingestible devices. 4.4 Assumptions, Constraints, and Trade-offs Despite significant progress in smart wireless capsule endoscopy (WCE), existing methodologies and technologies are inevitably shaped by underlying assumptions and practical constraints that impose fundamental trade-offs on system design and performance. These trade-offs arise from the capsule's ingestible nature, the variability of the gastrointestinal environment, and stringent requirements for safety, reliability, and clinical usability. A primary assumption underlying many WCE systems is that sufficiently informative sensory data are available under constrained operating conditions. Vision-based approaches typically assume adequate illumination, a stable imaging geometry, and acceptable motion blur to ensure reliable perception and analysis [50], [32]. In practice, however, variable transit speeds, occlusions by intestinal contents, and non-uniform lighting conditions can significantly degrade image quality. Similarly, magnetic and RF-based localization techniques often assume quasi-static or weakly perturbed field environments, yet real anatomical settings introduce distortions, interference, and patient-specific variability that challenge these assumptions [51],[36]. Energy availability constitutes one of the most critical constraints in WCE and directly governs the balance between onboard intelligence and operational lifetime. Most capsule platforms rely on limited battery capacity, necessitating careful trade-offs among sensing resolution, computation, and wireless transmission [52],[53]. Increasing frame rates, deploying deep neural networks, or enabling continuous localization inevitably raises power consumption, reducing examination duration or necessitating aggressive duty cycling. As a result, many systems assume intermittent processing or selective data transmission, prioritizing salient events over continuous high-fidelity monitoring [32]. These assumptions influence not only algorithmic design but also clinical protocols and expectations. Another important trade-off concerns accuracy versus interpretability. Data-driven models, particularly deep learning–based diagnostic systems, often achieve superior performance in lesion detection and classification tasks, yet operate as black-box models with limited transparency [32]. In contrast, physics-based localization and signal reconstruction approaches offer clearer interpretability and error characterization but may exhibit reduced robustness under modelling mismatch or noise [51]. Hybrid strategies implicitly assume that learned components can compensate for model inaccuracies while preserving sufficient structure for reliable inference; however, this balance remains sensitive to the quality of the training data and the deployment conditions [36]. Latency and communication bandwidth further constrain system design. High-resolution image transmission and real-time feedback demand substantial wireless bandwidth, which is often unavailable or energetically prohibitive in ingestible devices [53]. Consequently, many WCE systems assume delayed or offline analysis or rely on onboard preprocessing to reduce data volume prior to transmission. This introduces trade-offs among the immediacy of clinical feedback, computational complexity, and local autonomy versus external supervision. Finally, clinical deployment introduces assumptions about safety and robustness that constrain technological choices. Assumptions about biocompatibility, thermal safety, electromagnetic exposure, and mechanical reliability limit the integration of high-power actuators, aggressive locomotion mechanisms, or dense electronic components [50],[52]. These constraints often favour conservative designs that trade advanced functionality for proven safety and regulatory acceptance. Overall, the evolution of smart WCE systems is characterized by ongoing negotiation among competing objectives: intelligence versus endurance, accuracy versus interpretability, autonomy versus safety, and performance versus practicality. Recognizing these assumptions and trade-offs is essential for contextualizing reported results and guiding future research toward clinically viable, balanced, and scalable solutions. 5. Fundamentals and Challenges in Wireless Capsule Endoscopy 5.1 Overview of Wireless Capsule Endoscopy WCE is a minimally invasive diagnostic technique that visualizes the gastrointestinal tract without sedation or the risks associated with conventional endoscopy [54–56]. It is widely used to diagnose obscure gastrointestinal bleeding, Crohn’s disease, celiac disease, small bowel tumours, and polyposis syndromes [8]. The origins of WCE date back to early endoscopic experiments by Philipp Bozzini in the 19th century. Later, Gavriel Iddan [57] recognized the potential of fibre optics and, working with engineers, developed the first wireless capsule capable of imaging the small bowel. The basic structure and main components of a typical capsule endoscope are illustrated in Figure .2. In 2001, Given Imaging commercialized this innovation as the M2A system (“mouth to anus”), marking a milestone in non-invasive gastrointestinal diagnostics [58–61]. The capsule integrates miniature cameras, LEDs, RF transmitters, and batteries within a biocompatible shell of about 11×26 mm [62–64]. However, these devices still operate passively, moving only with natural peristalsis. This lack of control limits maneuverability and contributes to omission rates of up to 30% [65–67]. Such limitations underscore the need for next-generation “smart capsules” equipped with active navigation, targeted drug delivery, and biopsy capabilities [68]. While WCE has revolutionized gastrointestinal diagnostics, its passive locomotion remains a major barrier to realizing its full potential. 5.2 Power Supply Limitations Conventional wireless capsules typically rely on two silver-oxide button-cell batteries, which provide approximately 20 mW at 3 V for 8–10 hours of operation [69–71]. While sufficient for basic imaging, these batteries limit advanced functionality due to their limited energy capacity, output power, and safety concerns [72–74]. Earlier generations also faced rapid depletion from power-hungry components such as CCD sensors [75]. To overcome these constraints, researchers have explored custom lithium-ion polymer batteries that deliver much higher power density—up to 2000 times greater than silver oxide cells—while supporting peak currents [76–78]. However, these raise safety concerns, particularly risks of thermal runaway in vivo. Alternative approaches, such as self-powered systems that harvest energy from gastric fluids, have also been tested, as have edible electronics based on food-grade materials (e.g., edible batteries, supercapacitors, and nanogenerators). Despite their biocompatibility, these strategies still fail to provide enough power for continuous capsule operation [67, 79, 80]. Supercapacitors store only limited energy, and nanogenerators convert mechanical or thermal stimuli into electricity but remain suitable only for low-demand functions [81–83]. A promising approach is to embed piezoelectric nanogenerators in the capsule shell, which convert peristaltic motion into usable electricity [84]. Another major line of research is Wireless Power Transmission (WPT), where energy is delivered externally via electromagnetic waves. Although magnetic resonance–based WPT has improved coupling efficiency, practical barriers remain, including tissue absorption, organ motion, and transmitter–receiver misalignment [67, 85–88]. In summary, while lithium-ion batteries and WPT prototypes show great potential for extending capsule lifetime, their clinical applicability is limited by safety risks, alignment challenges, and regulatory constraints. Self-powered and edible systems demonstrate biocompatibility but remain far from supporting full diagnostic functionality. At present, energy supply remains the most critical bottleneck in WCE, highlighting the need for multidisciplinary innovations that balance power density, safety, and clinical feasibility. 5.3 Data Transmission and Telemetry Challenges Reliable high-throughput communication is critical for WCE, as demand for high-resolution imaging and for integrating multiple sensors continues to increase. Conventional systems typically operate near 400 MHz with only 300 kHz bandwidth, which restricts video quality and makes signals vulnerable to absorption, scattering, and multipath fading, particularly in deep gastrointestinal regions [89], [67],[51]. To address these limitations, UWB communication has been proposed. Operating in the 3.1–10 GHz range, UWB supports data rates exceeding 100 Mb/s and offers improved energy efficiency [90]. However, it requires complex RF hardware and is subject to strict regulatory compliance, which has limited its clinical adoption [91], [92]. In parallel, low-power protocols such as Bluetooth Low Energy (BLE) have been evaluated for their simplicity and energy efficiency. Yet, their performance is significantly constrained by severe attenuation at 2.4 GHz, with devices such as iMAG showing poor results through thick gastric walls. To overcome this, researchers have explored lower-frequency bands (400–915 MHz) that reduce absorption and extend range, though these come at the expense of lower throughput [93]. Hybrid solutions are also under investigation. Approaches such as dual-band modules and Intra Body Communication (IBC), leveraging capacitive or galvanic coupling, aim to achieve more robust two-way communication while maintaining energy efficiency [94]. Despite these advances, ensuring reliable bidirectional telemetry in the highly dynamic gastrointestinal environment remains a substantial challenge. Therefore, while UWB offers unmatched throughput [90], its complexity and regulatory barriers make near-term clinical deployment unrealistic [91]. BLE and sub-GHz protocols [93] are more practical for current WCE systems, although their limited bandwidth restricts continuous high-resolution imaging. Hybrid architectures that combine these modes may therefore provide the most feasible pathway toward reliable, energy-efficient capsule telemetry [94]. The relative trade-offs between throughput and penetration for UWB, BLE, sub-GHz, and hybrid protocols are illustrated in the figure. 3. 5.4 Locomotion and Navigation Challenges Controllable locomotion remains a major challenge in WCE. Current capsules rely on passive peristalsis, which lacks directional control and often results in incomplete inspection of target regions [95–97]. This passive movement restricts diagnostic reliability and limits the ability to focus on suspicious lesions. To address these limitations, researchers have investigated active locomotion strategies, including bio-inspired propulsion systems such as earthworm-like crawling, screw-jet mechanisms, and robotic actuation [95], [57]. While these methods demonstrate feasibility in laboratory conditions, their integration into capsule-sized devices is limited by available volume, battery capacity, and the risk of heat generation during operation. Among alternative approaches, magnetic guidance offers five degrees of freedom (5-DOF) by manipulating internal magnets in response to external magnetic fields [51]. Recent designs employ Reciprocally Rotating Magnetic Actuation (RRMA) to minimize drag in narrow spaces and electromagnetic coil arrays (e.g., Helmholtz and Maxwell systems) for dynamic orientation control [98], [99]. Although these methods improve maneuverability, they increase power demand and depend heavily on large external equipment, which reduces clinical practicality. Accurate localization also remains unresolved. Techniques based on radio-frequency signals, visual odometry, and magnetic tracking have been tested, but none consistently achieve real-time six-degree-of-freedom (6-DoF) accuracy. Hybrid approaches combining visual landmarks, magnetic mapping, and RF tracking show promise for sub-centimetre precision [28], [67], though most remain experimental and unvalidated in large-scale clinical trials. Therefore, while active propulsion and magnetic guidance significantly enhance maneuverability, their reliance on bulky external systems and high energy demand limits clinical adoption. Passive peristalsis remains the only approved method in practice. Future WCE systems must resolve this trade-off by developing compact, energy-efficient locomotion and robust hybrid localization, bridging the gap between experimental prototypes and real-world clinical feasibility. 5.5 Hardware Constraints and Miniaturization in WCE The push for miniaturization in WCE creates major engineering challenges. Capsules must integrate imaging units, batteries, and communication hardware within dimensions of no more than 32×12 mm, thereby imposing strict design constraints [51]. Adding advanced functions such as high-definition imaging, telemetry, or locomotion increases complexity. For example, Micromotors and Shape Memory Alloys (SMAs) can generate motion but occupy valuable space, produce heat, and respond slowly, limiting real-time navigation [100],[101]. Thermal regulation and biocompatibility add further constraints. Devices must avoid overheating and excessive Specific Absorption Rate (SAR), particularly during high-frequency transmission or Wireless Power Transfer (WPT). Magnetic actuation systems also require coils and magnets, which reduce the available space[41]. Advanced fabrication techniques, such as MEMS, have been used to increase functional density without increasing capsule size. Yet many next-generation capsules remain relatively bulky, sometimes exceeding safe ingestion thresholds, particularly in patients with strictures. Tethered or robotic capsules intended for therapeutic use also add cost and complexity [51]. A comparative overview of representative WCE systems is provided in Table 1 [67]– [51]. Table 1 Some of the most well-known WCEs with their corresponding characteristics. Capsule Name Dimensions (mm) Weight (g) Battery Life (hrs.) Navigation Drug Delivery Power Transfer Notes PillCam SB3 26 × 11 ~ 3.0 8–12 Passive No Battery Clinically established NaviCam 27 × 12 ~ 3.5 10–12 Magnetic No Battery Requires an external magnet Sayaka 31 × 13 ~ 4.0 8 Passive No Battery High-resolution imaging iMAG 23 × 9 < 3.0 ~ 6 Passive No Wireless (BLE) Experimental Endocapsule 10 26 × 11 3.5 12 Passive No Battery Improved transmission Experimental Prototype (RRMA) 32 × 13 ~ 5.0 ~ 6 Active (Magnetic) No Inductive WPT Research phase As shown in Table 1 , increasing capsule functionality often entails trade-offs among size, weight, and battery life. This highlights the need for innovative designs that balance miniaturization with performance. Furthermore, energy management strategies such as adaptive frame rates and data compression can extend battery life but may reduce image quality [102–104]. 5.6 Biopsy, Drug Delivery, and Functional Limitations Expanding WCE beyond diagnostics to therapeutic applications is among the most ambitious goals in this field. The primary targets include integrating biopsy tools, targeted drug delivery, and real-time procedures such as ablation or lesion removal. Early designs, such as the Crosby-Kugler capsule, attempted to perform a biopsy but failed due to inaccurate forceps operation, lack of feedback, and single-use limitations [97],[105]. Even advanced robotic prototypes still face trade-offs among actuation force, compactness, and biocompatibility[51]. Drug delivery systems—such as spring reservoirs, osmotic pumps, and pH-sensitive coatings—have also been tested, but variable GI motility and localization errors hinder precise release [106–108]. More advanced strategies, including wireless-controlled and magnetically triggered mechanisms, show promise yet remain constrained by anatomical variability and tissue conductivity [109]. Efforts toward real-time therapeutic interventions—such as laser ablation, microsurgery, or biopsy—pose additional engineering challenges. Micro-actuated blades and needles require power-hungry modules, sterilizable materials, and rigorous safety validation, particularly in sensitive tissue regions [110]. Moreover, achieving full-duplex telemetry for reliable bidirectional control remains unresolved in commercial systems [111]. Hybrid capsules that combine biopsy and drug delivery often increase in size, raising concerns about swallowability for patients with strictures or pediatric populations [112], [113]. Ultimately, the vision of a fully autonomous “surgical capsule” remains aspirational. Its realization will depend on advances in energy-efficient actuation, wireless power transfer, and accurate real-time localization, while ensuring compliance with stringent clinical safety and regulatory standards [51]. 5.7 Regulatory and Clinical Translation Barriers Despite major technical advances, translating WCE innovations into routine clinical practice remains hindered by significant regulatory and translational challenges. Approval processes for therapeutic capsules are lengthy because mechanical components, such as drug actuators, lasers, and biopsy tools, must comply with strict safety, reliability, and sterilization standards. Even minor modifications—such as changes to embedded electronics or locomotion systems—often necessitate full recertification, thereby prolonging commercialization timelines [51], [67]. Clinical readiness is further limited by insufficient validation. Many prototypes demonstrate high performance in controlled laboratory environments but fail under variable physiological conditions. The lack of large-scale clinical trials across diverse populations prevents consistent demonstration of safety and diagnostic efficacy, delaying regulatory endorsement [114–116]. The integration of AI introduces additional ethical and legal concerns. Deep learning models often function as “black boxes,” producing results that are difficult to interpret. In cases of diagnostic errors, accountability remains unclear. Clinicians also remain skeptical of automated outputs, particularly when dealing with ambiguous or borderline cases [51], [117]. Infrastructure represents another major barrier. Many hospitals lack the computational resources, network infrastructure, or trained personnel required to deploy AI-enhanced WCE. Lightweight AI frameworks optimized for edge computing have been proposed to address these gaps, while federated learning approaches aim to improve privacy and compliance with data protection laws [117–120]. However, these solutions remain in the early stages of adoption. Therefore, the regulatory and clinical translation of WCE is hindered not by a single factor but by the convergence of multiple unresolved issues. Although prototypes demonstrate impressive technical promise [51], their adoption is hindered by limited clinical validation [114], [115], unclear liability in AI-assisted diagnosis [121], [122], and persistent infrastructure gaps [118]. These combined challenges reinforce regulatory and translational barriers as the most critical bottleneck. Unless a coordinated framework for safety, standardization, and oversight is established [57], advanced WCE systems risk being confined to prototypes, widening the gap between research innovation and patient care. 5.8 Section Summary WCE is evolving from a passive diagnostic tool into a multifunctional robotic system. This transformation relies on advances in AI, Microelectromechanical Systems (MEMS), Microfabrication, and UWB communication [123–126]. A major research direction is the integration of therapeutic functions. Procedures such as Endoscopic Submucosal Dissection (ESD) and Endoscopic Mucosal Resection (EMR), once limited to tethered devices, are now being reimagined for capsules through innovations in miniaturized energy sources [127], [128]. Novel mobility strategies, including magnetic navigation and swarm robotics, are also under development. Concepts such as a “mothership” capsule coordinating smaller “soldier” units highlight the potential for parallel interventions. At the same time, soft magnetic materials may enable more flexible locomotion, controlled drug delivery, and site-specific adhesion [67]. Therapeutic functions are advancing, with efficient CNNs and knowledge distillation enabling on-board analysis. Yet, trade-offs between speed, memory, energy, and diagnostic accuracy remain unresolved [129], [130]. In parallel, WCE is emerging as a tool for therapeutic functions, particularly when combined with 5G and cloud infrastructures. Lessons from the COVID-19 pandemic underline the importance of scalable remote diagnostics for underserved populations [67]. In navigation, hybrid localization frameworks that combine magnetic mapping, visual landmarks, and RF tracking aim to provide robust and precise positioning, even in anatomically complex regions [131–133]. Looking ahead, next-generation WCE platforms are expected to feature: Embedded AI for autonomous interpretation. Integration of diagnostic and therapeutic modules. 3Enhanced navigational precision using magnetic and hybrid localization. Secure, high-speed communications with UWB and 5G. Collectively, these advances could transform WCE into an intelligent, self-guided, globally deployable platform that automates many diagnostic and therapeutic procedures [67], [51], [111]. Achieving this vision will require multidisciplinary collaboration among engineers, clinicians, regulators, and healthcare providers. 6. AI in Wireless Capsule Endoscopy WCE has GI diagnostics that generate more than 50,000 image frames per exam, posing major challenges for manual review [29], [134]. To address this workload, AI has emerged as a key tool for automating interpretation, improving diagnostic efficiency, and reducing physician fatigue. AI-based systems now perform real-time analysis in WCE, detecting abnormalities and assisting decision-making [52]. Compared with handcrafted feature methods such as LBP, SIFT, or HOG, deep learning—particularly CNNs—achieves higher accuracy and generalization [30], [43]. CNNs have surpassed human experts on benchmark tasks such as CAMELYON17 histology classification [30]. In endoscopy, these models demonstrate high sensitivity (> 90%) for detecting bleeding, polyps, and ulcers, using architectures such as VGG16, ResNet50, InceptionV3, and EfficientNet [31], [35]. Transfer learning further enhances performance: pre-trained CNNs on ImageNet, when adapted to WCE datasets, reduce training time, mitigate overfitting, and improve detection accuracy, particularly for polyp classification [134], [44]. Edge AI is also being explored, where frames can be prioritized or filtered in real time to reduce redundant transmission [32], [44]. Clinical benefits are evident. AI-assisted review reduces interpretation time by more than 40% while maintaining or improving diagnostic performance [135]. In trials, AI identified more than 95% of bleeding lesions missed by human reviewers [33], and hybrid systems combining deep learning with rule-based logic achieved higher specificity and F1 Scores [34]. Challenges remain, including inconsistent image quality, lighting issues, and limited annotated datasets [52]. Class imbalance is particularly problematic, as lesion frames may constitute < 2% of the data, increasing the risk of false negatives [136], [53]. To address this, methods such as augmentation, GAN-based synthesis, weak supervision, and anomaly detection have been applied [137], [138]. Ethical and regulatory aspects are also critical. These include ensuring interpretability, mitigating bias, and validating across diverse populations, with AI positioned as a decision-support tool rather than a replacement for clinicians [135], [139]. In summary, AI adoption in WCE is driven by the need to manage large volumes of images, detect subtle lesions, and streamline clinical workflows. While deep learning has demonstrated strong potential, future research must focus on robustness, transparency, and adaptability. Section 6.1 next examines CNN architectures and transfer learning as the foundation of AI-driven lesion detection in WCE. 6.1 CNNs and Transfer Learning in Wireless Capsule Endoscopy CNNs are the central technology for AI-based lesion detection in WCE. Their hierarchical architecture enables automatic feature extraction, thereby avoiding reliance on handcrafted descriptors such as LBP, Gabor filters, and SIFT [30]. These models can capture complex spatial and texture-based patterns in GI tract images. Early research trained CNN architectures, such as AlexNet, VGGNet, and GoogLeNet, on labelled WCE datasets. These implementations showed that CNNs can differentiate normal from abnormal frames by learning convolutional filters directly from image data. For example, a 6-layer CNN achieved 93.4% sensitivity in bleeding detection on the Kvasir dataset [31], and GoogLeNet reached 96.36% accuracy in polyp classification [35]. A major limitation of deep CNNs is the need for large labelled datasets, which are often unavailable. Transfer learning addresses this issue by fine-tuning models pre-trained on datasets such as ImageNet. This approach accelerates training and improves classification on small GI datasets [134]. Studies report that fine-tuned ResNet50 and InceptionV3 models outperform custom-built CNNs by up to 8% in ulcer and polyp detection [44]. Several architectures have been explored for WCE tasks, such as: InceptionV3 and EfficientNet [32]: Strong performance relative to model size. EfficientNetB0 delivered 94.2% accuracy with fewer FLOPs, making it suitable for portable systems. ResNet variants [33]: Skip connections ease gradient flow. ResNet50 achieved an AUC of 0.96 in ulcer classification, about 15% better than standard CNNs. VGG16 and VGG1[136]: Deep simple convolutional stacks with high sensitivity but high computational cost. Beyond classification, CNNs have been adapted for segmentation. Encoder–decoder architectures, such as U-Net variants, have achieved Dice scores above 0.90 in bleeding segmentation [135].Capsule Networks (CapsNet) also show promise for handling image rotations and geometric distortions in capsule footage [34]. Transfer learning has proven valuable in multi-class classification, including bleeding, polyps, erosions, and tumours—a DenseNet121-based model classified five lesion categories with an average F1-score of 91.8% [29]. To improve domain adaptation, strategies such as freezing early layers, applying differential learning rates, and visualizing intermediate features have been applied. Freezing shallow layers while fine-tuning deeper ones aligns models with WCE-specific colour tones and textures [52]. Visualization methods such as CAM and Grad-CAM help interpret predictions and build clinician confidence [139]. Despite these advances, CNNs face generalization issues. Models trained on one dataset often perform poorly on other datasets due to differences in hardware, lighting, and annotation [137]. Class imbalance is another challenge: normal frames vastly outnumber pathological ones, biasing models toward the majority class [53]. Solutions include cost-sensitive loss functions, oversampling, and hard example mining [43]. Lightweight CNNs have also been developed for real-time edge processing. A depthwise separable CNN achieved 28 FPS on a Raspberry Pi-based WCE receiver, showing the feasibility of on-device inference for lesion detection [44]. Such approaches can reduce transmission costs and power consumption in constrained settings. In conclusion, CNNs—particularly when augmented with transfer learning—underpin AI systems for WCE lesion detection and classification. Ongoing improvements in architectures, generalization, and interpretability continue to drive their adoption. While CNNs and transfer learning remain central, they are not sufficient to capture all lesion types. Section 6.2 , therefore, examines AI methods for detecting specific gastrointestinal lesions such as polyps, bleeding, tumours, and erosions. 6.2 AI Detection of Specific Lesions in WCE: Polyps, Bleeding, Tumours, and Erosions Deep learning methods have achieved strong performance in GI lesions in WCE, including polyps, bleeding, tumours, and mucosal erosions/ulcers. Each lesion type presents unique visual challenges for both clinicians and algorithms. 6.2.1 Polyp Detection Polyps are difficult to identify due to their flat structures and similarity to mucosa [52]. CNNs such as VGG16, ResNet, and InceptionV3 remain standard, with a fine-tuned ResNet50 reporting 96.2% accuracy and 94.7% sensitivity on CVC-Clinic and Kvasir [35]. EfficientNetB0 with patch-based augmentation achieved an F1-score of 93.1% [29]. Transparency is supported by Class Activation Maps (CAMs), which highlight key regions [33]. 6.2.2 Bleeding Detection Bleeding is visually distinct but prone to false positives from artifacts [134]. While early RGB-based CNNs underperformed, modern models such as DenseNet121 and MobileNetV2 exceed 95% accuracy [31]. CNN–LSTM hybrids capture temporal continuity across frames [139], and augmentation with red-hue variations further improves generalization [137]. 6.2.3 Tumour Detection Tumours are rare and often resemble folds or polyps [32]. GAN-based data augmentation addresses dataset scarcity, with GAN–CNN hybrids achieving > 91% precision [135]. CNN–Transformer systems, such as Swin blocks, improved multi-lesion classification to 92.6% [44]. 6.2.4 Ulcers and Erosions These lesions are subtle and easily missed under poor lighting. Attention-based CNNs improved ulcer classification with an AUC of 0.963 [43]. A U-Net + CNN cascade reached 90.4% accuracy [34], while multi-label frameworks achieved F1-scores of 91.2% for ulcers and 92.8% for erosions on CAD-CAP [136]. Table 2 summarizes the performance of state-of-the-art AI models for detecting polyps, bleeding, tumours, and ulcers/erosions in WCE. The results highlight the diversity of architectures, evaluation metrics, and datasets considered in recent studies. Table 2 Performance of different models in the detection of polyps, bleedings, tumours, and erosions. Lesion Type Model(s) Metric(s) Dataset(s) Ref(s) Polyp ResNet50 96.2% Accuracy, 94.7% Sensitivity Kvasir + CVC-Clinic [35] EfficientNetB0 F1-score 93.1% Kvasir [29] CAM-based CNN Interpretability (qualitative) Kvasir [33] Bleeding DenseNet121 > 95% Accuracy Custom BleedDB [31] MobileNetV2 > 95% Accuracy Custom BleedDB [31] CNN–LSTM hybrid Temporal continuity Video sequences [139] Augmentation (red-hue) Improved generalization Custom DB [137] Tumor CNN + Transformer (Swin) 92.6% Accuracy SynthTumor + Kvasir [44] GAN–CNN hybrid > 91% Precision Synthetic augmentation [135] Ulcer/Erosion Attention-based CNN AUC 0.963 CAD-CAP [43] CNN + U-Net Cascade 90.4% Accuracy CAD-CAP [34] Multi-label framework F1: 91.2% (ulcer), 92.8% (erosion) CAD-CAP [136] As shown in Table 2 , AI methods achieve consistently high performance across different lesion types. Polyp detection benefits from deep CNNs such as ResNet and EfficientNet, while bleeding detection achieves > 95% accuracy using DenseNet and lightweight CNNs. Tumour detection remains more challenging, with hybrid Transformer-based models and GAN augmentation addressing data scarcity. For ulcers and erosions, attention-based CNNs and multi-label frameworks yield promising results on CAD-CAP datasets. Overall, hybrid and attention-driven architectures are the most effective strategies for robust lesion detection. 6.3 Real-Time AI and Edge Models in Wireless Capsule Endoscopy The integration of AI in WCE must balance diagnostic accuracy with latency, memory, and energy use. With each exam producing more than 50,000 images, real-time processing reduces physician workload and supports on-board diagnostics. 6.3.1 YOLO-Based Detection The YOLO family has been widely applied in WCE for real-time lesion detection. YOLOv3 achieved 34 FPS and 93.7% mAP for polyps [134], while lightweight variants such as YOLOv5s and YOLOv4-tiny achieved > 91% accuracy for bleeding/ulcers on embedded devices, cutting image transmission by over 80% [35], [34]. Customized anchors and multi-head YOLO models further improve localization and throughput [136]. 6.3.2 Transformer Architectures Transformers capture long-range dependencies and temporal dynamics in video frames. A Swin Transformer improved AUC by 4.6% over ResNet50 for bleeding/erosion detection [52], while hybrid CNN–Transformer models such as combined EfficientNet and ViT reached 95.1% accuracy on Kvasir [33]. These methods enhance lesion tracking but remain computationally demanding [135]. 6.3.3 Edge-AI and Lightweight CNNs On embedded platforms, compact models such as MobileNetV3 achieve 28 FPS with < 40 MB of RAM on the Raspberry Pi [29]. Quantized or pruned CNNs (e.g., 8-bit ResNet18) deliver 4× faster inference with minimal accuracy loss [32]. Early-exit cascades reduce computation by discarding easy cases, cutting costs by up to 38% [137]. EfficientNet-lite3 achieved 94.5% bleeding detection with < 40 ms latency at 3.2 W [43]. 6.3.4 Toward Onboard Diagnostics Prototypes employing ASIC-based CNNs achieve real-time lesion detection at < 100 mW [44], demonstrating the potential for fully autonomous in-body diagnostics. In summary, YOLO, transformers, and lightweight CNNs provide feasible real-time solutions, but hardware, power, and reliability constraints still limit deployment. Section 6.4 examines these challenges in detail. 6.4 Challenges in AI-Based Lesion Detection in WCE Despite major advances in deep learning for WCE, several challenges continue to limit the robustness, reproducibility, and clinical adoption of AI-based lesion detection systems. 6.4.1 Generalization and Domain Shift A key obstacle is poor generalization across datasets. Models trained on Kvasir, for instance, showed accuracy drops of up to 17% when tested on CAD-CAP or hospital-specific data [134]. These declines are linked to differences in imaging hardware, lighting, and annotation protocols. Domain adaptation techniques, such as CycleGAN or unsupervised transfer learning, have been explored, but they often distort lesion characteristics [135]. Stronger generalization will require larger multi-center datasets and domain-invariant feature representations [32]. 6.4.2 Annotation Quality and Label Ambiguity Supervised learning relies heavily on accurate annotations, which are costly and time-consuming to obtain. A single WCE video may require hours for gastroenterologists to review, and even then, disagreements arise over lesion boundaries, particularly in subtle cases such as small polyps or early-stage ulcers [52]. Crowdsourcing and semi-supervised labelling have been tested, but noisy labels reduce model accuracy. In ulcer detection, for instance, inaccurate annotations lowered CNN performance by up to 9% [35]. 6.4.3 Class Imbalance and Rare Lesion Detection WCE datasets are highly imbalanced, with normal frames comprising more than 85%, whereas tumours, erosions, and angiectasia account for less than 1% [34]. This imbalance biases models toward the majority classes. Techniques such as focal loss, SMOTE, cost-sensitive training, and multi-head attention have been evaluated [43]. Despite these methods, rare classes remain difficult: for example, angiectasia F1-scores often remain below 70% despite focal loss [31]. 6.4.4 Interpretability and Clinical Trust The “black box” nature of deep learning models limits clinical acceptance. Even accurate predictions may be disregarded by clinicians when explanations are unclear. Methods such as Grad-CAM, SHAP, and saliency maps help visualize decision-making, but they often yield inconsistent results in noisy WCE frames [137]. In some cases, clinicians dismissed correct predictions because the highlighted regions did not align with their expectations [139]. 6.4.5 Clinical Integration and Workflow Compatibility Many AI systems remain offline and are used retrospectively, limiting their usefulness for real-time decision support. Edge-AI enables in situ predictions but still faces hardware and regulatory hurdles. The absence of standardized metrics and limited regulatory approval—such as FDA clearance for WCE-specific AI—further slows adoption [53]. In summary, while AI has demonstrated remarkable diagnostic accuracy in controlled settings, challenges with dataset generalization, annotation reliability, detection of rare lesions, interpretability, and workflow integration continue to hinder clinical deployment. Overcoming these issues will require not only technical advances but also stronger collaboration among engineers, clinicians, and regulatory bodies to ensure real-world applicability. 6.5 Comparative Analysis of AI, IoT, and Compressed Sensing in WCE The following section provides a comparative analysis of AI, IoT, and compressed sensing in WCE, emphasizing how their integration collectively enables next-generation capsule systems. 6.5.1 AI – Diagnosis-Oriented Processing Deep learning automates lesion detection and improves accuracy, but requires large, annotated datasets and faces challenges such as domain generalization and interpretability [134], [43], [31], [32]. 6.5.2 IoT – Connectivity-Driven Infrastructure The Internet of Things enables telemetry, cloud-based data integration, and remote feedback within capsule endoscopy systems. This connectivity facilitates efficient data exchange between the capsule, physicians, and monitoring platforms. However, the IoT framework remains constrained by high energy consumption, transmission latency, and signal attenuation across the gastrointestinal (GI) tract. These factors collectively limit the reliability and continuity of wireless communication in WCE applications [52], [30], [33]. 6.5.3 CS – Efficiency-Oriented Transmission Compressed Sensing improves transmission efficiency by reducing data volume and minimizing power consumption. It achieves this by exploiting the inherent sparsity of biomedical signals during acquisition and reconstruction. Despite these advantages, CS-based methods remain highly sensitive to noise and often demand substantial computational resources. These limitations can affect the real-time applicability of CS in wireless capsule endoscopy systems [35], [135], [139]. 6.5.4 Comparative Overview To better highlight the complementary roles of Artificial Intelligence, the Internet of Things, and Compressed Sensing in next-generation Wireless Capsule Endoscopy, Table 3 summarizes their key features with respect to functionality, energy consumption, data requirements, clinical relevance, and integration challenges. Table 3 Comparative key features of AI, IoT, and CS in next-generation WCE. Feature AI IoT CS Core Function Lesion detection Telemetry & connectivity Data compression Power Use Medium–High High Low Data Needs Very high Moderate Low Clinical Impact High High Medium Integration Complexity High Moderate High 6.5.5 Synergistic Opportunities As shown in Table 3 , hybrid designs demonstrate strong potential. For example, AI-guided CS-based region-of-interest sampling reduced bandwidth by 78% without compromising accuracy [35]. When combined with lightweight IoT protocols such as MQTT or LoRa, these approaches enable adaptive, low-power capsules capable of real-time monitoring. In summary, the convergence of AI, IoT, and CS enables “smart capsules” with intelligent detection, efficient data handling, and connected workflows. Section 6.6 expands on open challenges and future research opportunities. 6.6 Future Research Opportunities and Open Challenges in Smart Capsule Endoscopy Despite major progress enabled by AI, IoT, and CS, significant barriers remain to the realization of fully autonomous WCE systems. Future research must address the following priorities: 6.6.1 Edge-Aware AI for Real-Time Processing Shifting computation from cloud servers to capsule hardware is essential for real-time triage and adaptive imaging. Ultra-efficient CNNs or TinyML models are being explored, but balancing accuracy and power consumption on constrained hardware remains unresolved [134], [30]. 6.6.2 Standardization of Data and Evaluation Protocols The lack of consistent datasets and benchmarks hampers cross-study comparison. Large-scale, clinically validated datasets—such as CAD-CAP and Kvasir—are valuable but still limited. Future collections should support multi-label annotations and temporal lesion tracking [52], [136]. 6.6.3 Multimodal and Adaptive Imaging Capsules Expanding beyond RGB imaging toward hyperspectral, 3D, or biochemical sensing could enable richer diagnosis. AI-triggered frame capture may also conserve energy by focusing on abnormal events [35], [32], [140]. 6.6.4 Explainability, Trust, and Ethical AI The black-box nature of AI limits clinical adoption. Approaches such as uncertainty estimation, clinician-in-the-loop training, and bias mitigation are required to ensure transparency, fairness, and regulatory acceptance [29, 33, 34]. 6.6.5 Privacy and Federated Learning Federated Learning (FL) enables privacy-preserving model training across institutions, aligning with GDPR and HIPAA. However, challenges remain in managing communication overhead and heterogeneous client data [137], [139]. In summary, addressing efficiency, data quality, multimodal integration, transparency, and privacy will be central to advancing WCE toward intelligent, trustworthy, and widely deployable smart capsules. 6.7 Interim Summary This review summarizes the integration of Artificial Intelligence, Internet of Things, and Compressed Sensing in Wireless Capsule Endoscopy. AI has advanced lesion detection, segmentation, and decision support, often reaching clinician-level accuracy in polyp and ulcer detection [134], [43], [32]. However, limited generalization, annotation cost, and lack of interpretability remain barriers to clinical use. IoT has expanded WCE by enabling wireless data synchronization, real-time monitoring, and remote feedback [33], [34]. Yet it introduces challenges related to energy use, latency, and privacy. CS addresses bandwidth and power constraints through sparsity-based compression, reducing transmission load while retaining diagnostic content [140], [136]. Its limitations lie in its sensitivity to reconstruction and its vulnerability to noise. Looking forward, the convergence of these domains is expected to produce “smart capsules” capable of adaptive imaging, selective transmission, and seamless integration with electronic health records [52], [35]. Key directions include: AI combined with advanced sensing (multi- spectral, biochemical). Privacy-preserving AI via federated learning [139]. Event-driven reporting for urgent cases. Interpretable AI to strengthen clinician trust [29]. In the long term, these innovations are likely to improve diagnostic speed, accuracy for rare pathologies, and accessibility in low-resource settings. With progress in miniaturization, edge computing, and low-power communication, WCE is poised to become a central tool in next-generation gastrointestinal care. To provide a structured perspective on how the three enabling technologies — Artificial Intelligence (AI), the Internet of Things (IoT), and Compressed Sensing (CS) — contribute to the evolution of Wireless Capsule Endoscopy (WCE), Table 4 summarizes their core roles, advantages, challenges, and representative applications. This comparative view highlights the complementary nature of these technologies and the areas requiring further development. Table 4 Benefits, challenges and implications of different technologies in WCE. Technology Core Role in WCE Advantages Challenges Applications Ref AI Automated lesion detection & classification High accuracy, reduced workload Generalization, explainability, and annotation cost Polyp detection, bleeding segmentation, tumour classification, edge AI [134], [43], [32], [29] IoT Connectivity & remote interaction Real-time telemetry, monitoring Bandwidth, latency, privacy, energy use Capsule navigation, cloud sync, remote feedback [33], [34], [35] CS Data compression & efficient transmission Low energy, reduced bandwidth Reconstruction quality, noise sensitivity Selective ROI transmission, compressed imaging [140], [52], [136] As shown in Table 4 , AI contributes diagnostic intelligence but is limited by generalization and interpretability; IoT provides connectivity at the cost of energy and privacy; and CS improves efficiency but remains sensitive to noise. Their convergence underscores the complementary strengths of these technologies in advancing next-generation WCE toward smart, connected, and adaptive systems. 7. IoT in Wireless Capsule Endoscopy The Internet of Things refers to a network of interconnected devices equipped with sensors, software, and communication technologies that enable seamless data exchange over the internet [141]. In healthcare, this concept is extended to the Internet of Medical Things (IoMT), which supports continuous monitoring and real-time analytics, enabling personalized and preventive care. As part of healthcare 4.0, IoMT—along with edge computing and AI—has become a central driver of modern clinical systems [142]. Wireless Capsule Endoscopy, a noninvasive technique for gastrointestinal imaging, has benefited greatly from IoT integration. IoT-based telemetry enables real-time imaging and wireless data transfer, allowing remote access to diagnostic information by external devices [123], [143]. This connectivity enhances patient comfort by reducing in-clinic visits and supports extended monitoring outside hospital settings [18]. IoT also enables interaction through real-time remote-control features within WCE, such as device reconfiguration, firmware updates, and data offloading—functions that are essential for smart capsule development [144]. In addition, multiple sensors and cloud-based synchronization open new opportunities for telemedicine and remote diagnostics [145]. These advances align with the concept of connected healthcare, where patients, devices, and clinicians form feedback loops [142]. Having established the role of IoT in WCE, Section 7.1 examines the architectural components that enable this advanced connectivity. 7.1 IoT Architecture in Smart Capsule Systems The architecture of IoT-integrated WCE systems is organized into three core layers: perception, network, and application. At the perception layer, the capsule integrates biosensors and RFID modules to capture pH, temperature, pressure, and imaging data—critical inputs for diagnostic evaluation [143], [48]. The network layer handles communication between the capsule and external receivers. Low-power wireless modules transfer data securely to gateways, which relay information to cloud or edge platforms. Decentralized edge/fog units reduce latency and energy load, enhancing Quality of Service (quality of service) [142], [26], [36]. At the application layer, clinicians access dashboards and analytics tools that convert sensor data into actionable insights, supporting real-time diagnostic decision-making [123], [17]. Overall, this three-layer architecture provides modularity and interoperability, linking embedded sensing with secure transmission and clinically relevant visualization [142]. The general IoT-enabled architecture for WCE systems is depicted in Figure. 4. As shown in the figure. 4, this framework establishes the foundation for real-time monitoring, which is further examined in Section 7.2 . 7.2 Real-Time Monitoring and Data Acquisition Wireless Capsule Endoscopy enables continuous monitoring of the gastrointestinal tract by collecting image and physiological data as the capsule moves through the digestive system [143]. Embedded sensors—including imaging modules and instruments for temperature, pH, and pressure—collect signals at fixed intervals and transmit them wirelessly to an external receiver, typically worn by the patient [123], [48]. Real-time transmission provides physicians with immediate access to physiological changes, improving diagnostic accuracy and supporting timely interventions [145]. Within IoT-enabled frameworks, the data can be relayed to cloud platforms for storage and visualization [146]. Live dashboards allow clinicians to track capsule progression in real time via mobile or desktop interfaces [147]. To improve efficiency, wearable receivers equipped with edge modules can perform basic preprocessing tasks such as image compression and noise filtering before data are uploaded [142], [36]. This approach reduces bandwidth demand and conserves capsule energy by minimizing unnecessary transmissions. Furthermore, these systems can generate alerts for abnormal readings, enabling remote clinical response [148]. The combination of real-time acquisition, embedded sensing, and wireless analytics transforms WCE into a responsive diagnostic platform. By linking raw physiological data with actionable insights, WCE provides a more intelligent approach to gastrointestinal monitoring [18], [149]. 7.3 Wireless Communication Protocols and Standards Reliable wireless communication is essential in WCE to ensure that real-time data from the capsule is transmitted effectively to external receivers and diagnostic platforms [48]. Several low-power protocols—Bluetooth Low Energy (BLE), ZigBee, LoRa, NB-IoT, and Wi-Fi—have been explored in this context. Table 5 summarizes the main wireless communication protocols explored for WCE, highlighting their trade-offs in data rate, range, power consumption, suitability, and limitations [131]–[133]. Table 5 The most well-known protocols for WCE. Protocol Data Rate Range Power Use Suitability for WCE Limitation BLE Up to 2 Mbps ~ 10 m Very Low Excellent for short-range, low-power Limited range, interference-prone ZigBee ~ 250 kbps ~ 100 m Very Low Good for mesh-style in-body networks Low bandwidth LoRa 1 km (outdoor) Low Suitable for remote monitoring Very low data rate NB-IoT ~ 100 kbps Nationwide Moderate Secure, cloud-connected telemetry High power demand Wi-Fi Up to 54 Mbps ~ 50 m High Prototype experiments Not feasible for battery capsules As shown in Table 5 , selecting an appropriate protocol requires balancing trade-offs between range, throughput, energy, and security. In some designs, hybrid approaches are adopted—for example, BLE for short-range relay combined with NB-IoT for cloud uploads—providing both real-time monitoring and secure long-distance connectivity [144], [145]. 7.4 Edge Computing vs. Cloud Computing in WCE Cloud computing provides extensive storage and computational resources, making it well-suited to handling the large data volumes generated by WCE [141]. Capsule sensor outputs can be transmitted to cloud platforms, where advanced algorithms—including AI-based diagnostic models—interpret the data [26]. This approach also supports remote collaboration among clinicians and seamless integration with Electronic Health Records (EHRs) [142]. However, transmitting all raw data to the cloud raises major concerns. It consumes high bandwidth and introduces privacy risks, as sensitive medical information must be protected under strict regulations [150]. To mitigate these limitations, edge computing has emerged as a complementary approach, moving computational tasks closer to the data source—such as wearable receivers or local gateways [123]. Edge computing enables near-real-time functions such as image enhancement, anomaly detection, and preliminary pattern recognition before forwarding compressed or summarized data to the cloud [36]. This not only reduces the processing burden on the capsule but also conserves energy by minimizing unnecessary transmissions [143]. While the latency advantages of edge systems were discussed in Section 7.3 , this section focuses on computational offloading and security benefits. Hybrid architectures combine both paradigms: edge nodes manage time-sensitive tasks, while cloud servers perform resource-intensive analytics and provide long-term storage [17]. This model improves scalability, reliability, and adaptability to diverse clinical contexts [144]. Ultimately, the choice among edge, cloud, and hybrid frameworks depends on application-specific requirements, including energy constraints, bandwidth availability, and data security requirements [48]. Processing distribution decisions directly affects both the sustainability and the trustworthiness of WCE platforms. Section 7.5 further examines energy-consumption challenges and optimization strategies. 7.5 Energy Efficiency and Power Management via IoT Energy consumption is a critical design challenge in WCE, as capsules must operate autonomously inside the gastrointestinal tract for extended periods [143]. IoT-based architectures enhance power efficiency by dynamically adjusting parameters such as sampling frequency and transmission rate in response to real-time physiological signals [123]. For instance, wake-up mechanisms can trigger imaging or telemetry only when motion or temperature thresholds are exceeded [147]. Low-power communication standards such as BLE and ZigBee further reduce energy usage [26]. Hardware design also plays a major role: ultra-low-power microcontrollers and CMOS image sensors help extend battery life while preserving diagnostic quality [26]. When coupled with edge computing, these strategies have demonstrated more than 30% power reduction relative to traditional architectures [36]. On the software side, adaptive compression, event-driven activation, and smart transmission scheduling are increasingly adopted [150]. Energy-aware routing and sleep-mode algorithms within IoT frameworks also lower power consumption in capsule-like medical devices [151–154]. Beyond conservation, researchers are exploring energy-harvesting methods—including thermoelectric and biomechanical approaches—to enable semi-autonomous or self-powered capsules [48]. These technologies could extend the operational lifetime and enable richer, higher-resolution imaging without enlarging the capsule size [17]. While power efficiency is essential for capsule reliability, it must evolve in tandem with robust data protection and privacy safeguards. 7.6 Security and Privacy Concerns in IoT-Enabled WCE The integration of IoT into WCE introduces cybersecurity and privacy risks, mainly due to the wireless transfer of sensitive medical data [155],[156],[7]. As information moves across system layers—from embedded sensors to gateways and finally cloud platforms—it becomes vulnerable to interception, manipulation, and breaches [123]. Studies show that many existing protocols still lack sufficient encryption for healthcare applications [150]. A multi-layered security strategy is therefore essential. Strong encryption methods, such as AES, secure device authentication, and blockchain-based audit trails, can enable tamper-resistant logging [157]. End-to-end encryption between the capsule and cloud platforms is increasingly viewed as a baseline requirement for IoMT healthcare [26]. However, implementing advanced cryptography is technically challenging due to the capsule's limited computing power and energy constraints [48]. Access control represents another key concern. Without safeguards, unauthorized users could access diagnostic reports or alter transmitted data [17]. Modern systems address this by employing role-based access control, biometric verification, and secure dashboards [144]. Firmware-Over-The-Air (FOTA) updates are also critical for maintaining system integrity and patching vulnerabilities [18]. Legal and ethical compliance further underpins system security. Regulations such as HIPAA and GDPR ensure that patient data is handled responsibly and lawfully [142]. Effective cybersecurity must therefore be integrated into the design of WCE systems, balancing technical safeguards with regulatory and ethical accountability [36]. With these foundations in place, the next step is to review IoT-enabled prototypes that incorporate such security features into practical WCE deployments. Section 7.7 explores these case studies. 7.7 Case Studies and Recent Research in IoT-Enabled WCE Recent research has advanced the integration of IoT into WCE systems. One notable effort designed a smart capsule using BLE to stream both image and physiological data to mobile applications, enabling real-time ambulatory monitoring [123]. Another project developed a multi-sensor capsule with pH, temperature, and pressure sensors, transmitting data to the cloud via a local gateway. This approach supported remote diagnostics and reduced reliance on prolonged hospital stays [147], [18]. A 2021 study proposed an IoT-based WCE platform in which onboard machine-learning algorithms performed local anomaly detection and transmitted only flagged data to the cloud. This design reduced transmission volume and conserved battery life [36]. In a separate experiment, real-time localization was implemented using embedded inertial sensors and RF triangulation, managed through an IoT framework to improve capsule tracking [48]. Blockchain has also been explored to secure IoT-enabled WCE systems. Decentralized audit trails and secure access logs were integrated to enhance data privacy [157]. Another implementation employed NB-IoT over public cellular networks to transmit encrypted diagnostic data directly to the cloud [158],[159]. Hybrid architectures are also gaining traction. Edge units handle preprocessing, preliminary classification, and alert generation locally, while cloud resources provide long-term storage, historical trend analysis, and advanced diagnostics [142]. Together, these prototypes illustrate the convergence of IoT, edge intelligence, and real-time medical imaging in next-generation endoscopic diagnostics [26]. Despite these promising implementations, several technical and operational challenges persist, as outlined below. 7.8 Challenges and Future Perspectives Despite significant progress in IoT-enabled WCE, several technical and systemic challenges remain. Power limitation is a primary constraint, as capsules must function for extended periods without external energy sources [123]. Wireless transmission also faces reliability issues in the gastrointestinal environment, where tissue attenuation and interference are particularly pronounced in deeper regions [143], [147]. A major design trade-off exists between ensuring high data fidelity and conserving energy. Capturing and transmitting high-resolution images increases both power demand and memory requirements, which conflicts with the capsule’s low-power hardware [48]. Secure and private data transfer across sensors, networks, and cloud layers is also difficult to guarantee, given the capsule’s limited processing capacity [160],[16]. System-level barriers further complicate deployment. Interoperability challenges arise from diverse communication standards, inconsistent data formats, and fragmented IoT infrastructures [142]. To address this, researchers are pursuing standardized data ontologies and flexible middleware designed for medical IoT applications [26]. Future technologies offer promising solutions. Next-generation wireless systems, such as 6G, are expected to deliver ultra-reliable, low-latency communication (see Section 7.3 ) and higher spectral efficiency, thereby improving real-time telemetry in WCE [17]. Advances in AI are enabling autonomous IoT frameworks that can adjust acquisition strategies, perform onboard anomaly detection, and tailor diagnostics [36]. Energy autonomy is another frontier. Self-powered capsules that use biocompatible energy harvesters (e.g., thermoelectric or kinetic converters) are being investigated to extend operational life without increasing device size [150]. Looking ahead, the fusion of 6G wireless, edge AI, and biomedical IoT is poised to drive a new generation of intelligent diagnostic capsules [157]. This evolution marks a transition from passive data collection to proactive, connected, and personalized gastrointestinal diagnostics. Although challenges in energy efficiency, security, and standardization persist, the convergence of these technologies provides a clear pathway toward fully autonomous, secure, and clinically valuable WCE systems. 8. Compressed Sensing Compressed Sensing has emerged as a disruptive paradigm for signal acquisition and image compression, enabling efficient sampling and accurate reconstruction compared to conventional methods [161]. Its key elements include sparsity-driven acquisition strategies, advanced reconstruction algorithms, and specialized analog-to-digital converters [40]. In medical imaging, particularly MRI, CS integrated with parallel imaging reduces scan duration while preserving diagnostic quality [162]. In wireless biomedical systems, CS mitigates power and bandwidth constraints by compressing signals before transmission, thereby enabling real-time monitoring on resource-limited platforms [163]. Applications also extend to photoacoustic imaging, where CS achieves high-fidelity reconstructions at reduced sampling rates [23]. Recent advances combine compressed sensing with artificial intelligence, enabling diagnostic-quality imaging even under aggressive compression [164]. Hybrid strategies, such as CS-2FFT, have further enhanced signal characterization in capsule endoscopy [165]. These developments highlight CS as a foundation for next-generation WCE; however, most approaches remain restricted to research prototypes and require further validation for clinical integration. 8.1 Sparsity and Incoherence in Biomedical Signals Sparsity is fundamental for the efficient acquisition and transmission of biomedical signals. EEG- and ECG-based systems demonstrate that physiological data can often be represented with only a few dominant coefficients in wavelet or Fourier domains [166], [41]. Multi-layer convolutional sparse coding has further advanced compression performance in biomedical imaging [10]. MRI benefits greatly from sparsity-exploiting strategies, where reduced sampling shortens acquisition times without compromising diagnostic quality [162], [163]. Similar principles apply in wearable and implantable systems, where exploiting signal sparsity accelerates both transmission and downstream processing [161], [23]. A cornerstone of Compressed Sensing (CS) is incoherence—the sensing matrix must remain uncorrelated with the sparsifying basis to ensure accurate reconstruction from incomplete data [167]. Biomedical signals often meet these requirements naturally due to periodicity or structured patterns. To further enhance fidelity, techniques such as dictionary learning and transform-based sparsity models are increasingly adopted in clinical applications [42]. Having established the role of sparsity and incoherence, Section 8.2 explores the application of CS principles to image compression in WCE systems. 8.2 Compressed Sensing for Image Data Compression in Wireless Capsule Endoscopy CS allows image acquisition and reconstruction with far fewer samples than conventional Nyquist-based methods, a major advantage for WCE where energy, memory, and bandwidth are highly constrained [161], [41]. By reducing the amount of data transmitted, CS improves efficiency and enables near-real-time gastrointestinal monitoring. Recent developments integrate CS with deep learning-based reconstruction frameworks, achieving high-resolution images with low power consumption and confirming feasibility for capsule-based diagnostics [164], [168]. CS has also shown strong performance in biomedical photoacoustic imaging, further highlighting its diagnostic potential [23]. Performance evaluations often rely on metrics like PSNR, with reported values consistently above 33 dB, supporting clinical reliability [169]. Hardware-oriented innovations—such as modulo CS frameworks for ADCs—help mitigate dynamic-range limitations [40]. At the same time, the combination of CS with parallel imaging continues to enhance image fidelity and reduce scan times [162]. Building on these compression strategies, Section 8.3 discusses how CS contributes to localization and tracking in capsule endoscopy. 8.3 Compressed Sensing-Based Localization Techniques in Wireless Capsule Endoscopy Compressed sensing has been proposed for WCE to lower on-capsule power by reducing the number of transmitted/measured samples, while maintaining clinically useful reconstruction quality. Recent CS systems for wearable/implantable sensing report energy savings of 50% or more, supporting their feasibility for real-time or near-real-time WCE data reduction [170],[111]. Sparse Bayesian learning and graph-structured models further improve estimation reliability, allowing accurate trajectory recovery with fewer projection measurements [171]. Advanced methods such as compressed trajectory sensing and super-resolution combined with CS enhance angular precision and navigation fidelity [172], [37]. Compressive magnetic mapping has also shown robustness against disturbances, supporting clinical deployment [37]. Incorporating prior motion models or random projection matrices strengthens trajectory estimation under complex anatomical conditions [172], while sparse approximation enables path reconstruction even without GPS or visual markers [165]. Hybrid strategies integrate CS with Kalman filtering or compressive priors for predictive motion modelling [40]. Compressed magnetic signals have been shown to capture both positional and orientational data with minimal sensor input [167]. More recently, deep learning models trained on sparsified sensor data have improved geometry-aware localization, making systems more resilient to noise and patient variability [93]. In summary, CS-based localization offers precise, energy-aware, and robust solutions for capsule tracking, addressing critical clinical and technical requirements. Section 8.4 transitions to hardware–software co-design strategies for practical deployment. 8.4 Hardware and Algorithm Co-Design for Compressed Sensing in Wireless Capsule Endoscopy Implementing CS in WCE is constrained by capsule size, limited energy, and low computational capacity. Hardware–software co-design has therefore become essential. Ultra-low-voltage acquisition modules reduce power consumption while maintaining fidelity [169], and ASICs that embed CS reconstruction engines minimize computational load without compromising signal integrity [47]. FPGAs have been explored to exploit parallelism for real-time reconstructions, achieving sub-millisecond delays with application-specific sparsity models [42], [173]. Due to memory constraints, compressed-domain processing is often adopted, with dictionary learning and custom bases implemented directly in circuits to improve efficiency [174]. Adaptive methods, such as approximate computing and reconfigurable logic, also support dynamic adjustment to variations in resolution and sparsity [93]. At the sensor level, compressive ADCs encode signals via multiplexing, eliminating the need for raw data storage and enabling reconstruction using algorithms such as OMP and CoSaMP [47], [49]. These circuit-level innovations illustrate how CS can be embedded into compact hardware platforms while maintaining diagnostic fidelity. Overall, co-optimized pipelines that integrate CS algorithms with specialized electronics demonstrate strong potential to enable capsules capable of onboard processing and real-time in vivo diagnostics under strict resource constraints [42]. Section 8.5 next introduces evaluation metrics for assessing CS performance in WCE. 8.5 Evaluation Metrics for Compressed Sensing in WCE Evaluating CS in WCE requires examining both reconstruction fidelity and system-level performance. Image quality metrics remain the most widely used: Peak Signal-to-Noise Ratio (PSNR) values above 30 dB typically indicate strong similarity to original frames [175]. In contrast, Structural Similarity Index (SSIM) values exceeding 0.85 in block-based WCE experiments confirm preserved diagnostic reliability [162], [176]. Error-based measures such as Root Mean Square Error (RMSE) and Normalized Mean Square Error (NMSE) provide additional benchmarks, with RMSE < 5% generally regarded as clinically acceptable [177]. System-level indicators complement image fidelity. Studies have shown that combining CS with energy-aware modulation strategies reduces the per-bit transmission cost, and hybrid CS–inertial architectures significantly reduce frame-delivery latency [42, 131, 178–180]. Data compactness is often quantified by the sparsity ratio, with optimized pipelines achieving 10–20% reductions in the number of coefficients [169]. Beyond quantitative results, clinical validation remains essential. Physician-based evaluations report that more than 85% of CS-reconstructed frames are judged diagnostically equivalent to originals [37]. Taken together, these metrics highlight CS as a robust and clinically reliable method for efficient image acquisition, transmission, and interpretation in WCE. 8.7 Future Trends and Research Directions in Compressed Sensing for WCE The future of CS in WCE is expected to evolve through adaptive and intelligent frameworks that address real-time constraints, energy limitations, and clinical accuracy requirements. A central direction is the shift toward edge-computing capsules, in which compressed-domain AI modules analyze sparsified data directly on board. This approach eliminates the need for full reconstruction or continuous cloud support, thereby reducing latency and energy consumption while enabling closed-loop diagnostic operation [167], [170], [32], [181]. Another emerging trajectory is multimodal integration, in which CS frameworks jointly process imaging, inertial, and physiological signals. Instead of treating each modality separately, next-generation capsules will exploit joint sparsity to achieve efficient data use while preserving diagnostic and localization accuracy [165], [37]. Looking further ahead, federated learning, quantum-inspired CS algorithms, and direct compressed-domain classification represent forward-looking directions. These innovations aim to minimize power use, enhance real-time responsiveness, and ensure robust diagnostic reliability across diverse patient populations. Table 6 summarizes representative compressed sensing (CS) strategies in WCE, covering imaging, localization, AI-assisted recovery, hardware-aware implementations, graph-based models, and physiological signal compression. Each method highlights distinct design trade-offs with respect to application, advantages, and limitations. Table 6 CS-based methods for WCE problems and their corresponding characteristics. CS Method Application Advantages Challenges References Image-based CS (Image-CS) GI image sequences from WCE Reduced transmission load; preserved diagnostic quality Degradation at high compression ratios [182], [166], [168] Localization-based CS Magnetic/inertial data Energy-efficient trajectory estimation; fewer sensors Sensitive to motion models; noise susceptibility [37], [172], [93] AI-integrated CS (Deep-CS) Compressed data with AI Higher accuracy; learned priors; adaptability Requires large datasets; computational demand [183], [184], [161] Hardware-aware CS (HW-CS) FPGA / ASIC implementations Real-time feasibility; low-power deployment Memory/frequency limits; HW–algorithm mismatch [47], [42] Graph-based CS (Graph-CS) Spatially structured biomedical data Captures anatomical/structural relations Requires graph modelling; higher complexity [167], [93] Physiological Signal CS EEG, EMG, temperature, pulse Multi-parameter low-data-rate monitoring Risk of reduced clinical resolution at low sampling [40], [165] As shown in Table 6 , each CS strategy offers distinct advantages and limitations, emphasizing the importance of application-specific optimization. Collectively, these approaches underline CS as a transformative enabler of future WCE platforms by aligning advances in AI, hardware co-design, and multi-modal sensing. 9. Overview and Objectives of the Comparative Study Wireless Capsule Endoscopy is a pivotal tool for non-invasive gastrointestinal diagnostics; however, current systems still face challenges, including limited diagnostic accuracy, high energy consumption, and transmission delays [185], [8]. To address these issues, advanced approaches in Artificial Intelligence, Internet of Things, and Compressed Sensing are being increasingly adopted. AI enhances lesion detection by improving sensitivity and specificity for bleeding, polyps, and ulcers, while also reducing the review burden on physicians [186]. IoT enables real-time data transfer and remote monitoring, ensuring timely clinical response [187]. CS mitigates power and bandwidth constraints by enabling signal reconstruction from fewer samples, making it suitable for compact, low-energy devices [188]. These technologies not only provide distinct benefits but also demonstrate potential synergy, forming the backbone of next-generation WCE [189]. The objective of this comparative study is to evaluate their respective strengths, limitations, and clinical relevance in settings such as remote monitoring, Crohn’s disease surveillance, and the detection of obscure gastrointestinal bleeding [19]. The evaluation applies key performance indicators—diagnostic accuracy, energy efficiency, integration feasibility, scalability, and real-time capability—alongside Technological Readiness Levels (TRLs) and regulatory perspectives [51], [190]. For example, the study contrasts AI-based image classification with CS-driven compression in resource-limited contexts and examines the scalability of IoT-enabled platforms for chronic care [53]. Ultimately, the goal is to guide the selection—or integration—of these technologies to support context-specific diagnostic tasks in WCE [45], [191]. 9.1 Evaluation Criteria and Methodology To compare AI, IoT, and CS in WCE, a standardized framework was applied, focusing on diagnostic accuracy, computational demand, energy efficiency, latency, and integration feasibility [19]. For AI, diagnostic sensitivity is central, with advanced deep learning models achieving up to 95% accuracy for bleeding detection [186]. CS is assessed through compression ratio, reconstruction fidelity, and image quality; one framework reported 80% compression with preserved diagnostic value [188]. IoT evaluations emphasize energy use, reliability, and real-time data transmission, with packet loss and uptime as key indicators [187]. The methodology combines literature review, simulations, and clinical validation. Over 120 peer-reviewed studies were analyzed to compare TRL, energy demand, and clinical applicability [51], [190]. AI energy profiling used TensorFlow Lite on embedded processors [45], whereas CS frameworks were evaluated using PSNR and SSIM metrics in MATLAB [46]. The Analytical Hierarchy Process (AHP) balanced these indicators relative to system objectives [192]. Real-world cases, including BLE-enabled IoT capsules in telemedicine [193] and deep-learning polyp detection [8], were incorporated. Security and privacy were also considered, with blockchain-enhanced IoT platforms ensuring secure data transfer [194]. In summary, the framework defines six evaluation criteria—diagnostic accuracy, power consumption, real-time capability, integration feasibility, TRL, and data security—that underpin the subsequent cross-technology analysis. Table 7 defines the evaluation criteria applied in this comparative analysis, including diagnostic accuracy, power consumption, real-time capability, integration feasibility, technology readiness level (TRL), and security. Table 7 defined criteria for comparative analysis. Criterion Definition Diagnostic Accuracy Correct identification of abnormalities such as bleeding, ulcers, or polyps. Power Consumption Average operational power requirement, measured in milliwatts (mW). Real-Time Capability System responsiveness is measured as the latency between sensing and reporting. Integration Feasibility Ease of incorporating the technology into existing WCE hardware designs. TRL (Technology Readiness Level) Scale (1–9) indicating the maturity of a technology for deployment. Security Protection of patient data against unauthorized access or tampering. As shown in Table 7 , the evaluation framework incorporates both technical and clinical considerations, ensuring that AI, IoT, and CS are compared consistently with respect to diagnostic reliability, efficiency, integration feasibility, maturity, and data protection. 9.2 Diagnostic Accuracy and Clinical Effectiveness The integration of AI into WCE has markedly improved diagnostic precision. CNN models, such as Ding et al. [8], achieved nearly 100% accuracy for bleeding and normal frame classification. In comparison, other deep learning and ResNet-based methods reported > 98% specificity in polyp detection and robust multi-lesion performance [186], [189]. These results demonstrate AI’s ability to enhance reliability, reduce false positives, and minimize physician workload. IoT technologies, although they do not directly improve classification accuracy, enable continuous, real-time monitoring and rapid response. Prototypes have transmitted diagnostic data to mobile devices for emergencies [193], while BLE-enabled capsules maintained uninterrupted high-resolution streaming with < 5% packet loss over short ranges [187]. These advances highlight IoT’s role in timely diagnosis and expanded accessibility, particularly in remote settings. CS ensures image quality under strict resource limits. Studies reported > 94% fidelity retention at 70% undersampling [46], and diagnostic adequacy was validated through anonymous reviews [188]. CS-based acquisition also accelerated capsule transit without reducing accuracy [51]. In summary, AI delivers state-of-the-art diagnostic accuracy, IoT enables responsive real-time connectivity, and CS optimizes image acquisition under energy and bandwidth constraints. Together, they illustrate the importance of selecting or combining technologies according to clinical needs [19]. 9.3 System-Level Constraints and Hardware Feasibility Wireless Capsule Endoscopy systems are restricted by capsule dimensions (< 1 cm³), limited energy storage, and thermal challenges. These constraints hinder the integration of advanced AI models, since most deep networks exceed the computational capacity of embedded microcontrollers [195], [186]. To cope, lightweight approaches such as TinyML or quantized CNNs have been adopted, though often at the cost of reduced diagnostic accuracy [45]. Energy consumption remains a central bottleneck. Standard 3 V coin-cell batteries support only 8–12 hours of operation, while data transmission alone can require 50–100 mW [46]. IoT modules increase demand, as continuous communication requires BLE or Wi-Fi protocols to adopt intermittent or event-driven signalling to conserve energy [193]. Excessive consumption also poses a risk of overheating, which can affect both electronics and surrounding tissue [191]. Compressed sensing provides partial relief by reducing the number of samples and lowering the transmission frequency [188]. However, reconstruction typically must be offloaded to external processors, which requires specialized analog hardware, such as random-sampling circuits and tailored ADCs. These additions complicate miniaturized designs [53]. Cost is another critical factor. BLE-based microcontrollers are relatively inexpensive, whereas AI-oriented accelerators, such as EdgeTPU, are both costly and oversized for capsule integration [147]. AI frameworks tend to increase Bill of Materials (BOM) and energy requirements [185], whereas CS shifts complexity to external computation [188]. IoT module costs vary depending on chipset, protocol, and whether energy harvesting is included [187]. In summary, each enabling technology faces distinct trade-offs at the system level: AI increases diagnostic power but struggles with size and energy limits; IoT enables real-time connectivity but raises power consumption and cost; CS reduces energy and bandwidth load but relies on complex external reconstruction. Table 8 summarizes the main system-level trade-offs of AI, IoT, and CS in WCE, highlighting their respective advantages, limitations, and overall impact on capsule feasibility. Table 8 System-level comparison of AI, IoT, and CS in capsule endoscopy. Tech Advantages Limitations System-Level Impact AI High diagnostic accuracy; supports automated lesion detection; potential for real-time decision support High computational demand; poor fit for capsule microcontrollers; increases BOM cost and power usage [186], [45] Improves diagnostic reliability but is constrained by size, energy, and thermal limits IoT Enables real-time telemetry; supports remote monitoring and integration with EHRs; event-driven signalling reduces delays [187], [193] High continuous energy demand; risk of overheating; variable cost depending on protocols and chipsets [191] Enhances connectivity and accessibility, but drains limited capsule resources CS Reduces data transmission load; lowers energy and bandwidth usage; offloads computation externally [188] Reconstruction heavy; requires specialized analog hardware; increases design complexity [53] Efficient at managing resource limits, but depends on external processing for clinical usability As shown in Table 8 , AI offers superior diagnostic accuracy but is constrained by computational and energy demands; IoT enables real-time connectivity but significantly increases power consumption and cost; and CS reduces energy and bandwidth usage but relies heavily on external reconstruction. These trade-offs emphasize the need for hybrid approaches that balance diagnostic performance with the strict physical and energy constraints of capsule systems. 9.4 Data Handling and Transmission Dynamics Efficient data handling remains a major challenge in WCE, as a single session may generate 50,000–60,000 images (several gigabytes) that must be stored or transmitted [185]. This scale strains bandwidth, memory, and latency. IoT-based capsules use wireless protocols such as BLE and Wi-Fi, achieving rates up to 2 Mbps in line-of-sight conditions [193]. Within the body, however, packet loss can reach 7% [187], jeopardizing diagnostic continuity unless mitigated by error correction or retransmission. Compressed Sensing reduces this burden by reducing the number of samples by up to 80%, thereby lowering payload and latency [188]. Reconstructions with only 25% of data still achieve diagnostic quality (> 30 dB PSNR) [46]. By eliminating heavy encoding, CS suits power-limited capsules [53], though reliance on external reconstruction and sparsity assumptions raises concerns for real-time workflows. AI brings different trade-offs. Lightweight models such as MobileNet and SqueezeNet operate on compressed inputs [186] but still produce large intermediate feature maps, thereby increasing memory and communication requirements [45]. Saliency-based filtering addresses this by transmitting only the most relevant frames [19], balancing efficiency with clinical utility. In summary, IoT enables real-time connectivity but faces bandwidth constraints; CS reduces transmission load while relying on robust reconstruction; and AI supports intelligent prioritization while increasing computational demand. Hybrid systems—combining CS for compression, AI for selective transmission, and IoT for real-time links—emerge as the most promising solution for scalable WCE [195]. 9.5 Synergistic Integration and Hybrid Architectures Recent developments in Wireless Capsule Endoscopy emphasize the integration of AI, CS, and IoT into hybrid systems that leverage complementary strengths. For example, a capsule integrating AI-based lesion detection, CS-driven compression, and BLE-enabled IoT reduced power consumption by 45% while maintaining > 90% diagnostic accuracy [189]. The key advantage lies in task distribution: CS reduces image volume during acquisition, enabling lightweight AI models to run locally or with minimal transmission [188]. At the same time, IoT edge devices handle compressed data without relying heavily on cloud processing [187]. One prototype employed CS-based sampling, onboard CNN classification, and BLE transmission, achieving < 300 ms latency and more than 10 hours of operation. AI-driven control logic further optimized transmission modes based on anatomical context [186]. Other studies confirm this trend. In [67], TinyML-optimized CNNs combined with CS achieved a 43% power reduction compared with full-frame streaming while maintaining > 90% sensitivity. Another design [193] employed BLE telemetry and localized AI logic to activate modules only upon anomaly detection, conserving energy and supporting real-time alerts. Despite the promise, challenges remain. CS requires analog front-end hardware and specialized ADCs, complicating mass production [192]. Embedding AI inference engines within capsule constraints demands highly optimized, energy-efficient designs [45]. IoT modules must also ensure secure, real-time wireless communication, particularly in cloud-assisted workflows [194]. Overall, hybrid integration offers the most practical path forward. By integrating AI, CS, and IoT, capsules can evolve into adaptive closed-loop diagnostic platforms that detect abnormalities, adjust imaging strategies, and tailor communication in real time [67]. Such systems align with the broader vision of personalized, intelligent, and autonomous gastrointestinal diagnostics [190]. 9.6 Clinical Scenarios and Technology Readiness Evaluating the deployment of AI, IoT, and CS in WCE requires both clinical context and Technology Readiness Level (TRL) assessment. Among them, AI is the most mature, reaching TRL 7–8, with clinical trials consistently showing > 90% sensitivity for detecting bleeding, ulcers, and polyps [186]. IoT-enabled platforms using BLE or Wi-Fi are at TRL 6–7, having been validated in prototypes and pilot studies but not yet widely adopted [193]. CS remains at an earlier TRL (4–5) due to limited in vivo testing and hardware integration challenges, although its efficiency in energy and data reduction is promising [188]. Clinically, AI has demonstrated clear benefits. In a 200-patient study of obscure gastrointestinal bleeding, CNN-based detection reduced localization time by 35% compared with manual review [8]. IoT-enabled capsules have supported chronic disease monitoring, such as Crohn’s, by transmitting near real-time alerts during capsule progression [187]. These illustrate the impact of AI and IoT in enhancing precision, responsiveness, and accessibility, particularly for remote or emergency care. CS has demonstrated experimental utility, as in a pilot study of celiac disease in which sparsely sampled images (70% reduction) reconstructed via dictionary learning yielded 82% diagnostic acceptability [189]. However, inconsistent clarity limited physician confidence, underscoring the need for higher reconstruction fidelity. Hybrid AI–IoT capsules have also proven effective in emergency response; one prototype transmitted AI-based anomaly alerts via BLE with < 5-second latency, improving triage speed and diagnostic throughput [193]. Regulatory pathways remain decisive. AI-based imaging requires extensive validation, clinical trials, and certifications (FDA, CE), often delaying adoption [190]. IoT systems must comply with strict privacy and security rules (HIPAA, GDPR). CS, being post-processing oriented, faces fewer regulatory barriers but still depends on clinical validation and physician acceptance. In summary, readiness varies: AI is the most advanced and clinically proven, IoT is rapidly advancing in remote diagnostics, whereas CS remains experimental but shows strong potential for energy-efficient, miniaturized WCE systems. 9.7 Comparative Summary Table and Visualizations This section synthesizes insights from the previous analysis by comparing AI, IoT, and CS across key evaluation criteria, including diagnostic accuracy, power consumption, real-time capability, integration feasibility, security, hardware requirements, and technology readiness level (TRL). Table 9 provides a comparative summary of these technologies, highlighting their relative performance and clinical applicability in WCE. Table 1 Comparative performance metrics of AI, IoT, and CS in wireless capsule endoscopy systems. Dimension AI IoT CS Diagnostic Accuracy High (90–95%) [186] Medium [187] Moderate–High [188] Power Consumption High (100–250 mW) [45] Moderate–High (50–100 mW) [193] Low (20–50 mW) [188] Real-Time Capability Limited (offboard preferred) [186] Strong (BLE/Wi-Fi) [193] Post-processed preferred [189] Integration Feasibility Medium–Low [195], [147] High [193] Low–Medium [53], [192] Security & Privacy Medium (cloud dependency) [194] Variable (BLE vulnerabilities) [194] High (local, compressed data) [188] Hardware Demand High (EdgeTPU / CNN) [45] Moderate (MCU + BLE) [147] Low (sensor + ADC) [53] TRL TRL 7–8 [186] TRL 6–7 [193] TRL 4–5 [188] Key Clinical Use Cases OGIB, polyp and bleeding detection [8] Telemedicine, Crohn’s monitoring [187] Sparse imaging, celiac detection [189] To complement the tabular comparison, Fig. 5 provides a radar visualization that highlights the relative strengths and weaknesses of AI, IoT, and CS across the same evaluation dimensions. As shown in the figure. 5, each technology exhibits unique advantages—AI excels in diagnostic accuracy, IoT in real-time connectivity, and CS in efficiency. Hybrid architectures that integrate these strengths represent a strategic direction for next-generation WCE platforms. 10. Discussion The structured analysis presented in this review reveals a decisive paradigm shift in wireless capsule endoscopy (WCE) research—from isolated hardware-centric localization mechanisms toward intelligent, system-level architectures that integrate sensing, computation, and communication. Early-stage developments were predominantly grounded in physics-based localization strategies, including magnetic-field modelling, inverse electromagnetic formulations, and state-space estimation frameworks [18–20], [28]. These approaches provided mathematical interpretability and deterministic guarantees but were inherently sensitive to anatomical variability, magnetic distortion, and modelling inaccuracies in complex gastrointestinal environments. In contrast, recent research trajectories demonstrate a strong migration toward data-driven intelligence. AI-based lesion detection frameworks, particularly convolutional neural networks (CNNs), have substantially redefined diagnostic performance benchmarks. Reported accuracies of up to 95% for bleeding detection [186] and near-perfect classification performance in controlled settings [8] significantly exceed those of traditional feature-engineering pipelines. Furthermore, deep residual architectures have demonstrated > 98% specificity in polyp detection tasks [186], [189], confirming the robustness of learned hierarchical feature representations in high-dimensional perceptual domains. These findings collectively indicate that diagnostic intelligence in WCE is increasingly software-defined rather than sensor-limited. However, performance superiority alone does not equate to deployability. As highlighted within the comparative evaluation framework [19], learning-based systems remain heavily dependent on dataset composition, annotation quality, and domain consistency. Model generalization across diverse patient populations, imaging protocols, and device manufacturers remains insufficiently validated. Moreover, the limited interpretability of deep neural architectures raises regulatory and clinical trust concerns, particularly in safety-critical diagnostic workflows. Without standardized cross-institutional validation and explainability-aware model design, large-scale translational integration may remain constrained. Beyond perception, IoT-enabled architectures expand WCE from a passive imaging modality to an active node within smart healthcare ecosystems. BLE-enabled capsule platforms have demonstrated reliable real-time telemetry with packet loss rates below 5% in short-range environments [187], while telemedicine-oriented IoT infrastructures enable remote supervision and emergency intervention pathways [193]. Nevertheless, these capabilities introduce systemic trade-offs. Embedded AI execution increases computational demand and energy consumption [45], while wireless communication infrastructures introduce vulnerabilities related to data interception and protocol-level security weaknesses [194]. Thus, scalable IoT-WCE integration requires not only communication reliability but also co-optimization of power budgeting, cybersecurity, and embedded hardware design. Compressed sensing (CS) methodologies provide a complementary pathway by explicitly addressing acquisition and transmission constraints. By modelling signal recovery as a sparsity-constrained inverse problem [39–41], CS-based frameworks have achieved compression ratios approaching 80% while preserving diagnostic utility [188]. This capacity directly mitigates bandwidth and energy bottlenecks in resource-constrained ingestible platforms. However, reconstruction fidelity under biologically induced noise, motion artifacts, and non-ideal sampling remains a critical limitation, particularly under real-time processing constraints [46]. Importantly, the literature reveals limited exploration of tightly coupled CS–AI pipelines in which sparse acquisition, adaptive reconstruction, and diagnostic inference operate as a unified, latency-aware system. This represents a substantial opportunity for architectural innovation. A broader cross-domain insight emerging from this review is the fragmentation of technological maturity. While AI-based diagnostic modules and IoT communication layers have reached moderate-to-high Technology Readiness Levels (TRL 6–8) [186], [193], CS-based acquisition frameworks remain at comparatively earlier stages of maturity (TRL 4–5) [188]. More critically, fully integrated end-to-end architectures—where AI inference, IoT telemetry, and CS-driven data optimization are co-designed under shared energy and regulatory constraints—are rarely demonstrated in clinically validated environments. This translational gap underscores a systemic disconnect between algorithmic innovation and deployable medical-grade platforms. From a systems-engineering perspective, future progress in smart WCE will depend less on incremental performance gains within individual components and more on holistic architectural convergence. Energy-aware co-design, explainable AI, secure IoT communication layers, and mathematically grounded sparse acquisition must be jointly optimized rather than independently advanced. Only through such integrative frameworks can next-generation smart capsules transition from promising prototypes to clinically scalable, regulatory-compliant, and interoperable healthcare devices. 10.1 Limitations of This Review This review adopts a structured narrative synthesis approach based on publicly available peer-reviewed studies. Although more than 120 publications were systematically analyzed within a standardized comparative framework [51], [190], the heterogeneity of reported performance metrics, datasets, and evaluation methodologies precluded quantitative meta-analysis. Consequently, conclusions are interpretative and trend-oriented rather than statistically aggregated. Nevertheless, the comparative synthesis provides a coherent systems-level perspective that clarifies technological maturity, integration feasibility, and translational readiness across AI, IoT, and CS domains. 11. Future Research Opportunities and Open Challenges 11.1 Summary of Unresolved Issues Despite significant progress in WCE enabled by AI, the IoT, and CS, several barriers still limit full-scale clinical adoption. For AI, the most pressing concern is limited generalizability. Current models are typically trained on small, curated datasets that do not capture the full diversity of gastrointestinal conditions, which increases the risk of misclassification in rare or atypical cases [35]. Another critical challenge is interpretability: deep learning models often function as black boxes, reducing physician confidence and delaying regulatory approval [144]. IoT integration continues to face three persistent issues: power consumption, latency, and data security. Continuous real-time transmission via BLE or Wi-Fi can rapidly deplete capsule batteries, sometimes within only a few hours [167]. Latency also fluctuates due to patient movement and tissue interference, undermining reliability in time-sensitive scenarios such as bleeding detection [143]. Furthermore, cloud-based data sharing exposes patient information to cybersecurity threats if strong encryption and authentication measures are not implemented [194]. Although CS effectively reduces data acquisition and transmission loads, it also introduces drawbacks. High undersampling levels (e.g., 70% reduction) may distort image structures, limiting lesion detectability in edge cases [93]. At the hardware level, integrating CS remains immature: random-sampling circuits and miniaturized ADCs remain difficult to design within the strict spatial and thermal constraints of capsule devices [53]. Taken together, these unresolved issues emphasize the need for cross-domain strategies that combine algorithmic innovation, power optimization, enhanced security, and hardware–software co-design. Importantly, no single technology can address all these limitations; future progress will rely on the complementary strengths of AI for intelligent analysis, IoT for continuous connectivity, and CS for efficient data handling. 11.2 Open Technical Challenges As WCE systems advance toward greater autonomy and diagnostic precision, several unresolved technical barriers remain. One of the central challenges is the real-time execution of AI models on the capsule’s constrained hardware. Although CNNs deliver high diagnostic accuracy, their computational and memory requirements exceed the capabilities of capsule-grade microcontrollers [52]. Efforts to adopt lightweight variants, such as MobileNet or quantized CNNs, have shown promise; however, they still strain the limited power budget and may cause overheating during extended operation [186]. Ensuring safe, energy-efficient real-time inference, therefore, remains a major obstacle [17]. Wireless communication is another critical bottleneck, particularly for latency-sensitive tasks like gastrointestinal bleeding detection. BLE and Wi-Fi modules often experience significant degradation due to tissue attenuation and patient movement, leading to transmission interruptions [169]. Reliable low-latency connectivity without frequent retransmissions or buffering is essential [187]. Moreover, AI-triggered anomaly alerts can introduce sudden power spikes, underscoring the need for intelligent power management [123]. The implementation of CS at the hardware level also remains underdeveloped. While simulations demonstrate strong performance, analog front-end components—such as pseudo-random samplers, low-noise amplifiers, and CS-aware ADCs—must be miniaturized to operate within < 1 cm³ while remaining reliable in varying biological and electromagnetic environments [53], [23]. Thermal regulation adds another layer of complexity. High-frequency sensing, wireless bursts, and on-chip AI computations generate localized heating that may irritate surrounding tissue or impair sensor performance [191]. Existing strategies, including duty cycling and passive dissipation, may not suffice for future hybrid AI–CS–IoT systems operating continuously in real time. In summary, the evolution of WCE requires multi-layered innovation in hardware–software co-design, adaptive communication protocols, energy-aware AI algorithms, and robust thermal management. Addressing these technical bottlenecks is essential to enabling safe, reliable, and intelligent capsule diagnostics. 11.3 Research Opportunities The convergence of AI, IoT, and CS in WCE creates extensive research opportunities to overcome current limitations and move toward autonomous, personalized diagnostics. A major direction is the development of lightweight, energy-efficient AI architectures tailored to capsule hardware. Techniques such as TinyML, network pruning, weight quantization, and knowledge distillation have been shown to reduce memory and power consumption while maintaining diagnostic accuracy significantly [33]. Emerging neuromorphic accelerators and spiking neural networks could further improve performance, enabling real-time inference in the micro-Watt regime [45]. Such innovations would enhance onboard lesion detection while reducing dependence on external servers [109]. Hybrid pipelines that combine AI, CS, and IoT also represent a promising avenue. CS can reduce the data volume during acquisition, after which compact AI classifiers running on-chip determine whether full transmission is necessary [188]. This adaptive approach ensures that only diagnostically relevant information is transmitted, conserving energy and minimizing network congestion [171]. These cascading frameworks underpin closed-loop WCE systems, in which sensing, inference, and communication adapt dynamically to evolving diagnostic insights [194]. Privacy and data security remain critical areas for research. Federated learning enables AI models to be trained collaboratively without transferring raw patient data [194]. At the same time, blockchain mechanisms create immutable audit trails that support regulatory compliance, such as HIPAA and GDPR [143]. Another emerging frontier is patient-specific personalization. Capsules may eventually adjust imaging frequency, AI sensitivity thresholds, or transmission modes based on individual patient histories and clinical objectives [32]. For this vision to materialize, AI systems must not only be accurate but also interpretable, adaptable, and ethically aligned with medical practice. Taken together, these research opportunities highlight the potential for next-generation capsule platforms that are intelligent, energy-efficient, context-aware, and secure. However, realizing these advancements will ultimately require rigorous clinical validation and regulatory approval to ensure safety and trustworthiness in real-world deployments. 11.4 Clinical and Regulatory Considerations The translation of WCE systems from prototypes to routine clinical use depends not only on technological readiness but also on rigorous validation and regulatory compliance. For AI-powered capsules, agencies such as the FDA and CE require multi-stage clinical trials that demonstrate safety, efficacy, reproducibility, and superiority to existing diagnostic standards [32]. Adaptive learning models further complicate this process, as their evolution after deployment may necessitate continuous monitoring and periodic re-certification. IoT-enabled WCE platforms face additional requirements arising from privacy and data protection laws, such as HIPAA in North America and the GDPR in Europe. Real-time telemetry, remote alerts, and cloud synchronization must operate within secure communication frameworks that incorporate encryption and authentication to prevent unauthorized access [143]. Particularly in remote care applications, where data is transmitted via mobile or cloud networks, interoperability with hospital systems and cybersecurity auditing remain critical deployment concerns [194]. Compressed sensing, while less directly regulated, still demands validation to ensure diagnostic reliability. Even minor reconstruction artifacts can influence physician decision-making, underscoring the need for standardized image-quality benchmarks, such as PSNR and SSIM, combined with clinical scoring methods [167]. Furthermore, transparent documentation of reconstruction workflows and their limitations is vital to support informed consent practices [188]. Another pressing issue is the generalizability of performance across diverse populations and disease types. Many AI models are trained on region-specific datasets, which limits diversity and introduces risks of bias [57]. To mitigate this, large-scale multicenter trials across diverse demographics and pathologies are necessary to build trust in AI-assisted diagnostics [35]. Finally, explainability is increasingly central to both ethical and legal acceptance. Clinicians and regulators require transparency regarding how AI systems reach diagnostic decisions. Tools such as saliency maps, confidence scores, and traceable decision logs are gaining traction, aligning with explainable AI (XAI) initiatives in medicine [45]. In summary, the clinical deployment of smart WCE systems is not solely a technical achievement but also a regulatory and ethical undertaking. Closing validation gaps, ensuring data security, and embedding explainability into diagnostic pipelines will be essential for safe, scalable, and trustworthy adoption in healthcare practice. 11.5 Future Vision The convergence of AI, IoT, and CS in WCE signals a paradigm shift toward personalized, autonomous gastrointestinal diagnostics. The vision extends beyond incremental improvements, aiming for self-directed, context-aware, and therapeutically capable capsules. A central trajectory is the development of autonomous diagnostic capsules that adapt dynamically in real time—adjusting frame rates, lighting, or transmission protocols according to lesion detection or gastrointestinal motility [67]. Such systems enable closed-loop control, where AI actively governs data acquisition and telemetry strategies based on pathological findings [148]. The synergy between CS and AI offers a promising path to achieving real-time intelligence under strict hardware and energy constraints. By performing sparse sampling at the sensor level and reconstructing only diagnostically relevant features, computational demands are reduced while maintaining high-fidelity classification [167], [188]. This approach could extend battery life beyond 10 hours, enabling coverage of the entire gastrointestinal tract without compromising accuracy [30]. Advances in neuromorphic processors, which mimic brain-inspired, event-driven computation, may enable CNNs and RNNs to operate at micro-Watt energy profiles. Embedding such processors into WCE systems could deliver true edge-level intelligence while meeting thermal and spatial limitations [8]. From the IoT perspective, future capsules are expected to support bidirectional communication and adaptive alerts. This would enable clinicians to query or reprioritize data streams in real time [186]. In resource-limited regions, Low-Power Wide-Area Networks (LPWAN) could facilitate remote diagnostics, expanding access to care [147]. Personalized diagnostic models also represent a major frontier. By incorporating patient-specific factors such as prior medical history, imaging patterns, or microbiome data, WCE platforms could tailor their acquisition and diagnostic routines to individual risk profiles [43]. Looking ahead, capsules may evolve into theranostic systems —capable of localized drug delivery, thermal ablation, or biosample collection—supported by breakthroughs in power harvesting, actuation, and miniaturization [195],[51],[177]. In summary, the future of WCE lies in intelligent autonomy, energy-aware operation, and patient-centred personalization. Achieving this vision requires addressing unresolved issues, technical barriers, and regulatory demands while leveraging AI for accuracy, IoT for connectivity, and CS for efficiency. No single approach is sufficient; only through their integration and sustained interdisciplinary collaboration across engineering, clinical medicine, regulatory science, and ethics can smart capsules evolve into reliable, autonomous, and intelligent diagnostic platforms. The convergence of AI, IoT, and CS is not merely additive but catalytic, accelerating the transition of capsule endoscopy from an experimental technology into routine clinical practice. 12. Conclusion and Future Outlook This review provides a comprehensive, structured synthesis of the transformative roles that AI, IoT, and CS play in enabling the next generation of Wireless Capsule Endoscopy. By systematically examining their respective and combined contributions to lesion detection, telemetry, localization, and intelligent control, we have shown that these technologies offer not only incremental improvements but also the potential for a fundamental shift in how gastrointestinal diseases are diagnosed and managed. AI offers powerful tools for automated image interpretation, and IoT ensures real-time connectivity and remote supervision. At the same time, CS addresses the longstanding issue of data overload under strict bandwidth and energy limitations. Despite significant progress, several core limitations remain. As discussed in the AI section, AI systems continue to struggle with data scarcity and limited generalization across diverse patient populations and imaging conditions. Moreover, due to the capsule's limited space and energy budget, high-performance AI models must often be executed offline, thereby introducing diagnostic latency. Current power sources cannot sustain continuous operation, and wireless power transfer methods are still in early experimental stages. In addition, while compressed sensing offers promising data reduction, it must strike a delicate balance between compression efficiency and diagnostic interpretability. Equally concerning is the absence of active navigation and therapeutic functionalities, which restrict the capsule’s use to passive observation rather than actionable clinical intervention. These unresolved challenges underscore the urgent need for innovative architectures that integrate intelligence, connectivity, and efficiency in a balanced manner. By organizing the literature around unified methodological, mathematical, and computational frameworks, this review provides a coherent foundation for interpreting recent advances in smart wireless capsule endoscopy. This structured perspective clarifies how diverse AI-, IoT-, and compressive sensing–based approaches relate to one another and highlights their respective assumptions, constraints, and trade-offs. Looking ahead, the integration of ultra-low-power AI models, custom-designed embedded processors (e.g., ASICs, FPGAs), and hybrid AI–CS pipelines may enable on-board real-time intelligence without compromising energy efficiency. New paradigms like neuromorphic computing and bio-inspired sensors may further enhance embedded processing in next-generation capsules. Clinical translation will also require extensive human-centred design practices, multi-site trials, and adherence to emerging medical AI standards. Ethical considerations—such as patient data privacy, algorithm transparency, and risk mitigation in autonomous decision-making—must be built into the design pipeline from the outset. The advancement of smart capsule endoscopy is inherently interdisciplinary. Effective progress demands tight collaboration among biomedical engineers, gastroenterologists, AI researchers, hardware designers, clinical ethicists, and regulatory authorities. Initiatives to establish interoperable platforms, common benchmarking datasets, and shared validation protocols will play a key role in accelerating innovation while maintaining clinical rigour and safety. With persistent research, regulatory foresight, and collaborative innovation, smart capsule endoscopy has the potential to evolve from a passive imaging modality into a fully integrated diagnostic and therapeutic platform. This evolution promises not just a refinement of existing tools but a profound rethinking of minimally invasive medicine—where real-time sensing, targeted intervention, and patient-centred design converge in a single swallowable device. Ultimately, the convergence of AI, IoT, and CS provides not only incremental advances but also the foundation for a paradigm shift toward intelligent, connected, and energy-efficient gastrointestinal diagnostics. Abbreviations Abbreviation Full Term AI Artificial Intelligence ASIC Application-Specific Integrated Circuit BLE Bluetooth Low Energy CAM Class Activation Map CNN Convolutional Neural Network CS Compressed Sensing EMR Endoscopic Mucosal Resection ESD Endoscopic Submucosal Dissection FL Federated Learning GAN Generative Adversarial Network GI Gastrointestinal IoT Internet of Things LSTM Long Short-Term Memory MEMS Microelectromechanical Systems NB-IoT Narrowband Internet of Things RF Radio Frequency SAR Specific Absorption Rate SoC System-on-Chip UWB Ultra-Wideband WCE Wireless Capsule Endoscopy WPT Wireless Power Transfer Declarations Availability of data and materials. Data sharing not applicable to this article as no datasets were generated or analysed during the current study. Competing interests. The authors declare that they have no competing interests. Funding. No funding was received for this study . Authors’contributions. Zeinab Javid conceptualized the study, conducted the literature review, structured the analysis, and drafted the manuscript. Michel Kadoch supervised the research, contributed to critical revision of the manuscript, and approved the final version. All authors read and approved the final manuscript . References Wang Y, Huang Y, Chase RC, et al (2023) Global Burden of Digestive Diseases: A Systematic Analysis of the Global Burden of Diseases Study, 1990 to 2019. Gastroenterology 165:773–783.e15. https://doi.org/10.1053/j.gastro.2023.05.050 Endoscopy C, Size M, Region B, et al (2025) Capsule Endoscopy Market Summary Market Concentration & Characteristics. 2–11 Enns RA, Hookey L, Armstrong D, et al (2017) Clinical Practice Guidelines for the Use of Video Capsule Endoscopy. Gastroenterology 152:497–514. https://doi.org/10.1053/j.gastro.2016.12.032 Enns C, Galorport C, Ou G, Enns R (2021) Assessment of Capsule Endoscopy Utilizing Capsocam Plus in Patients with Suspected Small Bowel Disease, Including Pilot Study with Remote Access Patients during Pandemic. J Can Assoc Gastroenterol 4:269–273. https://doi.org/10.1093/jcag/gwaa042 Khattab Y, Pott PP (2025) Active/robotic capsule endoscopy - A review. Alexandria Eng J 127:431–451. https://doi.org/10.1016/j.aej.2025.05.032 Nomura S, Shimura T, Katano T, et al (2020) A multicenter, single-blind randomized controlled trial of endoscopic clipping closure for preventing coagulation syndrome after colorectal endoscopic submucosal dissection. Gastrointest Endosc 91:859–867.e1. https://doi.org/10.1016/j.gie.2019.11.030 Ali MA, Tom N, Alsunaydih FN, et al (2024). Recent Advancements in Localization Technologies for Wireless Capsule Endoscopy: A Technical Review. Sensors 15:1–21. https://doi.org/10.3390/bioengineering12060613 Dhali A, Kipkorir V, Maity R, et al (2025) Artificial Intelligence–Assisted Capsule Endoscopy Versus Conventional Capsule Endoscopy for Detection of Small Bowel Lesions: A Systematic Review and Meta-Analysis. J Gastroenterol Hepatol 40:1105–1118. https://doi.org/10.1111/jgh.16931 Wang F, Hu D, Sun H, et al (2023) Global, regional, and national burden of digestive diseases: findings from the global burden of disease study 2019. Front Public Heal 11:. https://doi.org/10.3389/fpubh.2023.1202980 Chen Q, Wang X, Peng J, et al (2025) The burden of digestive diseases in Asian countries and territories from 1990 to 2019: an analysis for the Global Burden of Disease 2019 study. npj Gut Liver 2:1–12. https://doi.org/10.1038/s44355-025-00019-x Chen W, Neely MJ, Mitra U (2008) Energy-Efficient Transmissions With Individual Packet Delay Constraints. IEEE Trans Inf Theory 54:2090–2109. https://doi.org/10.1109/TIT.2008.920344 Yung DE, Plevris JN, Koulaouzidis A (2017) Short article: Aspiration of capsule endoscopes: a comprehensive review of the existing literature. Eur J Gastroenterol Hepatol 29:428–434. https://doi.org/10.1097/MEG.0000000000000821 Mehmood T, Naeem N, Parveen S (2025) Survey on Integrated Power Optimization with Battery Friendly Algorithms Survey in Wireless Capsule Endoscopy. IJCSNS Int J Comput Sci Netw Secur 25: Chen X, Zhang X, Zhang L, et al (2009) A Wireless Capsule Endoscope System With Low-Power Controlling and Processing ASIC. IEEE Trans Biomed Circuits Syst 3:11–22. https://doi.org/10.1109/TBCAS.2008.2006493 Bouyaya D, Benierbah S, Khamadja M (2021) An intelligent compression system for wireless capsule endoscopy images. Biomed Signal Process Control 70:102929. https://doi.org/10.1016/j.bspc.2021.102929 Manolescu VD, AlZu’bi H, Secco EL (2024) Review on endoscopic capsules. Explor Digit Heal Technol 346–359. https://doi.org/10.37349/edht.2024.00033 Liu Y, Wang B (2025) Advanced applications in chronic disease monitoring using IoT mobile sensing device data, machine learning algorithms and frame theory: a systematic review. Front Public Heal 13:. https://doi.org/10.3389/fpubh.2025.1510456 Huang K, Qiu H, Deng Y, et al (2024) Capsule Endoscopy Technology: A New Era in Digestive Tract Examination. J Dig Endosc 15:243–249. https://doi.org/10.1055/s-0044-1800916 Chen Q, Wang XX, Peng J, et al (2025) Performance evaluation and future prospects of capsule robot localization technology. Front Public Heal 15:1–31. https://doi.org/10.1080/10095020.2024.2354239 Piccirelli S, Salvi D, Pugliano CL, et al (2025) Unmet Needs of Artificial Intelligence in Small Bowel Capsule Endoscopy. Diagnostics 15:1–11. https://doi.org/10.3390/diagnostics15091092 Wang X, Xu H, Ren Y, et al (2025) Advances in implantable capsule robots for monitoring and treatment of gastrointestinal diseases. Mater Sci Eng R Reports 163:. https://doi.org/10.1016/j.mser.2025.100943 Rehan M, Al-Bahadly I, Thomas DG, et al (2023) Smart capsules for sensing and sampling the gut: status, challenges and prospects. Gut 73:186–202. https://doi.org/10.1136/gutjnl-2023-329614 Wang Y, Chen Y, Zhao Y, Liu S (2024) Compressed Sensing for Biomedical Photoacoustic Imaging: A Review Krumb HJ, Mukhopadhyay A (2025) eNCApsulate: NCA for Precision Diagnosis on Capsule Endoscopes. Int J Comput Assist Radiol Surg. https://doi.org/10.1007/s11548-025-03425-x Tsuboi A, Oka S, Tanaka S (2024) Capsule endoscopy: clinical insights, challenges, and evolving perspectives in the 21st century. Mini-invasive Surg 8:. https://doi.org/10.20517/2574-1225.2023.94 Goel S, Guleria K, Panda SN (2025). Comprehensive Review on Routing Protocols for Wireless Body Area Networks. Int J Math Eng Manag Sci 10:797–855. https://doi.org/10.33889/IJMEMS.2025.10.4.040 Wei X, Xi P, Chen M, et al (2024). Capsule robots for the monitoring, diagnosis, and treatment of intestinal diseases. Mater Today Bio 29:101294. https://doi.org/10.1016/j.mtbio.2024.101294 Drzisga D, Köppl T, Pohl U, et al (2016) Numerical modelling of compensation mechanisms for peripheral arterial stenoses. Comput Biol Med 70:190–201. https://doi.org/10.1016/j.compbiomed.2016.01.015 Kim HE (2023) Enteroscopy versus Video Capsule Endoscopy for Automatic Diagnosis of Small Bowel Disorders—A Comparative Analysis of Artificial Intelligence Applications. Biomedicines 11:. https://doi.org/10.3390/biomedicines11112991 Kim HE, Cosa-Linan A, Santhanam N, et al (2022) Transfer learning for medical image classification: a literature review. BMC Med Imaging 22:69. https://doi.org/10.1186/s12880-022-00793-7 Gupta M, Mishra A (2024) A systematic review of deep learning based image segmentation to detect polyps. Artif Intell Rev 57:7. https://doi.org/10.1007/s10462-023-10621-1 Sahafi A, Wang Y, Rasmussen CLM, et al (2022) Edge artificial intelligence wireless video capsule endoscopy. Sci Rep 12:13723. https://doi.org/10.1038/s41598-022-17502-7 Habe TT, Haataja K, Toivanen P (2024) Review of Deep Learning Performance in Wireless Capsule Endoscopy Images for GI Disease Classification. F1000Research 13:201. https://doi.org/10.12688/f1000research.145950.2 Naskar S, Sharma S, Kuotsu K, et al (2025). The biomedical applications of artificial intelligence: an overview of decades of research. J Drug Target 33:717–748. https://doi.org/10.1080/1061186X.2024.2448711 Kim HE, Cosa-Linan A, Maros ME, et al (2021) A review of transfer learning for medical image classification. Int. J. Wirel. Inf. Networks 27:30–44 Munir A, Noor K, Shams MU, et al (2025). The Research of Medical Science Review. The Impact of AI TECHNOLOGIES IN MODERN HEALTHCARE : A Critical Analysis of Challenges, OPPORTUNITIES OF FUTURE PROSPECTS. The Research of Medical Science Review. 3:352–366 Miao L, Zhang T, Zuo C, et al (2024) A Rapid Localization Method Based on Super Resolution Magnetic Array Information for an Unknown Number of Magnetic Sources. Sensors 24:. https://doi.org/10.3390/s24103226 Vedaei SS, Wahid KA (2021) A localization method for wireless capsule endoscopy using side wall cameras and IMU sensor. Sci Rep 11:11204. https://doi.org/10.1038/s41598-021-90523-w Schulz A, Silva EAB (2009) Compressive Sensing. 1–41 Zhu J, Ma J, Liu Z, et al (2025) A Modulo Sampling Hardware Prototype and Reconstruction Algorithms Evaluation. IEEE Trans Instrum Meas 74:. https://doi.org/10.1109/TIM.2025.3551491 Li B, Wang Y, Zhao J, Shi J (2024) Ultra-Wideband Antennas for Wireless Capsule Endoscope System: A Review. IEEE Open J Antennas Propag 5:241–255. https://doi.org/10.1109/OJAP.2024.3355217 Jiang D, Zhu L, Tong S, et al (2023) Photoacoustic imaging plus X: a review. J Biomed Opt 29:1–27. https://doi.org/10.1117/1.jbo.29.s1.s11513 Eregowda N, Pruthviraja D (2023) Feature Extraction and Classification Techniques for Wireless Endoscopy Images: A Review. Rev d’Intelligence Artif 37:171–178. https://doi.org/10.18280/ria.370121 Kim HE (2022) Convolution neural network for the diagnosis of wireless capsule endoscopy: a systematic review and meta-analysis. Surg Endosc 36:16–31. https://doi.org/10.1007/s00464-021-08689-3 Yu G, Zhao S (2020) A new feature descriptor for multimodal image registration using phase congruency. Sensors (Switzerland) 20:1–21. https://doi.org/10.3390/s20185105 Xu B, Yu C (2025) Wireless, Battery-free, Implantable Inductor-Capacitor Based Sensors. Adv Electron Mater 2500184:1–14. https://doi.org/10.1002/aelm.202500184 Han S, Wang N, Guo Y, et al (2021) Application of Sparse Representation in Bioinformatics. Front Genet 12:1–12. https://doi.org/10.3389/fgene.2021.810875 Munagandla VB, Pochu S, Nersu SRK, Kathram SR (2024) Real-Time Data Integration for Emergency Response in Healthcare Systems. J AI-Powered Med Innov (International online ISSN 3078 − 1930) 3:25–38. https://doi.org/10.60087/japmi.vol.03.issue.01.id.002 Mart C, Weinreich W, Czernohorsky M, et al (2018) CMOS-compatible pyroelectric applications enabled by doped HfO2 films on deep-trench structures. Eur Solid-State Device Res Conf 2018-Septe:130–133. https://doi.org/10.1109/ESSDERC.2018.8486864 Cao Q, Deng R, Pan Y, et al (2024) Robotic wireless capsule endoscopy: recent advances and upcoming technologies. Nat Commun 15:4597. https://doi.org/10.1038/s41467-024-49019-0 Su C-C, Chou C-K, Mukundan A, et al (2025) Capsule Endoscopy: Current Trends, Technological Advancements, and Future Perspectives in Gastrointestinal Diagnostics. Bioengineering 12:613. https://doi.org/10.3390/bioengineering12060613 Kiran B, Thomas D, Parakkal R (2018) An Overview of Deep Learning Based Methods for Unsupervised and Semi-Supervised Anomaly Detection in Videos. J Imaging 4:36. https://doi.org/10.3390/jimaging4020036 Wang X, Hu X, Xu Y, et al (2023) A systematic review on diagnosis and treatment of gastrointestinal diseases by magnetically controlled capsule endoscopy and artificial intelligence. Therap Adv Gastroenterol 16:1–13. https://doi.org/10.1177/17562848231206991 Liu L, Towfighian S, Hila A (2015) A Review of Locomotion Systems for Capsule Endoscopy. IEEE Rev Biomed Eng 8:138–151. https://doi.org/10.1109/RBME.2015.2451031 Alam MW, Hasan MM, Mohammed SK, et al (2017) Are Current Advances of Compression Algorithms for Capsule Endoscopy Enough? A Technical Review. IEEE Rev Biomed Eng 10:26–43. https://doi.org/10.1109/RBME.2017.2757013 SWAIN P (2010) At a watershed? Technical developments in wireless capsule endoscopy. J Dig Dis 11:259–265. https://doi.org/10.1111/j.1751-2980.2010.00448.x Edet AN, Ben IAO, Imoh U (2020). Past, Present and Future of Robotic Capsule Endoscopy. In: International Journal of Scientific Research and Engineering Development. pp 935–941 Carpi F, Kastelein N, Talcott M, Pappone C (2011) Magnetically Controllable Gastrointestinal Steering of Video Capsules. IEEE Trans Biomed Eng 58:231–234. https://doi.org/10.1109/TBME.2010.2087332 Pan G, Wang L (2012) Swallowable Wireless Capsule Endoscopy: Progress and Technical Challenges. Gastroenterol Res Pract 2012:1–9. https://doi.org/10.1155/2012/841691 Nakamura T, Terano A (2008) Capsule endoscopy: past, present, and future. J Gastroenterol 43:93–99. https://doi.org/10.1007/s00535-007-2153-6 Adler SN, Metzger YC (2011) PillCam COLON capsule endoscopy: recent advances and new insights. Therap Adv Gastroenterol 4:265–268. https://doi.org/10.1177/1756283X11401645 Dupont PE, Nelson BJ, Goldfarb M, et al (2021) A decade retrospective of medical robotics research from 2010 to 2020. Sci Robot 6:. https://doi.org/10.1126/scirobotics.abi8017 Hakimian S, Raines D, Reed G, et al (2021) Assessment of Video Capsule Endoscopy in the Management of Acute Gastrointestinal Bleeding During the COVID-19 Pandemic. JAMA Netw Open 4:e2118796. https://doi.org/10.1001/jamanetworkopen.2021.18796 Iddan G, Meron G, Glukhovsky A, Swain P (2000) Wireless capsule endoscopy. Nature 405:417–417. https://doi.org/10.1038/35013140 Ding Z, Shi H, Zhang H, et al (2019) Gastroenterologist-Level Identification of Small-Bowel Diseases and Normal Variants by Capsule Endoscopy Using a Deep-Learning Model. Gastroenterology 157:1044–1054.e5. https://doi.org/10.1053/j.gastro.2019.06.025 Luo Y-Y, Pan J, Chen Y-Z, et al (2019) Magnetic Steering of Capsule Endoscopy Improves Small Bowel Capsule Endoscopy Completion Rate. Dig Dis Sci 64:1908–1915. https://doi.org/10.1007/s10620-019-5479-z Cao Q, Deng R, Pan Y, et al (2024) Robotic wireless capsule endoscopy: recent advances and upcoming technologies. Nat Commun 15:1–21. https://doi.org/10.1038/s41467-024-49019-0 Swain P (2008) The future of wireless capsule endoscopy. World J Gastroenterol 14:4142. https://doi.org/10.3748/wjg.14.4142 Ciuti G, Menciassi A, Dario P (2011) Capsule Endoscopy: From Current Achievements to Open Challenges. IEEE Rev Biomed Eng 4:59–72. https://doi.org/10.1109/RBME.2011.2171182 Pikul JH, Gang Zhang H, Cho J, et al (2013) High-power lithium ion microbatteries from interdigitated three-dimensional bicontinuous nanoporous electrodes. Nat Commun 4:1732. https://doi.org/10.1038/ncomms2747 Ma S, Jiang M, Tao P, et al (2018) Temperature effect and thermal impact in lithium-ion batteries: A review. Prog Nat Sci Mater Int 28:653–666. https://doi.org/10.1016/j.pnsc.2018.11.002 Mostafalu P, Sonkusale S (2014). Flexible and transparent gastric battery: Energy harvesting from gastric acid for endoscopy application. Biosens Bioelectron 54:292–296. https://doi.org/10.1016/j.bios.2013.10.040 Nadeau P, El-Damak D, Glettig D, et al (2017) Prolonged energy harvesting for ingestible devices. Nat Biomed Eng 1:0022. https://doi.org/10.1038/s41551-016-0022 Sharova AS, Melloni F, Lanzani G, et al (2021) Edible Electronics: The Vision and the Challenge. Adv Mater Technol 6:. https://doi.org/10.1002/admt.202000757 Ciuti G, Caliò R, Camboni D, et al (2016) Frontiers of robotic endoscopic capsules: a review. J Micro-Bio Robot 11:1–18. https://doi.org/10.1007/s12213-016-0087-x Ilic IK, Galli V, Lamanna L, et al (2023) An Edible Rechargeable Battery. Adv Mater 35:. https://doi.org/10.1002/adma.202211400 Wang F, Wu X, Yuan X, et al (2017). Latest advances in supercapacitors: from new electrode materials to novel device designs. Chem Soc Rev 46:6816–6854. https://doi.org/10.1039/C7CS00205J Chen K, Yan L, Sheng Y, et al (2022) An Edible and Nutritive Zinc-Ion Micro-supercapacitor in the Stomach with Ultrahigh Energy Density. ACS Nano 16:15261–15272. https://doi.org/10.1021/acsnano.2c06656 Zhang M, Du H, Liu K, et al (2021) Fabrication and applications of cellulose-based nanogenerators. Adv Compos Hybrid Mater 4:865–884. https://doi.org/10.1007/s42114-021-00312-2 Labib M, Grabowski L, Brüsseler C, et al (2022) Toward the Sustainable Production of the Active Pharmaceutical Ingredient Metaraminol. ACS Sustain Chem Eng 10:5117–5128. https://doi.org/10.1021/acssuschemeng.1c08275 Yang J, Zhang Y, Cao W, et al (2022). Analysis of voltage vectors and the realization method of torque ripple reduction for BLDCM. IET Power Electron 15:865–876. https://doi.org/10.1049/pel2.12274 Jia Q, Song T, Li Y, et al (2019) OAR Dose Distribution Prediction and gEUD-Based Automatic Treatment Planning Optimization for Intensity Modulated Radiotherapy. IEEE Access 7:141426–141437. https://doi.org/10.1109/ACCESS.2019.2942393 Gao J, Zhang Z, Yan G (2019). Development of a Capsule Robot for Exploring the Colon. Micromachines 10:456. https://doi.org/10.3390/mi10070456 Egger J, Tokuda J, Chauvin L, et al (2012) Integration of the OpenIGTLink Network Protocol for image-guided therapy with the medical platform MeVisLab. Int J Med Robot Comput Assist Surg 8:282–290. https://doi.org/10.1002/rcs.1415 Höög CM, Bark L-Å, Arkani J, et al (2012) Capsule Retentions and Incomplete Capsule Endoscopy Examinations: An Analysis of 2300 Examinations. Gastroenterol Res Pract 2012:1–7. https://doi.org/10.1155/2012/518718 Lo W, Liu Y, Elhajj IH, et al (2004) Cooperative Teleoperation of a Multirobot System With Force Reflection via Internet. IEEE/ASME Trans Mechatronics 9:661–670. https://doi.org/10.1109/TMECH.2004.839040 Chao-Chieh Lan, Kok-Meng Lee. Dynamic model of a compliant link with large deflection and shear deformation. In: Proceedings, 2005 IEEE/ASME International Conference on Advanced Intelligent Mechatronics. IEEE, pp 729–734 Johnson MJ, Xin Feng, Johnson LM, et al Robotic Systems that Rehabilitate as well as Motivate: Three Strategies for Motivating Impaired Arm Use. In: The First IEEE/RAS-EMBS International Conference on Biomedical Robotics and Biomechatronics, 2006. BioRob 2006. IEEE, pp 254–259 Gosselin B, Faniel L, Sawan M (2006) A high data rate telemetry system for multi-channel biosignal recording. In: 2006 IEEE Biomedical Circuits and Systems Conference. IEEE, pp 170–173 (2012) IEEE Standard for Local and metropolitan area networks - Part 15.6: Wireless Body Area Networks Won K, Choi H-J (2011) A novel diversity transmission technique using cooperative relay. In: The 17th Asia Pacific Conference on Communications. IEEE, pp 649–653 Sashiyama H, Hamahata Y, Matsuo K, et al (2008) Rectal burn caused by hot-water coffee enema. Gastrointest Endosc 68:1008–1009. https://doi.org/10.1016/j.gie.2008.04.017 Ali MA, Tom N, Alsunaydih FN, Yuce MR (2025) Recent Advancements in Localization Technologies for Wireless Capsule Endoscopy: A Technical Review. Sensors 25:1–31. https://doi.org/10.3390/s25010253 Taniuchi K, Ohba Y, Fajardo V, et al (2009) IEEE 802.21: Media independent handover: Features, applicability, and realization. IEEE Commun Mag 47:112–120. https://doi.org/10.1109/MCOM.2009.4752687 Slawinski PR (2015) Capsule endoscopy of the future: What’s on the horizon? World J Gastroenterol 21:10528. https://doi.org/10.3748/wjg.v21.i37.10528 Campisano F, Gramuglia F, Dawson IR, et al (2017) Gastric Cancer Screening in Low-Income Countries: System Design, Fabrication, and Analysis for an Ultralow-Cost Endoscopy Procedure. IEEE Robot Autom Mag 24:73–81. https://doi.org/10.1109/MRA.2017.2673852 Koulaouzidis A (2015) Wireless endoscopy in 2020: Will it still be a capsule? World J Gastroenterol 21:5119. https://doi.org/10.3748/wjg.v21.i17.5119 Borsdorf A, Raupach R, Flohr T, Hornegger J (2008) Wavelet-Based Noise Reduction in CT-Images Using Correlation Analysis. IEEE Trans Med Imaging 27:1685–1703. https://doi.org/10.1109/TMI.2008.923983 Wang C, Luo Z, Liu X, et al (2018) Organic Boundary Location Based on Colour-Texture of Visual Perception in Wireless Capsule Endoscopy Video. J Healthc Eng 2018:1–11. https://doi.org/10.1155/2018/3090341 Nister D, Naroditsky O, Bergen J Visual odometry. In: Proceedings of the 2004 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2004. CVPR 2004. IEEE, pp 652–659 Maggi N, Arrigo P, Ruggiero C (2013). Toll-like receptor structural determinants: Variability analysis by digital signal processing methods. In: 13th IEEE International Conference on BioInformatics and BioEngineering. IEEE, pp 1–4 Wahab H, Mehmood I, Ugail H, et al (2023). Machine learning based small bowel video capsule endoscopy analysis: Challenges and opportunities. Futur Gener Comput Syst 143:191–214. https://doi.org/10.1016/j.future.2023.01.011 Trasolini R, Byrne MF (2021) Artificial intelligence and deep learning for small bowel capsule endoscopy. Dig Endosc 33:290–297. https://doi.org/10.1111/den.13896 Muruganantham P, Balakrishnan SM (2022). Attention Aware Deep Learning Model for Wireless Capsule Endoscopy Lesion Classification and Localization. J Med Biol Eng 42:157–168. https://doi.org/10.1007/s40846-022-00686-8 Ingle SB, Alexander JA (2007). Recurrent obscure gastrointestinal bleeding: time for provocative thinking? Gastroenterol Hepatol (N Y) 3:571–3 Li X, Ren J, Jiang H (2017) Experimental Investigation of Endwall Heat Transfer With Film and Impingement Cooling. J Eng Gas Turbines Power 139:. https://doi.org/10.1115/1.4036361 Cummins G (2021) Smart pills for gastrointestinal diagnostics and therapy. Adv Drug Deliv Rev 177:113931. https://doi.org/10.1016/j.addr.2021.113931 Kurbanov B (2012) Apoptose und Melanom: Neue therapeutische Zielstrukturen. Aktuelle Derm 38:91–94. https://doi.org/10.1055/s-0031-1291547 Choudhury D (2017) Challenges for 5G?The Future of Wireless Communications [From the Guest Editor’s Desk]. IEEE Microw Mag 18:16–16. https://doi.org/10.1109/MMM.2017.2696838 Koo TH, Tee V, Lee YY, et al (2024) Green Endoscopy and Sustainable Practices: A Scoping Review. J Dig Endosc 15:184–191. https://doi.org/10.1055/s-0044-1790203 Intzes I, Meng H, Cosmas J (2020) An Ingenious Design of a High Performance-Low Complexity Image Compressor for Wireless Capsule Endoscopy. Sensors 20:1617. https://doi.org/10.3390/s20061617 Mimee M, Nadeau P, Hayward A, et al (2018). An ingestible bacterial-electronic system to monitor gastrointestinal health. Science (80- ) 360:915–918. https://doi.org/10.1126/science.aas9315 Kalantar-Zadeh K, Berean KJ, Ha N, et al (2018) A human pilot trial of ingestible electronic capsules capable of sensing different gases in the gut. Nat Electron 1:79–87. https://doi.org/10.1038/s41928-017-0004-x Ke Q, Luo W, Yan G, Yang K (2016). Analytical Model and Optimized Design of Power Transmitting Coil for Inductively Coupled Endoscope Robot. IEEE Trans Biomed Eng 63:694–706. https://doi.org/10.1109/TBME.2015.2469137 Sekiya N, Oshimoto N, Ebihara K, et al (2023) Wireless Power Transfer System Using High-Quality Factor Superconducting Transmitting Coil for Biomedical Capsule Endoscopy. IEEE Trans Appl Supercond 33:1–5. https://doi.org/10.1109/TASC.2023.3256346 Miah MS, Jayathurathnage P, Icheln C, et al (2019) High-Efficiency Wireless Power Transfer System for Capsule Endoscope. In: 2019 13th International Symposium on Medical Information and Communication Technology (ISMICT). IEEE, pp 1–5 Sumiyama K, Futakuchi T, Kamba S, et al (2021) Artificial intelligence in endoscopy: Present and future perspectives. Dig Endosc 33:218–230. https://doi.org/10.1111/den.13837 Xu Y, Li K, Zhao Z, Meng MQ-H (2021) On Reciprocally Rotating Magnetic Actuation of a Robotic Capsule in Unknown Tubular Environments. IEEE Trans Med Robot Bionics 3:919–927. https://doi.org/10.1109/TMRB.2021.3123407 Sliker L, Ciuti G, Rentschler M, Menciassi A (2015) Magnetically driven medical devices: a review. Expert Rev Med Devices 12:737–752. https://doi.org/10.1586/17434440.2015.1080120 Wang Z, Guo S, Fu Q, Guo J (2019) Characteristic evaluation of a magnetic-actuated microrobot in a pipe with screw jet motion. Microsyst Technol 25:719–727. https://doi.org/10.1007/s00542-018-4000-5 Zhang Y, Fang M, Feng X, et al (2025). Forging trust in AI-assisted disease diagnosis. Life. https://doi.org/10.1016/j.hlife.2025.05.011 Smith H (2021) Clinical AI: opacity, accountability, responsibility and liability. AI Soc 36:535–545. https://doi.org/10.1007/s00146-020-01019-6 Abdigazy A, Arfan M, Lazzi G, et al (2024). End-to-end design of ingestible electronics. Nat Electron 7:102–118. https://doi.org/10.1038/s41928-024-01122-2 Xu Y, Li K, Zhao Z, Meng MQ-H (2021) A Novel System for Closed-Loop Simultaneous Magnetic Actuation and Localization of WCE Based on External Sensors and Rotating Actuation. IEEE Trans Autom Sci Eng 18:1640–1652. https://doi.org/10.1109/TASE.2020.3013954 Garbay T, Chuquimia O, Pinna A, et al (2019). Distilling the knowledge in CNN for the WCE screening tool. In: 2019 Conference on Design and Architectures for Signal and Image Processing (DASIP). IEEE, pp 19–22 Wang Y, Yoo S, Braun J-M, Nadimi ES (2021) A locally-processed light-weight deep neural network for detecting colorectal polyps in wireless capsule endoscopes. J Real-Time Image Process 18:1183–1194. https://doi.org/10.1007/s11554-021-01126-7 Oo WM, Linklater JM, Bennell KL, et al (2020) Superb Microvascular Imaging in Low-Grade Inflammation of Knee Osteoarthritis Compared With Power Doppler: Clinical, Radiographic and MRI Relationship. Ultrasound Med Biol 46:566–574. https://doi.org/10.1016/j.ultrasmedbio.2019.11.017 Samel NS, Mashimo H (2019) Application of OCT in the Gastrointestinal Tract. Appl Sci 9:2991. https://doi.org/10.3390/app9152991 Kang CM, Lee S-H, Chung CC (2018) Discrete-Time LPV Observer With Nonlinear Bounded Varying Parameter and Its Application to the Vehicle State Observer. IEEE Trans Ind Electron 65:8768–8777. https://doi.org/10.1109/TIE.2018.2813961 Khan SR, Desmulliez MPY (2019) Towards a Miniaturized 3D Receiver WPT System for Capsule Endoscopy. Micromachines 10:545. https://doi.org/10.3390/mi10080545 Vedaei SS, Wahid KA (2021) A localization method for wireless capsule endoscopy using side wall cameras and IMU sensor. Sci Rep 11:1–16. https://doi.org/10.1038/s41598-021-90523-w Narmatha P, Thangavel V, Vidhya DS (2022) A Hybrid RF and Vision Aware Fusion Scheme for Multi-Sensor Wireless Capsule Endoscopic Localization. Wirel Pers Commun 123:1593–1624. https://doi.org/10.1007/s11277-021-09205-5 Vedaei SS, Wahid KA (2022) MagnetOFuse: A Hybrid Tracking Algorithm for Wireless Capsule Endoscopy Within the GI Track. IEEE Trans Instrum Meas 71:1–11. https://doi.org/10.1109/TIM.2022.3204103 Rochmawati N, Fatichah C, Amaliah B, et al (2025) Deep Learning-Based Lesion Detection in Endoscopy: A Systematic Literature Review. IEEE Access 13:43532–43556. https://doi.org/10.1109/ACCESS.2025.3548167 Rathnamala DS, Rajagopal A, Arunachalam M, Dhanasekaran V (2025) GI Bleeding Detection in WCE Images Using E-ORB and ME-DEEP CAPSNET. Int J Environ Sci 11:147–160. https://doi.org/10.64252/358k2389 Zhang Y, Zhang Y, Huang X (2021) Development and Application of Magnetically Controlled Capsule Endoscopy in Detecting Gastric Lesions. Gastroenterol Res Pract 2021:1–7. https://doi.org/10.1155/2021/2716559 Giordano A, Romero-Mascarell C, González-Suárez B, Guarner-Argente C (2025) Integration of Artificial Intelligence-Enhanced Capsule Endoscopy in Clinical Practice: A Review of Market-Available Tools for Clinical Practice. Dig Dis Sci. https://doi.org/10.1007/s10620-025-09099-4 Putra KT, Arrayyan AZ, Hayati N, et al (2024) A Review on the Application of Internet of Medical Things in Wearable Personal Health Monitoring: A Cloud-Edge Artificial Intelligence Approach. IEEE Access 12:21437–21452. https://doi.org/10.1109/ACCESS.2024.3358827 Gupta J, Pathak S, Kumar G (2022) Deep Learning (CNN) and Transfer Learning: A Review. J Phys Conf Ser 2273:012029. https://doi.org/10.1088/1742-6596/2273/1/012029 Cheng Z, Liao B, He Z, et al (2019) Joint Design of the Transmit and Receive Beamforming in MIMO Radar Systems. IEEE Trans Veh Technol 68:7919–7930. https://doi.org/10.1109/TVT.2019.2927045 Gubbi J, Buyya R, Marusic S, Palaniswami M (2013) Internet of Things (IoT): A vision, architectural elements, and future directions. Futur Gener Comput Syst 29:1645–1660. https://doi.org/10.1016/j.future.2013.01.010 Osama M, Ateya AA, Sayed MS, et al (2023) Internet of Medical Things and Healthcare 4.0: Trends, Requirements, Challenges, and Research Directions. Sensors 23:. https://doi.org/10.3390/s23177435 Yang LL, Wang XF, Ke YH, et al (2022). Dielectric Patch Antenna Self-Decoupling by Proper Structural Parameters. IEEE Antennas Wirel Propag Lett 21:1447–1451. https://doi.org/10.1109/LAWP.2022.3171198 Bollineni C, Sharma M, Hazra A, et al (2025) IoT for Next-Generation Smart Healthcare: A Comprehensive Survey. IEEE Internet Things J PP:1. https://doi.org/10.1109/JIOT.2025.3570188 Zhao G, Ban Y, Zhang Z, et al (2023) Phase Demodulation Strategy Based on Kalman Filter for Sinusoidal Encoders. IEEE Sens J 23:10625–10632. https://doi.org/10.1109/JSEN.2023.3264846 Anderson K (2024) ‘The role of edge computing and hybrid clouds in next-generation healthcare. Univ Cambridge, Cambridge, UK, Tech Rep Fontana S, D’Agostino S, Paffi A, et al (2024) State of the Art on Advancements in Wireless Capsule Endoscopy Telemetry: A Systematic Approach. IEEE Open J Antennas Propag 5:1282–1294. https://doi.org/10.1109/OJAP.2024.3409827 Huhulea EN, Huang L, Eng S, et al (2025) Artificial Intelligence Advancements in Oncology: A Review of Current Trends and Future Directions. Biomedicines 13:1–18. https://doi.org/10.3390/biomedicines13040951 Pitt J, Dryzek J, Ober J (2020) Algorithmic reflexive governance for socio-techno-ecological systems. IEEE Technol Soc Mag 39:52–59. https://doi.org/10.1109/MTS.2020.2991500 Nan J, Xu L (2022) Designing Interoperable Healthcare Services Based on HL7 FHIR : A Literature Review towards Motivations, Techniques, and Applications Table of Contents Mohseni M, Amirghafouri F, Pourghebleh B (2023) CEDAR: A cluster-based energy-aware data aggregation routing protocol in the internet of things using capuchin search algorithm and fuzzy logic. Peer-to-Peer Netw Appl 16:189–209. https://doi.org/10.1007/s12083-022-01388-3 Sankar S, Ramasubbareddy S, Luhach AK, et al (2022) NCCLA: new caledonian crow learning algorithm based cluster head selection for Internet of Things in smart cities. J Ambient Intell Humaniz Comput 13:4651–4661. https://doi.org/10.1007/s12652-021-03503-3 Srinivasulu M, Shivamurthy G, Venkataramana B (2023) Quality of service aware energy efficient multipath routing protocol for internet of things using hybrid optimization algorithm. Multimed Tools Appl 82:26829–26858. https://doi.org/10.1007/s11042-022-14285-x Lei C (2024) An energy-aware cluster-based routing in the Internet of things using particle swarm optimization algorithm and fuzzy clustering. J Eng Appl Sci 71:135. https://doi.org/10.1186/s44147-024-00464-0 Wahab H, Mehmood I, Ugail H, et al (2024) Federated Deep Learning for Wireless Capsule Endoscopy Analysis: Enabling Collaboration Across Multiple Data Centers for Robust Learning of Diverse Pathologies. Futur Gener Comput Syst 152:361–371. https://doi.org/10.1016/j.future.2023.10.007 Irshad RR, Sohail SS, Hussain S, et al (2023) Towards enhancing security of IoT-Enabled healthcare system. Heliyon 9:e22336. https://doi.org/10.1016/j.heliyon.2023.e22336 Flores Fernández A, Sánchez Morales E, Botsch M, et al (2023) Generation of Correction Data for Autonomous Driving by Means of Machine Learning and On-Board Diagnostics. Sensors 23:1–28. https://doi.org/10.3390/s23010159 Saif S, Das P, Biswas S, et al (2024) A secure data transmission framework for IoT enabled healthcare. Heliyon 10:e36269. https://doi.org/10.1016/j.heliyon.2024.e36269 Li C, Wang J, Wang S, Zhang Y (2024) A review of IoT applications in healthcare. Neurocomputing 565:127017. https://doi.org/10.1016/j.neucom.2023.127017 Kaur K, Kaur A, Gulzar Y, Gandhi V (2024) Unveiling the core of IoT: comprehensive review on data security challenges and mitigation strategies. Front Comput Sci 6:. https://doi.org/10.3389/fcomp.2024.1420680 Velho L (2009) Compressive Sensing In Image Compression Guermazi A, Omoumi P, Tordjman M, et al (2024) How AI May Transform Musculoskeletal Imaging. Radiology 310:. https://doi.org/10.1148/radiol.230764 Chen F, Chandrakasan AP, Stojanoviæ VM (2012) Design and analysis of a hardware-efficient compressed sensing architecture for data compression in wireless sensors. IEEE J Solid-State Circuits 47:744–756. https://doi.org/10.1109/JSSC.2011.2179451 Ziegelmayer S, Marka AW, Strenzke M, et al (2025) Speed and efficiency: evaluating pulmonary nodule detection with AI-enhanced 3D gradient echo imaging. Eur Radiol 35:2237–2244. https://doi.org/10.1007/s00330-024-11027-5 Banitalebi-Dehkordi M, Abouei J, Plataniotis KN (2014) Compressive-sampling-based positioning in wireless body area networks. IEEE J Biomed Heal Informatics 18:335–344. https://doi.org/10.1109/JBHI.2013.2261997 Duart X, Quiles E, Suay F, et al (2021) Evaluating the effect of stimuli color and frequency on SSVEP. Sensors (Switzerland) 21:1–19. https://doi.org/10.3390/s21010117 Ali HA, Rashed EA, Kudo H (2025) Compressed sensing-based image reconstruction for discrete tomography with sparse view and limited angle geometries. PLoS One 20:1–26. https://doi.org/10.1371/journal.pone.0327666 Ozgurun B, Lafci B, Razansky D (2025) Sparse optoacoustic sensing with convolutional dictionary learning. XX:1–11. https://doi.org/10.1109/TBME.2025.3589329 Park Y, Jeon B (2020) An acquisition method for visible and near infrared images from single CMYG color filter array-based sensor. Sensors (Switzerland) 20:1–18. https://doi.org/10.3390/s20195578 Saputra OD, Murti FW, Irfan M, et al (2018) Reducing power consumption of wireless capsule endoscopy utilizing compressive sensing under channel constraint. J Inf Commun Converg Eng 16:130–134. https://doi.org/10.6109/jicce.2018.16.2.130 Krishnan S, Abdel-Hafez M, Hämäläinen M (2025) Location estimation of UWB-based wireless capsule endoscopy using TDoA in various gastrointestinal simulation models. PLoS One 20:1–32. https://doi.org/10.1371/journal.pone.0319167 Suveren M, Akay R, Kanaan M (2022) Localization of an ultra wide band wireless endoscopy capsule inside the human body using received signal strength and centroid algorithm. Int J Optim Control Theor Appl 12:151–159. https://doi.org/10.11121/ijocta.2022.1146 Stoica BA, Sahoo SK, Larus JR, Adve VS (2019) Wok: Statistical program slicing in production. Proc − 2019 IEEE/ACM 41st Int Conf Softw Eng Companion, ICSE-Companion 2019 324–325. https://doi.org/10.1109/ICSE-Companion.2019.00136 Drakoulogkonas P, Apostolou D (2021) On the selection of process mining tools. Electron 10:1–24. https://doi.org/10.3390/electronics10040451 Yao Y, Tan J, Wu J, Zhang X (2022) A Unified Fuzzy Control Approach for Stochastic High-Order Nonlinear Systems With or Without State Constraints. IEEE Trans Fuzzy Syst 30:4530–4540. https://doi.org/10.1109/TFUZZ.2022.3155297 Liu G, Zhao P, Zhao M, et al (2020). Electromagnetic disturbed mechanism of electronic current transformer acquisition card under high frequency electromagnetic interference. Electron 9:1–18. https://doi.org/10.3390/electronics9081293 Salman D (2025). Optimized Image Compression Using Sparse Representations and Fourier Transform. East J Eng 1:62–75. https://doi.org/10.63496/eje.vol1.iss1.36 Razzaque M, Dobson S (2014) Energy-Efficient Sensing in Wireless Sensor Networks Using Compressed Sensing. Sensors 14:2822–2859. https://doi.org/10.3390/s140202822 Miao M, Ye C, Xu Z, et al (2025). Optimization Scheme for Modulation of Data Transmission Module in Endoscopic Capsule. Sensors 25:4738. https://doi.org/10.3390/s25154738 Barrett HH, Myers KJ, Hoeschen C, et al (2015). Task-based measures of image quality and their relation to radiation dose and patient risk. Phys Med Biol 60:R1–R75. https://doi.org/10.1088/0031-9155/60/2/R1 Ali B H B, Ramachandran P (2022) Compressive Domain Deep CNN for Image Classification and Performance Improvement Using Genetic Algorithm-Based Sensing Mask Learning. Appl Sci 12:6881. https://doi.org/10.3390/app12146881 Anand H, Mathur S (2019) Sustainable Communication Networks and Application Li Y, Li C, Tian M, et al (2020) Two-Step Thresholds TBD Algorithm for Time Sensitive Target Based on Dynamic Programming. IEEE Access 8:209267–209277. https://doi.org/10.1109/ACCESS.2020.3038190 Zanetti F, Bergamaschi L (2020). Scalable block preconditioners for linearized Navier-Stokes equations at high Reynolds number. Algorithms 13:1–27. https://doi.org/10.3390/A13080199 O’Hara FJ, Mc Namara D (2023) Capsule endoscopy with artificial intelligence-assisted technology: Real-world usage of a validated AI model for capsule image review. Endosc Int Open 11:E970–E975. https://doi.org/10.1055/a-2161-1816 El-Gammal EM, El-Shafai W, Taha TE, et al (2025) A survey of artificial intelligence models for wireless capsule endoscopy videos for superior automatic diagnosis: problems and solutions. Springer US Jameil AK, Al-Raweshidy H (2025) A digital twin framework for real-time healthcare monitoring: leveraging AI and secure systems for enhanced patient outcomes. Discover Internet Things 5: https://doi.org/10.1007/s43926-025-00135-3 Monika R, Dhanalakshmi S, Rajamanickam N, et al (2024) Coefficient-Shuffled Variable Block Compressed Sensing for Medical Image Compression in Telemedicine Systems. Bioengineering 11:1–15. https://doi.org/10.3390/bioengineering11111101 Chen X, Huang L, Ren L (2025) Multitasking Smart Intestinal Capsule Robot : A Cutting-Edge Platform for Sampling, Diagnosis, and Therapy. 1–6. https://doi.org/10.1002/adrr.202500056 McDermott O, Stam DL, Duarte S, Sony M (2025) Readiness for Industry 4.0 in a Medical Device Manufacturer as an Enabler for Sustainability, a Case Study. Sustain 17:1–20. https://doi.org/10.3390/su17010357 Hajjar A El, Rey JF (2020) Artificial intelligence in gastrointestinal endoscopy: General overview. Chin Med J (Engl) 133:326–334. https://doi.org/10.1097/CM9.0000000000000623 Hanscom M, Cave DR (2022) Endoscopic capsule robot-based diagnosis, navigation and localization in the gastrointestinal tract. Front Robot AI 9:1–16. https://doi.org/10.3389/frobt.2022.896028 McCausland C, Biglarbeigi P, Bond R, et al (2022) Time-Frequency Ridge Analysis of Sleep Stage Transitions. In: 2022 IEEE Signal Processing in Medicine and Biology Symposium (SPMB). IEEE, pp 1–5 Ren W, Dimarogonas D V. (2020) Symbolic Abstractions for Periodic Event-triggered Linear Control Systems. In: 2020 European Control Conference (ECC). IEEE, pp 2062–2067 Chen J, Xia K, Zhang Z, et al (2024) Establishing an AI model and application for automated capsule endoscopy recognition based on convolutional neural networks (with video). BMC Gastroenterol 24:. https://doi.org/10.1186/s12876-024-03482-7 Cite Share Download PDF Status: Under Review Version 1 posted Reviewers agreed at journal 17 May, 2026 Reviewers invited by journal 06 May, 2026 Editor assigned by journal 25 Feb, 2026 First submitted to journal 20 Feb, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8896919","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":635227877,"identity":"2d2304c0-50b3-44b4-8fd9-52cc42181721","order_by":0,"name":"Zeinab Javid","email":"data:image/png;base64,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","orcid":"","institution":"École de technologie supérieure: Ecole de technologie superieure","correspondingAuthor":true,"prefix":"","firstName":"Zeinab","middleName":"","lastName":"Javid","suffix":""},{"id":635227878,"identity":"1329dd62-3564-451e-bac4-304f4821a8a8","order_by":1,"name":"Michel Kadoch","email":"","orcid":"","institution":"École de technologie supérieure: Ecole de technologie superieure","correspondingAuthor":false,"prefix":"","firstName":"Michel","middleName":"","lastName":"Kadoch","suffix":""}],"badges":[],"createdAt":"2026-02-17 01:49:55","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8896919/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8896919/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":109296500,"identity":"87336a98-9668-4d43-8739-4ba1d3675c74","added_by":"auto","created_at":"2026-05-15 08:47:34","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":118759,"visible":true,"origin":"","legend":"\u003cp\u003eCapsule endoscopy market size by region (2018–2030, US$B). Redrawn from data reported by Grand View Research [2].\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-8896919/v1/61d05cbd3e4d8b1175a42532.png"},{"id":109286712,"identity":"3d6f094b-40a0-40a7-a698-9f8d2a8d7c16","added_by":"auto","created_at":"2026-05-15 02:36:24","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":159702,"visible":true,"origin":"","legend":"\u003cp\u003eSchematic representation of a wireless capsule endoscope showing its main components: optical dome, camera, light sources, ASIC/processor, batteries, and antenna.\u003c/p\u003e","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-8896919/v1/10aff80c9005d1cef6ac77cc.png"},{"id":109297871,"identity":"19230f16-5a53-4152-aaab-9c7538783f27","added_by":"auto","created_at":"2026-05-15 09:07:11","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":37212,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of WCE communication methods (UWB, BLE, Sub-GHz, Hybrid) in terms of throughput (Mb/s, log scale) and penetration (relative). Data adapted from representative studies [90]\u003cstrong\u003e–\u003c/strong\u003e[94].\u003c/p\u003e","description":"","filename":"Figure3.png","url":"https://assets-eu.researchsquare.com/files/rs-8896919/v1/c081c26c5bead9865a4163e0.png"},{"id":109286714,"identity":"6a18429e-4e22-4b68-960b-913c9bed1b96","added_by":"auto","created_at":"2026-05-15 02:36:24","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":96193,"visible":true,"origin":"","legend":"\u003cp\u003eIoT-enabled architecture of Wireless Capsule Endoscopy systems, illustrating the three core layers: perception (smart capsule sensors), network (wearable/edge gateways), and application (cloud platforms and clinical dashboards).\u003c/p\u003e","description":"","filename":"Figure4.png","url":"https://assets-eu.researchsquare.com/files/rs-8896919/v1/bc43c3f67021f2d2ad979a5f.png"},{"id":109296334,"identity":"78277561-63e3-4ba1-9554-b9fa3b7f4f9c","added_by":"auto","created_at":"2026-05-15 08:46:32","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":125183,"visible":true,"origin":"","legend":"\u003cp\u003eComparative radar chart of AI, IoT, and CS in WCE, illustrating diagnostic accuracy, power consumption, real-time capability, integration feasibility, hardware constraints, security, data efficiency, and TRL\u003c/p\u003e","description":"","filename":"Figure5.png","url":"https://assets-eu.researchsquare.com/files/rs-8896919/v1/484adbfa31954782c46fa50e.png"},{"id":109300615,"identity":"92645b45-7ff4-4f4d-89a9-57e628ae394c","added_by":"auto","created_at":"2026-05-15 09:22:34","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1108265,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8896919/v1/a1a139b7-5d09-4d53-8da2-732558b9034d.pdf"}],"financialInterests":"","formattedTitle":"A Review of Novel Approaches to Indoor Localization Based on Wireless Capsule Endoscopy","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eGastrointestinal (GI) disorders represent a significant global health burden, with the Global Burden of Disease Study reporting that in 2019 alone, digestive diseases caused over 1.3\u0026nbsp;million deaths and accounted for more than 245\u0026nbsp;million incident cases worldwide. Beyond mortality, these conditions are also among the leading contributors to Disability-Adjusted Life Years (DALYs), imposing substantial healthcare costs and reducing workforce productivity [1]. These alarming figures underscore the urgent need for non-invasive, efficient, and patient-friendly diagnostic solutions. Capsule Endoscopy (CE) has transformed small bowel imaging by providing a wireless, less invasive alternative to conventional endoscopic procedures. The potential of CE is further amplified through integration with advanced technologies such as Artificial Intelligence (AI), the Internet of Things (IoT), and Compressed Sensing (CS), which collectively enhance diagnostic accuracy, connectivity, and energy efficiency. The market growth trend is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e1\u003c/span\u003e, based on Grand View Research\u0026rsquo;s report [2].\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eWireless Capsule Endoscopy (WCE) is specifically designed to evaluate Small Bowel (SB) disorders that are often inaccessible with conventional endoscopy [3], [4]. Its ease of use\u0026mdash;requiring neither auxiliary tools nor sedation\u0026mdash;minimizes patient discomfort and enables visualization of otherwise unreachable GI segments [5]. Recent advances in capsule design, such as multi-camera systems that provide 360\u0026deg; imaging, have expanded diagnostic coverage [3], [4]. In parallel with hardware innovations, AI-driven tools such as Express View and Suspected Blood Indicator are increasingly integrated into wireless capsule endoscopy systems [6], [7]. A single capsule can generate more than 60,000 images during transit, and convolutional neural networks have achieved detection sensitivities approaching 99.9%, reducing physician review times from nearly 100 minutes to just a few minutes [8]. Despite these advances, a persistent unmet need is highlighted in recent reviews: clinicians still face 30\u0026ndash;120 minutes of video analysis per patient, with fatigue and diagnostic variability reducing efficiency [9], [10]. In addition, Ultra-Wide-Band (UWB) communication has been investigated to support high-volume data transmission at rates exceeding 100 Mb/s while improving energy efficiency [7].\u003c/p\u003e \u003cp\u003eManaging the large volume of imaging data remains a challenge in WCE, requiring efficient compression algorithms that reduce data volume without degrading image quality [11\u0026ndash;13]. Designers must balance image resolution and frame rate with limited hardware resources and battery capacity[14], [15]. Application-Specific Integrated Circuits ASICs) have emerged as a key solution, enabling optimized image processing, compression, and signal transmission at minimal energy cost.\u003c/p\u003e \u003cp\u003eSuch developments are critical to improving capsule autonomy while maintaining diagnostic accuracy [16].\u003c/p\u003e \u003cp\u003eArtificial Intelligence (AI) has further advanced automated lesion detection in WCE. CNN-based models have demonstrated high sensitivity in identifying abnormalities while significantly reducing physicians' time burden [8]. Automated bleeding detection has been a particular focus, as rapid identification is crucial to prevent adverse outcomes [7]. Nonetheless, AI adoption in clinical practice remains limited by professional resistance, the lack of standardized diagnostic protocols, and insufficient validation across diverse populations [8], [17]. These limitations reflect the broader challenge that, while AI excels in sensitivity, it still falls short of replacing expert decision-making in real-world workflows [9], [10].\u003c/p\u003e \u003cp\u003eAccurate localization of the capsule is another critical requirement for clinical translation. Visual Odometry methods estimate position by tracking feature-point motion across consecutive frames [18], whereas magnetic-based localization uses miniature internal magnets that are tracked by external sensors [7]. These approaches support motion trajectory reconstruction and even three-dimensional mapping of the gastrointestinal tract, with reported localization errors as low as 3.5 mm [19]. However, state-of-the-art reviews emphasize that no existing technique can consistently guarantee sub-5 mm positional error or \u0026lt;\u0026thinsp;5\u0026deg; orientation error in vivo [20], while technical analyses highlight that hybrid multi-sensor fusion offers a promising solution, though it remains largely experimental. Hybrid systems that combine visual, magnetic, and Radio-Frequency (RF) data, along with multi-sensor fusion, have therefore become a primary research direction [19].\u003c/p\u003e \u003cp\u003eThe integration of capsule-based sensors has further expanded diagnostic and monitoring capabilities. These devices incorporate sensing elements, microcontrollers, RF transmission units, and power sources, enabling real-time physiological monitoring and communication [21]. Some capsules can transmit data directly to smartphones via Bluetooth, while advanced RF modules ensure reliable connectivity with external receivers [16]. By leveraging mobile cloud computing and 5G technologies, WCE platforms are poised to support remote monitoring and personalized healthcare. To address energy limitations, research has focused on wireless power transfer\u0026mdash;particularly near-field systems\u0026mdash;that deliver contactless energy to capsules, as well as self-powered designs that harvest chemical or mechanical energy [21]. In addition, edible electronic components fabricated from biocompatible materials provide safe, degradable options for in-body applications [7].\u003c/p\u003e \u003cp\u003eDespite rapid progress, miniaturization, energy efficiency, and biocompatibility remain persistent challenges. Many laboratory prototypes still struggle to integrate all essential functionalities into a safe, swallowable capsule, limiting their translation to large-scale in vivo use [21], [22]. Addressing these constraints will be essential to enable multifunctional capsules that combine imaging, sensing, actuation, and communication within clinically viable dimensions.\u003c/p\u003e \u003cp\u003eAI and IoT are expected to play a central role in the next generation of capsule robots. Although AI has improved diagnostic accuracy, it cannot yet fully replace expert endoscopists in patient-specific decision-making [18]. Clinical translation is further challenged by high costs, professional resistance, and limited availability of standardized diagnostic protocols [8], [17]. The absence of large, high-quality datasets, coupled with the \u0026ldquo;black box\u0026rdquo; nature of many algorithms, also hampers trust and adoption. Nonetheless, capsule robots highlight the potential of AI to enhance precision, personalization, and real-time navigation [5], [21]. Ethical considerations such as accountability, equitable access, and overreliance on automated tools must also be addressed [17]. To complement clinical expertise effectively, future research should prioritize hardware-integrated, energy-efficient AI systems capable of reliable real-world performance [22], [23].\u003c/p\u003e \u003cp\u003eAnother major barrier is the enormous volume of video data generated during WCE procedures. Reviewing such data is time-intensive, and accurate localization after ingestion remains technically challenging. While tools such as bleeding detection and depth estimation can provide guidance, most deep learning models are too computationally heavy to be implemented directly within capsules. Recent optimizations, however, have enabled compact models capable of onboard bleeding segmentation and depth estimation, marking a significant step toward real-time in situ analysis [24]. Hybrid localization strategies that fuse complementary sensing technologies are also under investigation to achieve high-accuracy, real-time tracking, thereby supporting both diagnostic precision and emerging telemedicine applications [19], [20].\u003c/p\u003e \u003cp\u003eElectromagnetic safety is another fundamental consideration. Parameters such as Specific Absorption Rate (SAR) and current density are strictly regulated to ensure safe operation. Recent innovations, including biodegradable \u0026ldquo;transient electronics,\u0026rdquo; demonstrate the potential of devices that safely dissolve under physiological conditions [21]. Future capsule robots will likely incorporate these materials alongside advanced energy management strategies to balance inference accuracy, memory capacity, computational speed, and energy efficiency\u0026mdash;challenges that must be resolved to achieve reliable, real-time performance [7].\u003c/p\u003e \u003cp\u003eSmart capsules are rapidly emerging as a next-generation diagnostic and monitoring platform for gastrointestinal health [22]. These devices are expected to play a central role in intelligent healthcare systems by improving resource efficiency, reducing hospitalizations, and empowering patients to manage their health actively [16], [17]. The SmartPill\u0026reg; system exemplifies this progress by measuring pH, temperature, and pressure during GI transit [25]. Beyond this, next-generation capsules are being equipped with sensors to monitor diverse physiological parameters, enabling more comprehensive assessment of the gut [21], [22]. By combining AI, IoT, and biomedical engineering, smart capsules hold promise for individualized treatments, real-time disease tracking, and timely therapeutic interventions [26]. Emerging robotic platforms, such as those designed for microbiota sampling, illustrate new directions with strong potential to transform diagnostics, monitoring, and therapy [27], [7].\u003c/p\u003e \u003cp\u003ePrevious reviews have often focused on imaging hardware or AI developments in isolation [9], [10]. To the best of our knowledge, few have systematically integrated perspectives from AI, IoT, and CS. This review uniquely bridges these domains, highlighting how their convergence accelerates the development of smart capsule platforms and paves the way for multifunctional robotic capsules with diagnostic and therapeutic potential [27]. By mapping these cross-disciplinary advances, the present review aims to serve as both a comprehensive reference for researchers and a roadmap for future innovations in autonomous, intelligent capsule endoscopy.\u003c/p\u003e \u003cp\u003eSpecifically, this review highlights the synergistic role of AI, IoT, and compressed sensing in advancing smart capsule endoscopy, with a focus on data transmission, lesion detection, and localization.\u003c/p\u003e \u003cp\u003eThe paper is organized as follows:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003epresents the review methodology, including literature selection, screening, and synthesis procedures.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003ereports the results of the structured literature analysis and categorizes the reviewed studies.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003ediscusses the key methodologies, mathematical frameworks, and analytical tools employed across AI, IoT, and CS in WCE research.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eoutlines the fundamentals and clinical challenges of wireless capsule endoscopy.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eexamines AI-based automated lesion detection and intelligent diagnostic models.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003ereviews the role of IoT in connectivity, telemetry, and remote monitoring architectures.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003epresents compressed sensing approaches for efficient data transmission and localization.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eintroduces the objectives and framework of the comparative study.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eoffers a critical discussion of current limitations and translational gaps.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003ehighlights future research opportunities and open challenges.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003econcludes the paper with a synthesis of findings and a summary of core enabling technologies.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e"},{"header":"2. Review Methodology (Methods / Experimental)","content":"\u003cp\u003eThis review was conducted using a structured literature-selection and synthesis workflow to identify recent and influential contributions related to wireless capsule endoscopy (WCE), with an emphasis on indoor localization and enabling technologies such as artificial intelligence (AI), the Internet of Things (IoT), and compressed sensing (CS).\u003c/p\u003e \u003cp\u003eSearch Strategy and Sources: Relevant studies were identified through searches in major scientific databases commonly used in engineering and biomedical research (e.g., IEEE Xplore, PubMed, Scopus, and Google Scholar). Search queries combined terms associated with capsule endoscopy and localization (e.g., wireless capsule endoscopy, capsule localization, indoor localization, magnetic tracking, visual odometry, RF-based localization), together with enabling-technology keywords (e.g., deep learning, CNN, IoT, telemetry, edge computing, compressed sensing, sparse reconstruction). Reference lists of highly relevant papers and recent surveys were also screened to capture additional key sources (snowballing).\u003c/p\u003e \u003cp\u003eInclusion and Exclusion Criteria: Articles were included if they (i) addressed WCE localization and/or closely related sensing and tracking approaches, or (ii) contributed enabling methods that directly impact WCE performance (e.g., AI-based lesion detection and triage, IoT connectivity and remote monitoring, CS-based data reduction and transmission efficiency). Priority was given to peer-reviewed journal and conference publications in English. Studies that were out of scope (e.g., unrelated endoscopic modalities, non-capsule localization settings without transferable methodology, or papers lacking sufficient technical description) were excluded.\u003c/p\u003e \u003cp\u003eScreening and Synthesis: Titles and abstracts were initially screened to exclude clearly irrelevant records, followed by full-text evaluation to confirm eligibility. The initial search yielded more than 200 publications. After duplicate removal and relevance screening, a subset was retained for in-depth qualitative synthesis based on technical rigor and thematic relevance. The selected studies were subsequently organized into five thematic categories aligned with the structure of this review: (1) WCE fundamentals and system constraints, (2) AI-driven lesion detection and intelligent processing, (3) IoT-enabled connectivity and remote monitoring architectures, (4) compressed sensing-based acquisition, compression, and localization efficiency, and (5) comparative integration strategies and open research challenges. These five thematic categories were later synthesized into broader technological domains for high-level trend analysis. The evidence was synthesized using a narrative analytical approach supported by comparative tables, with particular attention to methodological assumptions, implementation constraints (e.g., energy consumption, bandwidth limitations, and miniaturization), and reported performance characteristics where available.\u003c/p\u003e"},{"header":"3. Results of Literature Analysis","content":"\u003cp\u003eThe structured search and screening process resulted in the identification of a substantial body of literature addressing wireless capsule endoscopy (WCE) localization and enabling technologies. After removing duplicates and out-of-scope studies, for high-level analytical synthesis, the selected publications were further grouped into three principal technological domains: (i) AI-based lesion detection and intelligent interpretation, (ii) IoT-enabled connectivity and system-level integration, and (iii) compressed sensing (CS)-based data reduction and efficient localization frameworks.\u003c/p\u003e \u003cp\u003eTrend Analysis:\u003c/p\u003e \u003cp\u003eA clear temporal progression was observed in the literature. Early studies primarily focused on hardware-based tracking mechanisms such as magnetic field localization and RF triangulation. Over time, the research focus expanded toward software-driven intelligence, including convolutional neural networks (CNNs), hybrid AI-detection models, and real-time anomaly classification. In recent years, increasing attention has been directed toward integrated system architectures combining AI inference, IoT telemetry, and energy-aware data compression techniques.\u003c/p\u003e \u003cp\u003eTechnology Distribution:\u003c/p\u003e \u003cp\u003eAmong the reviewed studies, AI-based detection approaches represent the most rapidly growing category, particularly in automated lesion detection and real-time classification tasks. IoT-related contributions focus on system connectivity, telemetry reliability, and low-power communication strategies. CS-based approaches primarily address bandwidth limitations, sparse signal reconstruction, and efficient transmission under strict energy constraints.\u003c/p\u003e \u003cp\u003eReported Performance Trends:\u003c/p\u003e \u003cp\u003eAcross AI-driven lesion detection studies, deep learning architectures are generally associated with improved classification performance compared to earlier feature-engineering-based approaches. Nevertheless, practical deployment remains constrained by real-time processing requirements, model interpretability concerns, and dataset heterogeneity. IoT-oriented frameworks contribute to enhanced system connectivity and remote monitoring capabilities, while continuing to face limitations related to power efficiency and secure data transmission. Similarly, compressed sensing strategies support transmission load reduction and bandwidth optimization; however, reconstruction fidelity and robustness under noisy or variable clinical conditions remain active areas of investigation.\u003c/p\u003e \u003cp\u003eIdentified Gaps:\u003c/p\u003e \u003cp\u003eDespite significant progress, few studies demonstrate fully integrated end-to-end smart capsule architectures combining AI, IoT, and CS in a clinically validated environment. Most contributions remain domain-specific, indicating a translational gap between laboratory validation and real-world deployment.\u003c/p\u003e \u003cp\u003eThese results provide the foundation for the structured technical discussion presented in the following sections.\u003c/p\u003e"},{"header":"4. Methodologies, Mathematical Models, and Tools","content":"\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Methodological Taxonomy in Smart Wireless Capsule Endoscopy\u003c/h2\u003e \u003cp\u003eResearch on smart wireless capsule endoscopy (WCE) has evolved along several distinct methodological paradigms, reflecting different assumptions about data availability, system observability, and computational constraints. Existing approaches can be broadly categorized into model-based, data-driven, and hybrid methodologies, each addressing specific challenges in localization, perception, and decision support for WCE systems.\u003c/p\u003e \u003cp\u003eModel-based methodologies rely on explicit physical or mathematical representations of the underlying system dynamics and sensing mechanisms. In the context of WCE localization, magnetic-field\u0026ndash;based approaches exemplify this paradigm by formulating the capsule-positioning problem as a parameter-estimation (inverse) problem grounded in electromagnetic field models. Studies based on magnetic dipole formulations, gradient tensor analysis, or normalized source strength exploit known physical laws to infer capsule position and orientation from sensor measurements, often employing optimization or inversion techniques to estimate unknown states [18\u0026ndash;20], [28]. Similarly, vision-based and kinematic localization methods use geometric constraints, motion models, or state-space representations to estimate capsule trajectories, typically under observability assumptions and bounded noise [19], [20]. These model-based approaches offer interpretability and theoretical guarantees but may be sensitive to modelling inaccuracies, environmental disturbances, and incomplete prior knowledge.\u003c/p\u003e \u003cp\u003eIn contrast, data-driven methodologies treat WCE tasks as pattern recognition or function approximation problems, leveraging large volumes of labelled or unlabelled data to learn mappings directly from observations to outputs. Recent advances in artificial intelligence have accelerated the adoption of deep-learning\u0026ndash;based pipelines for lesion detection, frame classification, and auxiliary localization cues. Convolutional neural networks and related architectures have been widely applied to endoscopic image streams to reduce diagnostic workload and improve detection accuracy by learning discriminative features without explicit physical modelling [29\u0026ndash;34]. These approaches demonstrate strong performance in high-dimensional perceptual tasks and are particularly effective when physical models are difficult to define. However, their dependence on extensive training data, limited interpretability, and sensitivity to domain shifts remain important methodological considerations.\u003c/p\u003e \u003cp\u003eBridging these two paradigms, hybrid methodologies integrate model-based reasoning with data-driven learning to exploit the complementary strengths of both approaches. Representative studies combine physics-informed models with neural networks, for example, by applying learning-based super-resolution techniques to enhance sensor data before model inversion, or by constraining optimization with data-driven initial estimates [35\u0026ndash;37]. In magnetic localization, hybrid strategies have been proposed in which deep neural networks improve spatial resolution or estimate coarse target parameters, followed by deterministic or trust-region optimization to refine capsule position and magnetic moment estimates [37]. Such hybrid formulations reduce the dimensionality of the solution space, mitigate convergence issues, and improve robustness under real-time constraints, making them particularly attractive for clinical WCE applications.\u003c/p\u003e \u003cp\u003eOverall, the methodological landscape of smart WCE reflects a progression from purely physics-based formulations toward learning-centric and hybrid frameworks. While model-based approaches emphasize interpretability and physical consistency, data-driven methods prioritize scalability and perceptual accuracy. Hybrid methodologies are increasingly emerging as a unifying strategy, enabling robust performance by embedding learned representations within structured mathematical models. This taxonomy provides a conceptual foundation for analyzing subsequent developments in mathematical modelling and computational toolchains for smart WCE systems.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Mathematical Modelling Frameworks\u003c/h2\u003e \u003cp\u003eMathematical modelling plays a central role in smart wireless capsule endoscopy (WCE) by providing a principled foundation for localization, signal reconstruction, image understanding, and decision-making under severe physical and computational constraints. Across the literature, the dominant modelling frameworks can be grouped into state-space estimation models, inverse and sparse reconstruction models, and learning-based function approximation models, each reflecting different assumptions about system observability, noise statistics, and data availability.\u003c/p\u003e \u003cp\u003eState-space and estimation-based models are widely used for capsule localization and motion inference, in which the capsule pose is treated as a latent state that evolves over time. In magnetic and sensor-based localization, the relationship between the capsule state and measured magnetic field intensities is captured by nonlinear measurement equations derived from dipole or multipole field models [18\u0026ndash;20], [28], [38]. These formulations naturally lead to recursive estimation frameworks in which capsule position and orientation are inferred by minimizing prediction\u0026ndash;measurement mismatches under assumed noise models. Although explicit filters are not always stated, the underlying mathematical structure corresponds to nonlinear state estimation problems, where identifiability and convergence depend on excitation conditions, sensor geometry, and modelling fidelity [19], [20]. Such models emphasize physical consistency and enable uncertainty-aware inference but may degrade when the assumed field or motion models are violated in complex anatomical environments.\u003c/p\u003e \u003cp\u003eA second class of formulations arises from inverse problems and sparse reconstruction theory and is particularly relevant to signal acquisition, compression, and recovery under bandwidth and power constraints. Compressive sensing\u0026ndash;based approaches model the acquired measurements as underdetermined linear systems in which the underlying signal admits a sparse representation in a suitable dictionary or frame [39\u0026ndash;41]. Within this framework, signal recovery is cast as an optimization problem\u0026mdash;often ℓ₁-regularized or constrained\u0026mdash;whose solvability relies on sparsity assumptions and measurement incoherence. Extensions of modulo and nonlinear sampling further interpret signal recovery as an ill-posed inverse problem, in which uniqueness and stability depend on sampling rates, folding mechanisms, and prior knowledge of signal structure [42]. These mathematical models are attractive for WCE systems because they explicitly encode acquisition constraints while providing theoretical recovery guarantees, albeit at the cost of increased computational complexity.\u003c/p\u003e \u003cp\u003eIn parallel, learning-based mathematical models reformulate WCE tasks as nonlinear function approximation problems. Deep neural networks, particularly convolutional architectures, are used to approximate mappings from high-dimensional image or signal spaces to diagnostic labels, saliency maps, or auxiliary localization cues [30],[43],[44],[32]. From a modelling perspective, these approaches replace explicit physical equations with parameterized nonlinear operators optimized through empirical risk minimization. Techniques such as transfer learning and feature reuse can be interpreted as imposing implicit priors on the learned function space, mitigating data scarcity and stabilizing training dynamics [40]. While these models often achieve superior performance in perceptual tasks, their mathematical guarantees are largely empirical, and interpretability remains limited compared to physics-based formulations.\u003c/p\u003e \u003cp\u003eRecent studies increasingly explore hybrid modelling frameworks that integrate learning-based components into structured mathematical models. Examples include combining sparse reconstruction principles with neural network\u0026ndash;assisted estimation or using learned features to initialize or constrain optimization-based solvers [37]. In such cases, learning components reduce model mismatch or dimensionality, while the underlying optimization or estimation framework preserves physical interpretability and robustness. These hybrid formulations highlight a broader trend toward unifying data-driven adaptability with mathematically grounded inference in smart WCE systems.\u003c/p\u003e \u003cp\u003eOverall, the mathematical modelling landscape in WCE reflects a balance between rigour and flexibility. State-space and inverse problem formulations provide interpretability and theoretical structure, whereas learning-based models offer scalability and performance in complex perceptual domains. Hybrid frameworks are a promising approach that leverages complementary strengths to address the inherent challenges of localization, signal recovery, and diagnosis in resource-constrained ingestible platforms.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e4.3 Computational and Hardware Tools\u003c/h2\u003e \u003cp\u003eThe tight coupling between computational tools and hardware platforms fundamentally constrains the practical deployment of smart wireless capsule endoscopy (WCE) systems. Unlike conventional medical imaging pipelines that rely on off-device processing, WCE requires embedded, energy-aware, and real-time computational infrastructure that operates under severe size, power, and thermal constraints. Consequently, existing studies adopt a co-design philosophy in which algorithms, software frameworks, and hardware architectures are jointly optimized.\u003c/p\u003e \u003cp\u003eFrom a computational perspective, software toolchains for WCE can be broadly divided into offline training environments and on-device inference pipelines. Deep learning\u0026ndash;based diagnostic and perception models are typically developed and trained using high-level frameworks such as TensorFlow or PyTorch, enabling rapid prototyping, transfer learning, and large-scale optimization on external computing resources [29],[44],[32]. Once trained, these models are converted into lightweight representations suitable for embedded execution, often through model compression, quantization, or architectural simplification. In edge-AI capsule systems, such as those enabling real-time lesion detection and adaptive task execution, inference pipelines are tightly optimized to meet strict latency and memory budgets while maintaining clinically acceptable accuracy [32].\u003c/p\u003e \u003cp\u003eComplementary to learning-based pipelines, signal processing and reconstruction tasks rely on numerical optimization and linear algebra tools that are frequently implemented in environments such as MATLAB or Python-based scientific libraries [45], [46]. These tools support compressed-sensing\u0026ndash;based recovery, sparse optimization, and the simulation of acquisition constraints before hardware deployment. Although such environments are rarely embedded directly in the capsule, they play a critical role in validating reconstruction fidelity, tuning regularization parameters, and assessing robustness under realistic noise and sampling conditions.\u003c/p\u003e \u003cp\u003eAt the hardware level, system-on-chip (SoC) platforms form the backbone of modern smart WCE devices. Recent designs integrate microcontrollers, digital signal processors, and dedicated neural processing units to support on-capsule intelligence while minimizing energy consumption [32],[47]. Application-specific integrated circuits (ASICs) and field-programmable gate arrays (FPGAs) are increasingly used to accelerate convolutional operations and feature-extraction tasks, enabling real-time processing under tight power envelopes [48],[49]. These hardware accelerators enable critical computations to be executed locally, reducing reliance on continuous wireless transmission and thereby extending operational lifetime.\u003c/p\u003e \u003cp\u003eWireless communication modules are another essential hardware component, bridging embedded computation with external receivers. Technologies such as Bluetooth Low Energy (BLE), ultra-wideband (UWB), and custom RF telemetry links are used to transmit compressed data streams, intermediate inference results, or event-driven alerts [47],[49]. The choice of communication protocol directly influences computational design decisions, as bandwidth limitations motivate on-device preprocessing and selective data transmission rather than raw video streaming.\u003c/p\u003e \u003cp\u003eBeyond active electronic components, passive and semi-passive sensing hardware also plays a significant role in WCE ecosystems. Inductor\u0026ndash;capacitor (LC)\u0026ndash;based sensor architectures enable battery-free or low-power sensing through resonant coupling, enabling continuous physiological monitoring without dedicated power sources [45]. These architectures reduce computational overhead by encoding sensed information as frequency or phase shifts, which can be decoded externally with minimal on-capsule processing.\u003c/p\u003e \u003cp\u003eOverall, the computational and hardware tools employed in smart WCE systems reflect a strong emphasis on edge intelligence, hardware\u0026ndash;software co-design, and energy efficiency. Rather than treating computation as an isolated layer, existing approaches tightly integrate algorithmic complexity with hardware capabilities and communication constraints. This integrated toolchain provides the technological foundation for advanced modelling, perception, and diagnostic functions in next-generation ingestible devices.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e4.4 Assumptions, Constraints, and Trade-offs\u003c/h2\u003e \u003cp\u003eDespite significant progress in smart wireless capsule endoscopy (WCE), existing methodologies and technologies are inevitably shaped by underlying assumptions and practical constraints that impose fundamental trade-offs on system design and performance. These trade-offs arise from the capsule's ingestible nature, the variability of the gastrointestinal environment, and stringent requirements for safety, reliability, and clinical usability.\u003c/p\u003e \u003cp\u003eA primary assumption underlying many WCE systems is that sufficiently informative sensory data are available under constrained operating conditions. Vision-based approaches typically assume adequate illumination, a stable imaging geometry, and acceptable motion blur to ensure reliable perception and analysis [50], [32]. In practice, however, variable transit speeds, occlusions by intestinal contents, and non-uniform lighting conditions can significantly degrade image quality. Similarly, magnetic and RF-based localization techniques often assume quasi-static or weakly perturbed field environments, yet real anatomical settings introduce distortions, interference, and patient-specific variability that challenge these assumptions [51],[36].\u003c/p\u003e \u003cp\u003eEnergy availability constitutes one of the most critical constraints in WCE and directly governs the balance between onboard intelligence and operational lifetime. Most capsule platforms rely on limited battery capacity, necessitating careful trade-offs among sensing resolution, computation, and wireless transmission [52],[53]. Increasing frame rates, deploying deep neural networks, or enabling continuous localization inevitably raises power consumption, reducing examination duration or necessitating aggressive duty cycling. As a result, many systems assume intermittent processing or selective data transmission, prioritizing salient events over continuous high-fidelity monitoring [32]. These assumptions influence not only algorithmic design but also clinical protocols and expectations.\u003c/p\u003e \u003cp\u003eAnother important trade-off concerns accuracy versus interpretability. Data-driven models, particularly deep learning\u0026ndash;based diagnostic systems, often achieve superior performance in lesion detection and classification tasks, yet operate as black-box models with limited transparency [32]. In contrast, physics-based localization and signal reconstruction approaches offer clearer interpretability and error characterization but may exhibit reduced robustness under modelling mismatch or noise [51]. Hybrid strategies implicitly assume that learned components can compensate for model inaccuracies while preserving sufficient structure for reliable inference; however, this balance remains sensitive to the quality of the training data and the deployment conditions [36].\u003c/p\u003e \u003cp\u003eLatency and communication bandwidth further constrain system design. High-resolution image transmission and real-time feedback demand substantial wireless bandwidth, which is often unavailable or energetically prohibitive in ingestible devices [53]. Consequently, many WCE systems assume delayed or offline analysis or rely on onboard preprocessing to reduce data volume prior to transmission. This introduces trade-offs among the immediacy of clinical feedback, computational complexity, and local autonomy versus external supervision.\u003c/p\u003e \u003cp\u003eFinally, clinical deployment introduces assumptions about safety and robustness that constrain technological choices. Assumptions about biocompatibility, thermal safety, electromagnetic exposure, and mechanical reliability limit the integration of high-power actuators, aggressive locomotion mechanisms, or dense electronic components [50],[52]. These constraints often favour conservative designs that trade advanced functionality for proven safety and regulatory acceptance.\u003c/p\u003e \u003cp\u003eOverall, the evolution of smart WCE systems is characterized by ongoing negotiation among competing objectives: intelligence versus endurance, accuracy versus interpretability, autonomy versus safety, and performance versus practicality. Recognizing these assumptions and trade-offs is essential for contextualizing reported results and guiding future research toward clinically viable, balanced, and scalable solutions.\u003c/p\u003e \u003c/div\u003e"},{"header":"5. Fundamentals and Challenges in Wireless Capsule Endoscopy","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e5.1 Overview of Wireless Capsule Endoscopy\u003c/h2\u003e \u003cp\u003eWCE is a minimally invasive diagnostic technique that visualizes the gastrointestinal tract without sedation or the risks associated with conventional endoscopy [54\u0026ndash;56]. It is widely used to diagnose obscure gastrointestinal bleeding, Crohn\u0026rsquo;s disease, celiac disease, small bowel tumours, and polyposis syndromes [8]. The origins of WCE date back to early endoscopic experiments by Philipp Bozzini in the 19th century. Later, Gavriel Iddan [57] recognized the potential of fibre optics and, working with engineers, developed the first wireless capsule capable of imaging the small bowel. The basic structure and main components of a typical capsule endoscope are illustrated in Figure .2.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn 2001, Given Imaging commercialized this innovation as the M2A system (\u0026ldquo;mouth to anus\u0026rdquo;), marking a milestone in non-invasive gastrointestinal diagnostics [58\u0026ndash;61]. The capsule integrates miniature cameras, LEDs, RF transmitters, and batteries within a biocompatible shell of about 11\u0026times;26 mm [62\u0026ndash;64]. However, these devices still operate passively, moving only with natural peristalsis. This lack of control limits maneuverability and contributes to omission rates of up to 30% [65\u0026ndash;67].\u003c/p\u003e \u003cp\u003eSuch limitations underscore the need for next-generation \u0026ldquo;smart capsules\u0026rdquo; equipped with active navigation, targeted drug delivery, and biopsy capabilities [68]. While WCE has revolutionized gastrointestinal diagnostics, its passive locomotion remains a major barrier to realizing its full potential.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e5.2 Power Supply Limitations\u003c/h2\u003e \u003cp\u003eConventional wireless capsules typically rely on two silver-oxide button-cell batteries, which provide approximately 20 mW at 3 V for 8\u0026ndash;10 hours of operation [69\u0026ndash;71]. While sufficient for basic imaging, these batteries limit advanced functionality due to their limited energy capacity, output power, and safety concerns [72\u0026ndash;74]. Earlier generations also faced rapid depletion from power-hungry components such as CCD sensors [75].\u003c/p\u003e \u003cp\u003eTo overcome these constraints, researchers have explored custom lithium-ion polymer batteries that deliver much higher power density\u0026mdash;up to 2000 times greater than silver oxide cells\u0026mdash;while supporting peak currents [76\u0026ndash;78]. However, these raise safety concerns, particularly risks of thermal runaway in vivo. Alternative approaches, such as self-powered systems that harvest energy from gastric fluids, have also been tested, as have edible electronics based on food-grade materials (e.g., edible batteries, supercapacitors, and nanogenerators). Despite their biocompatibility, these strategies still fail to provide enough power for continuous capsule operation [67, 79, 80].\u003c/p\u003e \u003cp\u003eSupercapacitors store only limited energy, and nanogenerators convert mechanical or thermal stimuli into electricity but remain suitable only for low-demand functions [81\u0026ndash;83]. A promising approach is to embed piezoelectric nanogenerators in the capsule shell, which convert peristaltic motion into usable electricity [84]. Another major line of research is Wireless Power Transmission (WPT), where energy is delivered externally via electromagnetic waves. Although magnetic resonance\u0026ndash;based WPT has improved coupling efficiency, practical barriers remain, including tissue absorption, organ motion, and transmitter\u0026ndash;receiver misalignment [67, 85\u0026ndash;88].\u003c/p\u003e \u003cp\u003eIn summary, while lithium-ion batteries and WPT prototypes show great potential for extending capsule lifetime, their clinical applicability is limited by safety risks, alignment challenges, and regulatory constraints. Self-powered and edible systems demonstrate biocompatibility but remain far from supporting full diagnostic functionality. At present, energy supply remains the most critical bottleneck in WCE, highlighting the need for multidisciplinary innovations that balance power density, safety, and clinical feasibility.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e5.3 Data Transmission and Telemetry Challenges\u003c/h2\u003e \u003cp\u003eReliable high-throughput communication is critical for WCE, as demand for high-resolution imaging and for integrating multiple sensors continues to increase. Conventional systems typically operate near 400 MHz with only 300 kHz bandwidth, which restricts video quality and makes signals vulnerable to absorption, scattering, and multipath fading, particularly in deep gastrointestinal regions [89], [67],[51].\u003c/p\u003e \u003cp\u003eTo address these limitations, UWB communication has been proposed. Operating in the 3.1\u0026ndash;10 GHz range, UWB supports data rates exceeding 100 Mb/s and offers improved energy efficiency [90]. However, it requires complex RF hardware and is subject to strict regulatory compliance, which has limited its clinical adoption [91], [92]. In parallel, low-power protocols such as Bluetooth Low Energy (BLE) have been evaluated for their simplicity and energy efficiency. Yet, their performance is significantly constrained by severe attenuation at 2.4 GHz, with devices such as iMAG showing poor results through thick gastric walls. To overcome this, researchers have explored lower-frequency bands (400\u0026ndash;915 MHz) that reduce absorption and extend range, though these come at the expense of lower throughput [93].\u003c/p\u003e \u003cp\u003eHybrid solutions are also under investigation. Approaches such as dual-band modules and Intra Body Communication (IBC), leveraging capacitive or galvanic coupling, aim to achieve more robust two-way communication while maintaining energy efficiency [94]. Despite these advances, ensuring reliable bidirectional telemetry in the highly dynamic gastrointestinal environment remains a substantial challenge. Therefore, while UWB offers unmatched throughput [90], its complexity and regulatory barriers make near-term clinical deployment unrealistic [91]. BLE and sub-GHz protocols [93] are more practical for current WCE systems, although their limited bandwidth restricts continuous high-resolution imaging. Hybrid architectures that combine these modes may therefore provide the most feasible pathway toward reliable, energy-efficient capsule telemetry [94]. The relative trade-offs between throughput and penetration for UWB, BLE, sub-GHz, and hybrid protocols are illustrated in the figure. 3.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e5.4 Locomotion and Navigation Challenges\u003c/h2\u003e \u003cp\u003eControllable locomotion remains a major challenge in WCE. Current capsules rely on passive peristalsis, which lacks directional control and often results in incomplete inspection of target regions [95\u0026ndash;97]. This passive movement restricts diagnostic reliability and limits the ability to focus on suspicious lesions.\u003c/p\u003e \u003cp\u003eTo address these limitations, researchers have investigated active locomotion strategies, including bio-inspired propulsion systems such as earthworm-like crawling, screw-jet mechanisms, and robotic actuation [95], [57]. While these methods demonstrate feasibility in laboratory conditions, their integration into capsule-sized devices is limited by available volume, battery capacity, and the risk of heat generation during operation.\u003c/p\u003e \u003cp\u003eAmong alternative approaches, magnetic guidance offers five degrees of freedom (5-DOF) by manipulating internal magnets in response to external magnetic fields [51]. Recent designs employ Reciprocally Rotating Magnetic Actuation (RRMA) to minimize drag in narrow spaces and electromagnetic coil arrays (e.g., Helmholtz and Maxwell systems) for dynamic orientation control [98], [99]. Although these methods improve maneuverability, they increase power demand and depend heavily on large external equipment, which reduces clinical practicality.\u003c/p\u003e \u003cp\u003eAccurate localization also remains unresolved. Techniques based on radio-frequency signals, visual odometry, and magnetic tracking have been tested, but none consistently achieve real-time six-degree-of-freedom (6-DoF) accuracy. Hybrid approaches combining visual landmarks, magnetic mapping, and RF tracking show promise for sub-centimetre precision [28], [67], though most remain experimental and unvalidated in large-scale clinical trials.\u003c/p\u003e \u003cp\u003eTherefore, while active propulsion and magnetic guidance significantly enhance maneuverability, their reliance on bulky external systems and high energy demand limits clinical adoption. Passive peristalsis remains the only approved method in practice. Future WCE systems must resolve this trade-off by developing compact, energy-efficient locomotion and robust hybrid localization, bridging the gap between experimental prototypes and real-world clinical feasibility.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e5.5 Hardware Constraints and Miniaturization in WCE\u003c/h2\u003e \u003cp\u003eThe push for miniaturization in WCE creates major engineering challenges. Capsules must integrate imaging units, batteries, and communication hardware within dimensions of no more than 32\u0026times;12 mm, thereby imposing strict design constraints [51]. Adding advanced functions such as high-definition imaging, telemetry, or locomotion increases complexity. For example, Micromotors and Shape Memory Alloys (SMAs) can generate motion but occupy valuable space, produce heat, and respond slowly, limiting real-time navigation [100],[101].\u003c/p\u003e \u003cp\u003eThermal regulation and biocompatibility add further constraints. Devices must avoid overheating and excessive Specific Absorption Rate (SAR), particularly during high-frequency transmission or Wireless Power Transfer (WPT). Magnetic actuation systems also require coils and magnets, which reduce the available space[41].\u003c/p\u003e \u003cp\u003eAdvanced fabrication techniques, such as MEMS, have been used to increase functional density without increasing capsule size. Yet many next-generation capsules remain relatively bulky, sometimes exceeding safe ingestion thresholds, particularly in patients with strictures. Tethered or robotic capsules intended for therapeutic use also add cost and complexity [51]. A comparative overview of representative WCE systems is provided in Table\u0026nbsp;\u003cspan refid=\"Tab9\" class=\"InternalRef\"\u003e1\u003c/span\u003e [67]\u0026ndash; [51].\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSome of the most well-known WCEs with their corresponding characteristics.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026times;\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCapsule Name\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDimensions (mm)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWeight (g)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eBattery Life (hrs.)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNavigation\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eDrug Delivery\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003ePower Transfer\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eNotes\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePillCam SB3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026times;\" colname=\"c2\"\u003e \u003cp\u003e26 \u0026times; 11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e~\u0026thinsp;3.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e8\u0026ndash;12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePassive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eBattery\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eClinically established\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNaviCam\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026times;\" colname=\"c2\"\u003e \u003cp\u003e27 \u0026times; 12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e~\u0026thinsp;3.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e10\u0026ndash;12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eMagnetic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eBattery\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eRequires an external magnet\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSayaka\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026times;\" colname=\"c2\"\u003e \u003cp\u003e31 \u0026times; 13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e~\u0026thinsp;4.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePassive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eBattery\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eHigh-resolution imaging\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eiMAG\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026times;\" colname=\"c2\"\u003e \u003cp\u003e23 \u0026times; 9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;3.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e~\u0026thinsp;6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePassive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eWireless (BLE)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eExperimental\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEndocapsule 10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026times;\" colname=\"c2\"\u003e \u003cp\u003e26 \u0026times; 11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePassive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eBattery\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eImproved transmission\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eExperimental Prototype (RRMA)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026times;\" colname=\"c2\"\u003e \u003cp\u003e32 \u0026times; 13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e~\u0026thinsp;5.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e~\u0026thinsp;6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eActive (Magnetic)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eInductive WPT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eResearch phase\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAs shown in Table\u0026nbsp;\u003cspan refid=\"Tab9\" class=\"InternalRef\"\u003e1\u003c/span\u003e, increasing capsule functionality often entails trade-offs among size, weight, and battery life. This highlights the need for innovative designs that balance miniaturization with performance. Furthermore, energy management strategies such as adaptive frame rates and data compression can extend battery life but may reduce image quality [102\u0026ndash;104].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e5.6 Biopsy, Drug Delivery, and Functional Limitations\u003c/h2\u003e \u003cp\u003eExpanding WCE beyond diagnostics to therapeutic applications is among the most ambitious goals in this field. The primary targets include integrating biopsy tools, targeted drug delivery, and real-time procedures such as ablation or lesion removal. Early designs, such as the Crosby-Kugler capsule, attempted to perform a biopsy but failed due to inaccurate forceps operation, lack of feedback, and single-use limitations [97],[105]. Even advanced robotic prototypes still face trade-offs among actuation force, compactness, and biocompatibility[51].\u003c/p\u003e \u003cp\u003eDrug delivery systems\u0026mdash;such as spring reservoirs, osmotic pumps, and pH-sensitive coatings\u0026mdash;have also been tested, but variable GI motility and localization errors hinder precise release [106\u0026ndash;108]. More advanced strategies, including wireless-controlled and magnetically triggered mechanisms, show promise yet remain constrained by anatomical variability and tissue conductivity [109].\u003c/p\u003e \u003cp\u003eEfforts toward real-time therapeutic interventions\u0026mdash;such as laser ablation, microsurgery, or biopsy\u0026mdash;pose additional engineering challenges. Micro-actuated blades and needles require power-hungry modules, sterilizable materials, and rigorous safety validation, particularly in sensitive tissue regions [110]. Moreover, achieving full-duplex telemetry for reliable bidirectional control remains unresolved in commercial systems [111]. Hybrid capsules that combine biopsy and drug delivery often increase in size, raising concerns about swallowability for patients with strictures or pediatric populations [112], [113]. Ultimately, the vision of a fully autonomous \u0026ldquo;surgical capsule\u0026rdquo; remains aspirational. Its realization will depend on advances in energy-efficient actuation, wireless power transfer, and accurate real-time localization, while ensuring compliance with stringent clinical safety and regulatory standards [51].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e5.7 Regulatory and Clinical Translation Barriers\u003c/h2\u003e \u003cp\u003eDespite major technical advances, translating WCE innovations into routine clinical practice remains hindered by significant regulatory and translational challenges. Approval processes for therapeutic capsules are lengthy because mechanical components, such as drug actuators, lasers, and biopsy tools, must comply with strict safety, reliability, and sterilization standards. Even minor modifications\u0026mdash;such as changes to embedded electronics or locomotion systems\u0026mdash;often necessitate full recertification, thereby prolonging commercialization timelines [51], [67].\u003c/p\u003e \u003cp\u003eClinical readiness is further limited by insufficient validation. Many prototypes demonstrate high performance in controlled laboratory environments but fail under variable physiological conditions. The lack of large-scale clinical trials across diverse populations prevents consistent demonstration of safety and diagnostic efficacy, delaying regulatory endorsement [114\u0026ndash;116].\u003c/p\u003e \u003cp\u003eThe integration of AI introduces additional ethical and legal concerns. Deep learning models often function as \u0026ldquo;black boxes,\u0026rdquo; producing results that are difficult to interpret. In cases of diagnostic errors, accountability remains unclear. Clinicians also remain skeptical of automated outputs, particularly when dealing with ambiguous or borderline cases [51], [117].\u003c/p\u003e \u003cp\u003eInfrastructure represents another major barrier. Many hospitals lack the computational resources, network infrastructure, or trained personnel required to deploy AI-enhanced WCE. Lightweight AI frameworks optimized for edge computing have been proposed to address these gaps, while federated learning approaches aim to improve privacy and compliance with data protection laws [117\u0026ndash;120]. However, these solutions remain in the early stages of adoption. Therefore, the regulatory and clinical translation of WCE is hindered not by a single factor but by the convergence of multiple unresolved issues. Although prototypes demonstrate impressive technical promise [51], their adoption is hindered by limited clinical validation [114], [115], unclear liability in AI-assisted diagnosis [121], [122], and persistent infrastructure gaps [118]. These combined challenges reinforce regulatory and translational barriers as the most critical bottleneck. Unless a coordinated framework for safety, standardization, and oversight is established [57], advanced WCE systems risk being confined to prototypes, widening the gap between research innovation and patient care.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e5.8 Section Summary\u003c/h2\u003e \u003cp\u003eWCE is evolving from a passive diagnostic tool into a multifunctional robotic system. This transformation relies on advances in AI, Microelectromechanical Systems (MEMS), Microfabrication, and UWB communication [123\u0026ndash;126]. A major research direction is the integration of therapeutic functions. Procedures such as Endoscopic Submucosal Dissection (ESD) and Endoscopic Mucosal Resection (EMR), once limited to tethered devices, are now being reimagined for capsules through innovations in miniaturized energy sources [127], [128]. Novel mobility strategies, including magnetic navigation and swarm robotics, are also under development. Concepts such as a \u0026ldquo;mothership\u0026rdquo; capsule coordinating smaller \u0026ldquo;soldier\u0026rdquo; units highlight the potential for parallel interventions. At the same time, soft magnetic materials may enable more flexible locomotion, controlled drug delivery, and site-specific adhesion [67].\u003c/p\u003e \u003cp\u003eTherapeutic functions are advancing, with efficient CNNs and knowledge distillation enabling on-board analysis. Yet, trade-offs between speed, memory, energy, and diagnostic accuracy remain unresolved [129], [130]. In parallel, WCE is emerging as a tool for therapeutic functions, particularly when combined with 5G and cloud infrastructures. Lessons from the COVID-19 pandemic underline the importance of scalable remote diagnostics for underserved populations [67].\u003c/p\u003e \u003cp\u003eIn navigation, hybrid localization frameworks that combine magnetic mapping, visual landmarks, and RF tracking aim to provide robust and precise positioning, even in anatomically complex regions [131\u0026ndash;133].\u003c/p\u003e \u003cp\u003eLooking ahead, next-generation WCE platforms are expected to feature:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eEmbedded AI for autonomous interpretation.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eIntegration of diagnostic and therapeutic modules.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e3Enhanced navigational precision using magnetic and hybrid localization.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eSecure, high-speed communications with UWB and 5G.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eCollectively, these advances could transform WCE into an intelligent, self-guided, globally deployable platform that automates many diagnostic and therapeutic procedures [67], [51], [111]. Achieving this vision will require multidisciplinary collaboration among engineers, clinicians, regulators, and healthcare providers.\u003c/p\u003e \u003c/div\u003e"},{"header":"6. AI in Wireless Capsule Endoscopy","content":"\u003cp\u003eWCE has GI diagnostics that generate more than 50,000 image frames per exam, posing major challenges for manual review [29], [134]. To address this workload, AI has emerged as a key tool for automating interpretation, improving diagnostic efficiency, and reducing physician fatigue. AI-based systems now perform real-time analysis in WCE, detecting abnormalities and assisting decision-making [52]. Compared with handcrafted feature methods such as LBP, SIFT, or HOG, deep learning\u0026mdash;particularly CNNs\u0026mdash;achieves higher accuracy and generalization [30], [43]. CNNs have surpassed human experts on benchmark tasks such as CAMELYON17 histology classification [30]. In endoscopy, these models demonstrate high sensitivity (\u0026gt;\u0026thinsp;90%) for detecting bleeding, polyps, and ulcers, using architectures such as VGG16, ResNet50, InceptionV3, and EfficientNet [31], [35].\u003c/p\u003e \u003cp\u003eTransfer learning further enhances performance: pre-trained CNNs on ImageNet, when adapted to WCE datasets, reduce training time, mitigate overfitting, and improve detection accuracy, particularly for polyp classification [134], [44]. Edge AI is also being explored, where frames can be prioritized or filtered in real time to reduce redundant transmission [32], [44].\u003c/p\u003e \u003cp\u003eClinical benefits are evident. AI-assisted review reduces interpretation time by more than 40% while maintaining or improving diagnostic performance [135]. In trials, AI identified more than 95% of bleeding lesions missed by human reviewers [33], and hybrid systems combining deep learning with rule-based logic achieved higher specificity and F1 Scores [34].\u003c/p\u003e \u003cp\u003eChallenges remain, including inconsistent image quality, lighting issues, and limited annotated datasets [52]. Class imbalance is particularly problematic, as lesion frames may constitute\u0026thinsp;\u0026lt;\u0026thinsp;2% of the data, increasing the risk of false negatives [136], [53]. To address this, methods such as augmentation, GAN-based synthesis, weak supervision, and anomaly detection have been applied [137], [138].\u003c/p\u003e \u003cp\u003eEthical and regulatory aspects are also critical. These include ensuring interpretability, mitigating bias, and validating across diverse populations, with AI positioned as a decision-support tool rather than a replacement for clinicians [135], [139].\u003c/p\u003e \u003cp\u003eIn summary, AI adoption in WCE is driven by the need to manage large volumes of images, detect subtle lesions, and streamline clinical workflows. While deep learning has demonstrated strong potential, future research must focus on robustness, transparency, and adaptability. Section \u003cspan refid=\"Sec19\" class=\"InternalRef\"\u003e6.1\u003c/span\u003e next examines CNN architectures and transfer learning as the foundation of AI-driven lesion detection in WCE.\u003c/p\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003e6.1 CNNs and Transfer Learning in Wireless Capsule Endoscopy\u003c/h2\u003e \u003cp\u003eCNNs are the central technology for AI-based lesion detection in WCE. Their hierarchical architecture enables automatic feature extraction, thereby avoiding reliance on handcrafted descriptors such as LBP, Gabor filters, and SIFT [30]. These models can capture complex spatial and texture-based patterns in GI tract images. Early research trained CNN architectures, such as AlexNet, VGGNet, and GoogLeNet, on labelled WCE datasets. These implementations showed that CNNs can differentiate normal from abnormal frames by learning convolutional filters directly from image data. For example, a 6-layer CNN achieved 93.4% sensitivity in bleeding detection on the Kvasir dataset [31], and GoogLeNet reached 96.36% accuracy in polyp classification [35].\u003c/p\u003e \u003cp\u003eA major limitation of deep CNNs is the need for large labelled datasets, which are often unavailable. Transfer learning addresses this issue by fine-tuning models pre-trained on datasets such as ImageNet. This approach accelerates training and improves classification on small GI datasets [134]. Studies report that fine-tuned ResNet50 and InceptionV3 models outperform custom-built CNNs by up to 8% in ulcer and polyp detection [44]. Several architectures have been explored for WCE tasks, such as:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eInceptionV3 and EfficientNet [32]: Strong performance relative to model size. EfficientNetB0 delivered 94.2% accuracy with fewer FLOPs, making it suitable for portable systems.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eResNet variants [33]: Skip connections ease gradient flow. ResNet50 achieved an AUC of 0.96 in ulcer classification, about 15% better than standard CNNs.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eVGG16 and VGG1[136]: Deep simple convolutional stacks with high sensitivity but high computational cost.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eBeyond classification, CNNs have been adapted for segmentation. Encoder\u0026ndash;decoder architectures, such as U-Net variants, have achieved Dice scores above 0.90 in bleeding segmentation [135].Capsule Networks (CapsNet) also show promise for handling image rotations and geometric distortions in capsule footage [34]. Transfer learning has proven valuable in multi-class classification, including bleeding, polyps, erosions, and tumours\u0026mdash;a DenseNet121-based model classified five lesion categories with an average F1-score of 91.8% [29]. To improve domain adaptation, strategies such as freezing early layers, applying differential learning rates, and visualizing intermediate features have been applied. Freezing shallow layers while fine-tuning deeper ones aligns models with WCE-specific colour tones and textures [52]. Visualization methods such as CAM and Grad-CAM help interpret predictions and build clinician confidence [139].\u003c/p\u003e \u003cp\u003eDespite these advances, CNNs face generalization issues. Models trained on one dataset often perform poorly on other datasets due to differences in hardware, lighting, and annotation [137]. Class imbalance is another challenge: normal frames vastly outnumber pathological ones, biasing models toward the majority class [53]. Solutions include cost-sensitive loss functions, oversampling, and hard example mining [43]. Lightweight CNNs have also been developed for real-time edge processing. A depthwise separable CNN achieved 28 FPS on a Raspberry Pi-based WCE receiver, showing the feasibility of on-device inference for lesion detection [44]. Such approaches can reduce transmission costs and power consumption in constrained settings. In conclusion, CNNs\u0026mdash;particularly when augmented with transfer learning\u0026mdash;underpin AI systems for WCE lesion detection and classification. Ongoing improvements in architectures, generalization, and interpretability continue to drive their adoption. While CNNs and transfer learning remain central, they are not sufficient to capture all lesion types. Section \u003cspan refid=\"Sec20\" class=\"InternalRef\"\u003e6.2\u003c/span\u003e, therefore, examines AI methods for detecting specific gastrointestinal lesions such as polyps, bleeding, tumours, and erosions.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003e6.2 AI Detection of Specific Lesions in WCE: Polyps, Bleeding, Tumours, and Erosions\u003c/h2\u003e \u003cp\u003eDeep learning methods have achieved strong performance in GI lesions in WCE, including polyps, bleeding, tumours, and mucosal erosions/ulcers. Each lesion type presents unique visual challenges for both clinicians and algorithms.\u003c/p\u003e \u003cdiv id=\"Sec21\" class=\"Section3\"\u003e \u003ch2\u003e6.2.1 Polyp Detection\u003c/h2\u003e \u003cp\u003ePolyps are difficult to identify due to their flat structures and similarity to mucosa [52]. CNNs such as VGG16, ResNet, and InceptionV3 remain standard, with a fine-tuned ResNet50 reporting 96.2% accuracy and 94.7% sensitivity on CVC-Clinic and Kvasir [35]. EfficientNetB0 with patch-based augmentation achieved an F1-score of 93.1% [29]. Transparency is supported by Class Activation Maps (CAMs), which highlight key regions [33].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section3\"\u003e \u003ch2\u003e6.2.2 Bleeding Detection\u003c/h2\u003e \u003cp\u003eBleeding is visually distinct but prone to false positives from artifacts [134]. While early RGB-based CNNs underperformed, modern models such as DenseNet121 and MobileNetV2 exceed 95% accuracy [31]. CNN\u0026ndash;LSTM hybrids capture temporal continuity across frames [139], and augmentation with red-hue variations further improves generalization [137].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec23\" class=\"Section3\"\u003e \u003ch2\u003e6.2.3 Tumour Detection\u003c/h2\u003e \u003cp\u003eTumours are rare and often resemble folds or polyps [32]. GAN-based data augmentation addresses dataset scarcity, with GAN\u0026ndash;CNN hybrids achieving\u0026thinsp;\u0026gt;\u0026thinsp;91% precision [135]. CNN\u0026ndash;Transformer systems, such as Swin blocks, improved multi-lesion classification to 92.6% [44].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec24\" class=\"Section3\"\u003e \u003ch2\u003e6.2.4 Ulcers and Erosions\u003c/h2\u003e \u003cp\u003eThese lesions are subtle and easily missed under poor lighting. Attention-based CNNs improved ulcer classification with an AUC of 0.963 [43]. A U-Net\u0026thinsp;+\u0026thinsp;CNN cascade reached 90.4% accuracy [34], while multi-label frameworks achieved F1-scores of 91.2% for ulcers and 92.8% for erosions on CAD-CAP [136].\u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e summarizes the performance of state-of-the-art AI models for detecting polyps, bleeding, tumours, and ulcers/erosions in WCE. The results highlight the diversity of architectures, evaluation metrics, and datasets considered in recent studies.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePerformance of different models in the detection of polyps, bleedings, tumours, and erosions.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLesion Type\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eModel(s)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMetric(s)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eDataset(s)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eRef(s)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003ePolyp\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eResNet50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e96.2% Accuracy, 94.7% Sensitivity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eKvasir\u0026thinsp;+\u0026thinsp;CVC-Clinic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e[35]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEfficientNetB0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eF1-score 93.1%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eKvasir\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e[29]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCAM-based CNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eInterpretability (qualitative)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eKvasir\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e[33]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eBleeding\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDenseNet121\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026gt;\u0026thinsp;95% Accuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCustom BleedDB\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e[31]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMobileNetV2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026gt;\u0026thinsp;95% Accuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCustom BleedDB\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e[31]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCNN\u0026ndash;LSTM hybrid\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTemporal continuity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eVideo sequences\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e[139]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAugmentation (red-hue)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eImproved generalization\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCustom DB\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e[137]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTumor\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCNN\u0026thinsp;+\u0026thinsp;Transformer (Swin)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e92.6% Accuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSynthTumor\u0026thinsp;+\u0026thinsp;Kvasir\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e[44]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGAN\u0026ndash;CNN hybrid\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026gt;\u0026thinsp;91% Precision\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSynthetic augmentation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e[135]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eUlcer/Erosion\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAttention-based CNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAUC 0.963\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCAD-CAP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e[43]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCNN\u0026thinsp;+\u0026thinsp;U-Net Cascade\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e90.4% Accuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCAD-CAP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e[34]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMulti-label framework\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eF1: 91.2% (ulcer), 92.8% (erosion)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCAD-CAP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e[136]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAs shown in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, AI methods achieve consistently high performance across different lesion types. Polyp detection benefits from deep CNNs such as ResNet and EfficientNet, while bleeding detection achieves\u0026thinsp;\u0026gt;\u0026thinsp;95% accuracy using DenseNet and lightweight CNNs. Tumour detection remains more challenging, with hybrid Transformer-based models and GAN augmentation addressing data scarcity. For ulcers and erosions, attention-based CNNs and multi-label frameworks yield promising results on CAD-CAP datasets. Overall, hybrid and attention-driven architectures are the most effective strategies for robust lesion detection.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec25\" class=\"Section2\"\u003e \u003ch2\u003e6.3 Real-Time AI and Edge Models in Wireless Capsule Endoscopy\u003c/h2\u003e \u003cp\u003eThe integration of AI in WCE must balance diagnostic accuracy with latency, memory, and energy use. With each exam producing more than 50,000 images, real-time processing reduces physician workload and supports on-board diagnostics.\u003c/p\u003e \u003cdiv id=\"Sec26\" class=\"Section3\"\u003e \u003ch2\u003e6.3.1 YOLO-Based Detection\u003c/h2\u003e \u003cp\u003eThe YOLO family has been widely applied in WCE for real-time lesion detection. YOLOv3 achieved 34 FPS and 93.7% mAP for polyps [134], while lightweight variants such as YOLOv5s and YOLOv4-tiny achieved\u0026thinsp;\u0026gt;\u0026thinsp;91% accuracy for bleeding/ulcers on embedded devices, cutting image transmission by over 80% [35], [34]. Customized anchors and multi-head YOLO models further improve localization and throughput [136].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec27\" class=\"Section3\"\u003e \u003ch2\u003e6.3.2 Transformer Architectures\u003c/h2\u003e \u003cp\u003eTransformers capture long-range dependencies and temporal dynamics in video frames. A Swin Transformer improved AUC by 4.6% over ResNet50 for bleeding/erosion detection [52], while hybrid CNN\u0026ndash;Transformer models such as combined EfficientNet and ViT reached 95.1% accuracy on Kvasir [33]. These methods enhance lesion tracking but remain computationally demanding [135].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec28\" class=\"Section3\"\u003e \u003ch2\u003e6.3.3 Edge-AI and Lightweight CNNs\u003c/h2\u003e \u003cp\u003eOn embedded platforms, compact models such as MobileNetV3 achieve 28 FPS with \u0026lt;\u0026thinsp;40 MB of RAM on the Raspberry Pi [29]. Quantized or pruned CNNs (e.g., 8-bit ResNet18) deliver 4\u0026times; faster inference with minimal accuracy loss [32]. Early-exit cascades reduce computation by discarding easy cases, cutting costs by up to 38% [137]. EfficientNet-lite3 achieved 94.5% bleeding detection with \u0026lt;\u0026thinsp;40 ms latency at 3.2 W [43].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec29\" class=\"Section3\"\u003e \u003ch2\u003e6.3.4 Toward Onboard Diagnostics\u003c/h2\u003e \u003cp\u003ePrototypes employing ASIC-based CNNs achieve real-time lesion detection at \u0026lt;\u0026thinsp;100 mW [44], demonstrating the potential for fully autonomous in-body diagnostics.\u003c/p\u003e \u003cp\u003eIn summary, YOLO, transformers, and lightweight CNNs provide feasible real-time solutions, but hardware, power, and reliability constraints still limit deployment. Section \u003cspan refid=\"Sec30\" class=\"InternalRef\"\u003e6.4\u003c/span\u003e examines these challenges in detail.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec30\" class=\"Section2\"\u003e \u003ch2\u003e6.4 Challenges in AI-Based Lesion Detection in WCE\u003c/h2\u003e \u003cp\u003eDespite major advances in deep learning for WCE, several challenges continue to limit the robustness, reproducibility, and clinical adoption of AI-based lesion detection systems.\u003c/p\u003e \u003cdiv id=\"Sec31\" class=\"Section3\"\u003e \u003ch2\u003e6.4.1 Generalization and Domain Shift\u003c/h2\u003e \u003cp\u003eA key obstacle is poor generalization across datasets. Models trained on Kvasir, for instance, showed accuracy drops of up to 17% when tested on CAD-CAP or hospital-specific data [134]. These declines are linked to differences in imaging hardware, lighting, and annotation protocols. Domain adaptation techniques, such as CycleGAN or unsupervised transfer learning, have been explored, but they often distort lesion characteristics [135]. Stronger generalization will require larger multi-center datasets and domain-invariant feature representations [32].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec32\" class=\"Section3\"\u003e \u003ch2\u003e6.4.2 Annotation Quality and Label Ambiguity\u003c/h2\u003e \u003cp\u003eSupervised learning relies heavily on accurate annotations, which are costly and time-consuming to obtain. A single WCE video may require hours for gastroenterologists to review, and even then, disagreements arise over lesion boundaries, particularly in subtle cases such as small polyps or early-stage ulcers [52]. Crowdsourcing and semi-supervised labelling have been tested, but noisy labels reduce model accuracy. In ulcer detection, for instance, inaccurate annotations lowered CNN performance by up to 9% [35].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec33\" class=\"Section3\"\u003e \u003ch2\u003e6.4.3 Class Imbalance and Rare Lesion Detection\u003c/h2\u003e \u003cp\u003eWCE datasets are highly imbalanced, with normal frames comprising more than 85%, whereas tumours, erosions, and angiectasia account for less than 1% [34]. This imbalance biases models toward the majority classes. Techniques such as focal loss, SMOTE, cost-sensitive training, and multi-head attention have been evaluated [43]. Despite these methods, rare classes remain difficult: for example, angiectasia F1-scores often remain below 70% despite focal loss [31].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec34\" class=\"Section3\"\u003e \u003ch2\u003e6.4.4 Interpretability and Clinical Trust\u003c/h2\u003e \u003cp\u003eThe \u0026ldquo;black box\u0026rdquo; nature of deep learning models limits clinical acceptance. Even accurate predictions may be disregarded by clinicians when explanations are unclear. Methods such as Grad-CAM, SHAP, and saliency maps help visualize decision-making, but they often yield inconsistent results in noisy WCE frames [137]. In some cases, clinicians dismissed correct predictions because the highlighted regions did not align with their expectations [139].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec35\" class=\"Section3\"\u003e \u003ch2\u003e6.4.5 Clinical Integration and Workflow Compatibility\u003c/h2\u003e \u003cp\u003eMany AI systems remain offline and are used retrospectively, limiting their usefulness for real-time decision support. Edge-AI enables in situ predictions but still faces hardware and regulatory hurdles. The absence of standardized metrics and limited regulatory approval\u0026mdash;such as FDA clearance for WCE-specific AI\u0026mdash;further slows adoption [53].\u003c/p\u003e \u003cp\u003eIn summary, while AI has demonstrated remarkable diagnostic accuracy in controlled settings, challenges with dataset generalization, annotation reliability, detection of rare lesions, interpretability, and workflow integration continue to hinder clinical deployment. Overcoming these issues will require not only technical advances but also stronger collaboration among engineers, clinicians, and regulatory bodies to ensure real-world applicability.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec36\" class=\"Section2\"\u003e \u003ch2\u003e6.5 Comparative Analysis of AI, IoT, and Compressed Sensing in WCE\u003c/h2\u003e \u003cp\u003eThe following section provides a comparative analysis of AI, IoT, and compressed sensing in WCE, emphasizing how their integration collectively enables next-generation capsule systems.\u003c/p\u003e \u003cdiv id=\"Sec37\" class=\"Section3\"\u003e \u003ch2\u003e6.5.1 AI \u0026ndash; Diagnosis-Oriented Processing\u003c/h2\u003e \u003cp\u003eDeep learning automates lesion detection and improves accuracy, but requires large, annotated datasets and faces challenges such as domain generalization and interpretability [134], [43], [31], [32].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec38\" class=\"Section3\"\u003e \u003ch2\u003e6.5.2 IoT \u0026ndash; Connectivity-Driven Infrastructure\u003c/h2\u003e \u003cp\u003eThe Internet of Things enables telemetry, cloud-based data integration, and remote feedback within capsule endoscopy systems. This connectivity facilitates efficient data exchange between the capsule, physicians, and monitoring platforms. However, the IoT framework remains constrained by high energy consumption, transmission latency, and signal attenuation across the gastrointestinal (GI) tract. These factors collectively limit the reliability and continuity of wireless communication in WCE applications [52], [30], [33].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec39\" class=\"Section3\"\u003e \u003ch2\u003e6.5.3 CS \u0026ndash; Efficiency-Oriented Transmission\u003c/h2\u003e \u003cp\u003eCompressed Sensing improves transmission efficiency by reducing data volume and minimizing power consumption. It achieves this by exploiting the inherent sparsity of biomedical signals during acquisition and reconstruction. Despite these advantages, CS-based methods remain highly sensitive to noise and often demand substantial computational resources. These limitations can affect the real-time applicability of CS in wireless capsule endoscopy systems [35], [135], [139].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec40\" class=\"Section3\"\u003e \u003ch2\u003e6.5.4 Comparative Overview\u003c/h2\u003e \u003cp\u003eTo better highlight the complementary roles of Artificial Intelligence, the Internet of Things, and Compressed Sensing in next-generation Wireless Capsule Endoscopy, Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e summarizes their key features with respect to functionality, energy consumption, data requirements, clinical relevance, and integration challenges.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparative key features of AI, IoT, and CS in next-generation WCE.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFeature\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAI\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eIoT\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCS\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCore Function\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLesion detection\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTelemetry \u0026amp; connectivity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eData compression\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePower Use\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMedium\u0026ndash;High\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLow\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eData Needs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVery high\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eModerate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLow\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eClinical Impact\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMedium\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIntegration Complexity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eModerate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHigh\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec41\" class=\"Section3\"\u003e \u003ch2\u003e6.5.5 Synergistic Opportunities\u003c/h2\u003e \u003cp\u003eAs shown in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, hybrid designs demonstrate strong potential. For example, AI-guided CS-based region-of-interest sampling reduced bandwidth by 78% without compromising accuracy [35]. When combined with lightweight IoT protocols such as MQTT or LoRa, these approaches enable adaptive, low-power capsules capable of real-time monitoring.\u003c/p\u003e \u003cp\u003eIn summary, the convergence of AI, IoT, and CS enables \u0026ldquo;smart capsules\u0026rdquo; with intelligent detection, efficient data handling, and connected workflows. Section \u003cspan refid=\"Sec42\" class=\"InternalRef\"\u003e6.6\u003c/span\u003e expands on open challenges and future research opportunities.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec42\" class=\"Section2\"\u003e \u003ch2\u003e6.6 Future Research Opportunities and Open Challenges in Smart Capsule Endoscopy\u003c/h2\u003e \u003cp\u003eDespite major progress enabled by AI, IoT, and CS, significant barriers remain to the realization of fully autonomous WCE systems. Future research must address the following priorities:\u003c/p\u003e \u003cdiv id=\"Sec43\" class=\"Section3\"\u003e \u003ch2\u003e6.6.1 Edge-Aware AI for Real-Time Processing\u003c/h2\u003e \u003cp\u003eShifting computation from cloud servers to capsule hardware is essential for real-time triage and adaptive imaging. Ultra-efficient CNNs or TinyML models are being explored, but balancing accuracy and power consumption on constrained hardware remains unresolved [134], [30].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec44\" class=\"Section3\"\u003e \u003ch2\u003e6.6.2 Standardization of Data and Evaluation Protocols\u003c/h2\u003e \u003cp\u003eThe lack of consistent datasets and benchmarks hampers cross-study comparison. Large-scale, clinically validated datasets\u0026mdash;such as CAD-CAP and Kvasir\u0026mdash;are valuable but still limited. Future collections should support multi-label annotations and temporal lesion tracking [52], [136].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec45\" class=\"Section3\"\u003e \u003ch2\u003e6.6.3 Multimodal and Adaptive Imaging Capsules\u003c/h2\u003e \u003cp\u003eExpanding beyond RGB imaging toward hyperspectral, 3D, or biochemical sensing could enable richer diagnosis. AI-triggered frame capture may also conserve energy by focusing on abnormal events [35], [32], [140].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec46\" class=\"Section3\"\u003e \u003ch2\u003e6.6.4 Explainability, Trust, and Ethical AI\u003c/h2\u003e \u003cp\u003eThe black-box nature of AI limits clinical adoption. Approaches such as uncertainty estimation, clinician-in-the-loop training, and bias mitigation are required to ensure transparency, fairness, and regulatory acceptance [29, 33, 34].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec47\" class=\"Section3\"\u003e \u003ch2\u003e6.6.5 Privacy and Federated Learning\u003c/h2\u003e \u003cp\u003eFederated Learning (FL) enables privacy-preserving model training across institutions, aligning with GDPR and HIPAA. However, challenges remain in managing communication overhead and heterogeneous client data [137], [139].\u003c/p\u003e \u003cp\u003eIn summary, addressing efficiency, data quality, multimodal integration, transparency, and privacy will be central to advancing WCE toward intelligent, trustworthy, and widely deployable smart capsules.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec48\" class=\"Section2\"\u003e \u003ch2\u003e6.7 Interim Summary\u003c/h2\u003e \u003cp\u003eThis review summarizes the integration of Artificial Intelligence, Internet of Things, and Compressed Sensing in Wireless Capsule Endoscopy. AI has advanced lesion detection, segmentation, and decision support, often reaching clinician-level accuracy in polyp and ulcer detection [134], [43], [32]. However, limited generalization, annotation cost, and lack of interpretability remain barriers to clinical use. IoT has expanded WCE by enabling wireless data synchronization, real-time monitoring, and remote feedback [33], [34]. Yet it introduces challenges related to energy use, latency, and privacy. CS addresses bandwidth and power constraints through sparsity-based compression, reducing transmission load while retaining diagnostic content [140], [136]. Its limitations lie in its sensitivity to reconstruction and its vulnerability to noise. Looking forward, the convergence of these domains is expected to produce \u0026ldquo;smart capsules\u0026rdquo; capable of adaptive imaging, selective transmission, and seamless integration with electronic health records [52], [35].\u003c/p\u003e \u003cp\u003eKey directions include:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eAI combined with advanced sensing (multi- spectral, biochemical).\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003ePrivacy-preserving AI via federated learning [139].\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eEvent-driven reporting for urgent cases.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eInterpretable AI to strengthen clinician trust [29].\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eIn the long term, these innovations are likely to improve diagnostic speed, accuracy for rare pathologies, and accessibility in low-resource settings. With progress in miniaturization, edge computing, and low-power communication, WCE is poised to become a central tool in next-generation gastrointestinal care.\u003c/p\u003e \u003cp\u003eTo provide a structured perspective on how the three enabling technologies \u0026mdash; Artificial Intelligence (AI), the Internet of Things (IoT), and Compressed Sensing (CS) \u0026mdash; contribute to the evolution of Wireless Capsule Endoscopy (WCE), Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e summarizes their core roles, advantages, challenges, and representative applications. This comparative view highlights the complementary nature of these technologies and the areas requiring further development.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eBenefits, challenges and implications of different technologies in WCE.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTechnology\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCore Role in WCE\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAdvantages\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eChallenges\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eApplications\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eRef\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAI\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAutomated lesion detection \u0026amp; classification\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHigh accuracy, reduced workload\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eGeneralization, explainability, and annotation cost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePolyp detection, bleeding segmentation, tumour classification, edge AI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e[134], [43], [32], [29]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eIoT\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eConnectivity \u0026amp; remote interaction\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eReal-time telemetry, monitoring\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eBandwidth, latency, privacy, energy use\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eCapsule navigation, cloud sync, remote feedback\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e[33], [34], [35]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCS\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eData compression \u0026amp; efficient transmission\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLow energy, reduced bandwidth\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eReconstruction quality, noise sensitivity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSelective ROI transmission, compressed imaging\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e[140], [52], [136]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAs shown in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, AI contributes diagnostic intelligence but is limited by generalization and interpretability; IoT provides connectivity at the cost of energy and privacy; and CS improves efficiency but remains sensitive to noise. Their convergence underscores the complementary strengths of these technologies in advancing next-generation WCE toward smart, connected, and adaptive systems.\u003c/p\u003e \u003c/div\u003e"},{"header":"7. IoT in Wireless Capsule Endoscopy","content":"\u003cp\u003eThe Internet of Things refers to a network of interconnected devices equipped with sensors, software, and communication technologies that enable seamless data exchange over the internet [141]. In healthcare, this concept is extended to the Internet of Medical Things (IoMT), which supports continuous monitoring and real-time analytics, enabling personalized and preventive care. As part of healthcare 4.0, IoMT\u0026mdash;along with edge computing and AI\u0026mdash;has become a central driver of modern clinical systems [142]. Wireless Capsule Endoscopy, a noninvasive technique for gastrointestinal imaging, has benefited greatly from IoT integration. IoT-based telemetry enables real-time imaging and wireless data transfer, allowing remote access to diagnostic information by external devices [123], [143]. This connectivity enhances patient comfort by reducing in-clinic visits and supports extended monitoring outside hospital settings [18]. IoT also enables interaction through real-time remote-control features within WCE, such as device reconfiguration, firmware updates, and data offloading\u0026mdash;functions that are essential for smart capsule development [144]. In addition, multiple sensors and cloud-based synchronization open new opportunities for telemedicine and remote diagnostics [145]. These advances align with the concept of connected healthcare, where patients, devices, and clinicians form feedback loops [142]. Having established the role of IoT in WCE, Section \u003cspan refid=\"Sec50\" class=\"InternalRef\"\u003e7.1\u003c/span\u003e examines the architectural components that enable this advanced connectivity.\u003c/p\u003e \u003cdiv id=\"Sec50\" class=\"Section2\"\u003e \u003ch2\u003e7.1 IoT Architecture in Smart Capsule Systems\u003c/h2\u003e \u003cp\u003eThe architecture of IoT-integrated WCE systems is organized into three core layers: perception, network, and application. At the perception layer, the capsule integrates biosensors and RFID modules to capture pH, temperature, pressure, and imaging data\u0026mdash;critical inputs for diagnostic evaluation [143], [48]. The network layer handles communication between the capsule and external receivers. Low-power wireless modules transfer data securely to gateways, which relay information to cloud or edge platforms. Decentralized edge/fog units reduce latency and energy load, enhancing Quality of Service (quality of service) [142], [26], [36]. At the application layer, clinicians access dashboards and analytics tools that convert sensor data into actionable insights, supporting real-time diagnostic decision-making [123], [17]. Overall, this three-layer architecture provides modularity and interoperability, linking embedded sensing with secure transmission and clinically relevant visualization [142]. The general IoT-enabled architecture for WCE systems is depicted in Figure. 4.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAs shown in the figure. 4, this framework establishes the foundation for real-time monitoring, which is further examined in Section \u003cspan refid=\"Sec51\" class=\"InternalRef\"\u003e7.2\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec51\" class=\"Section2\"\u003e \u003ch2\u003e7.2 Real-Time Monitoring and Data Acquisition\u003c/h2\u003e \u003cp\u003eWireless Capsule Endoscopy enables continuous monitoring of the gastrointestinal tract by collecting image and physiological data as the capsule moves through the digestive system [143]. Embedded sensors\u0026mdash;including imaging modules and instruments for temperature, pH, and pressure\u0026mdash;collect signals at fixed intervals and transmit them wirelessly to an external receiver, typically worn by the patient [123], [48]. Real-time transmission provides physicians with immediate access to physiological changes, improving diagnostic accuracy and supporting timely interventions [145]. Within IoT-enabled frameworks, the data can be relayed to cloud platforms for storage and visualization [146]. Live dashboards allow clinicians to track capsule progression in real time via mobile or desktop interfaces [147]. To improve efficiency, wearable receivers equipped with edge modules can perform basic preprocessing tasks such as image compression and noise filtering before data are uploaded [142], [36]. This approach reduces bandwidth demand and conserves capsule energy by minimizing unnecessary transmissions. Furthermore, these systems can generate alerts for abnormal readings, enabling remote clinical response [148]. The combination of real-time acquisition, embedded sensing, and wireless analytics transforms WCE into a responsive diagnostic platform. By linking raw physiological data with actionable insights, WCE provides a more intelligent approach to gastrointestinal monitoring [18], [149].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec52\" class=\"Section2\"\u003e \u003ch2\u003e7.3 Wireless Communication Protocols and Standards\u003c/h2\u003e \u003cp\u003eReliable wireless communication is essential in WCE to ensure that real-time data from the capsule is transmitted effectively to external receivers and diagnostic platforms [48]. Several low-power protocols\u0026mdash;Bluetooth Low Energy (BLE), ZigBee, LoRa, NB-IoT, and Wi-Fi\u0026mdash;have been explored in this context. Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e summarizes the main wireless communication protocols explored for WCE, highlighting their trade-offs in data rate, range, power consumption, suitability, and limitations [131]\u0026ndash;[133].\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe most well-known protocols for WCE.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eProtocol\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eData Rate\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRange\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePower Use\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSuitability for WCE\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eLimitation\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBLE\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUp to 2 Mbps\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e~\u0026thinsp;10 m\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eVery Low\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eExcellent for short-range, low-power\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eLimited range, interference-prone\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZigBee\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e~\u0026thinsp;250 kbps\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e~\u0026thinsp;100 m\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eVery Low\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eGood for mesh-style in-body networks\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eLow bandwidth\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLoRa\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;50 kbps\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026gt;\u0026thinsp;1 km (outdoor)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSuitable for remote monitoring\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eVery low data rate\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNB-IoT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e~\u0026thinsp;100 kbps\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNationwide\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eModerate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSecure, cloud-connected telemetry\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eHigh power demand\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWi-Fi\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUp to 54 Mbps\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e~\u0026thinsp;50 m\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePrototype experiments\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNot feasible for battery capsules\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAs shown in Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, selecting an appropriate protocol requires balancing trade-offs between range, throughput, energy, and security. In some designs, hybrid approaches are adopted\u0026mdash;for example, BLE for short-range relay combined with NB-IoT for cloud uploads\u0026mdash;providing both real-time monitoring and secure long-distance connectivity [144], [145].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec53\" class=\"Section2\"\u003e \u003ch2\u003e7.4 Edge Computing vs. Cloud Computing in WCE\u003c/h2\u003e \u003cp\u003eCloud computing provides extensive storage and computational resources, making it well-suited to handling the large data volumes generated by WCE [141]. Capsule sensor outputs can be transmitted to cloud platforms, where advanced algorithms\u0026mdash;including AI-based diagnostic models\u0026mdash;interpret the data [26]. This approach also supports remote collaboration among clinicians and seamless integration with Electronic Health Records (EHRs) [142]. However, transmitting all raw data to the cloud raises major concerns. It consumes high bandwidth and introduces privacy risks, as sensitive medical information must be protected under strict regulations [150]. To mitigate these limitations, edge computing has emerged as a complementary approach, moving computational tasks closer to the data source\u0026mdash;such as wearable receivers or local gateways [123]. Edge computing enables near-real-time functions such as image enhancement, anomaly detection, and preliminary pattern recognition before forwarding compressed or summarized data to the cloud [36]. This not only reduces the processing burden on the capsule but also conserves energy by minimizing unnecessary transmissions [143]. While the latency advantages of edge systems were discussed in Section \u003cspan refid=\"Sec52\" class=\"InternalRef\"\u003e7.3\u003c/span\u003e, this section focuses on computational offloading and security benefits. Hybrid architectures combine both paradigms: edge nodes manage time-sensitive tasks, while cloud servers perform resource-intensive analytics and provide long-term storage [17]. This model improves scalability, reliability, and adaptability to diverse clinical contexts [144]. Ultimately, the choice among edge, cloud, and hybrid frameworks depends on application-specific requirements, including energy constraints, bandwidth availability, and data security requirements [48]. Processing distribution decisions directly affects both the sustainability and the trustworthiness of WCE platforms. Section \u003cspan refid=\"Sec54\" class=\"InternalRef\"\u003e7.5\u003c/span\u003e further examines energy-consumption challenges and optimization strategies.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec54\" class=\"Section2\"\u003e \u003ch2\u003e7.5 Energy Efficiency and Power Management via IoT\u003c/h2\u003e \u003cp\u003eEnergy consumption is a critical design challenge in WCE, as capsules must operate autonomously inside the gastrointestinal tract for extended periods [143]. IoT-based architectures enhance power efficiency by dynamically adjusting parameters such as sampling frequency and transmission rate in response to real-time physiological signals [123]. For instance, wake-up mechanisms can trigger imaging or telemetry only when motion or temperature thresholds are exceeded [147]. Low-power communication standards such as BLE and ZigBee further reduce energy usage [26]. Hardware design also plays a major role: ultra-low-power microcontrollers and CMOS image sensors help extend battery life while preserving diagnostic quality [26]. When coupled with edge computing, these strategies have demonstrated more than 30% power reduction relative to traditional architectures [36]. On the software side, adaptive compression, event-driven activation, and smart transmission scheduling are increasingly adopted [150]. Energy-aware routing and sleep-mode algorithms within IoT frameworks also lower power consumption in capsule-like medical devices [151\u0026ndash;154]. Beyond conservation, researchers are exploring energy-harvesting methods\u0026mdash;including thermoelectric and biomechanical approaches\u0026mdash;to enable semi-autonomous or self-powered capsules [48]. These technologies could extend the operational lifetime and enable richer, higher-resolution imaging without enlarging the capsule size [17]. While power efficiency is essential for capsule reliability, it must evolve in tandem with robust data protection and privacy safeguards.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec55\" class=\"Section2\"\u003e \u003ch2\u003e7.6 Security and Privacy Concerns in IoT-Enabled WCE\u003c/h2\u003e \u003cp\u003eThe integration of IoT into WCE introduces cybersecurity and privacy risks, mainly due to the wireless transfer of sensitive medical data [155],[156],[7]. As information moves across system layers\u0026mdash;from embedded sensors to gateways and finally cloud platforms\u0026mdash;it becomes vulnerable to interception, manipulation, and breaches [123]. Studies show that many existing protocols still lack sufficient encryption for healthcare applications [150]. A multi-layered security strategy is therefore essential. Strong encryption methods, such as AES, secure device authentication, and blockchain-based audit trails, can enable tamper-resistant logging [157]. End-to-end encryption between the capsule and cloud platforms is increasingly viewed as a baseline requirement for IoMT healthcare [26]. However, implementing advanced cryptography is technically challenging due to the capsule's limited computing power and energy constraints [48]. Access control represents another key concern. Without safeguards, unauthorized users could access diagnostic reports or alter transmitted data [17]. Modern systems address this by employing role-based access control, biometric verification, and secure dashboards [144]. Firmware-Over-The-Air (FOTA) updates are also critical for maintaining system integrity and patching vulnerabilities [18].\u003c/p\u003e \u003cp\u003eLegal and ethical compliance further underpins system security. Regulations such as HIPAA and GDPR ensure that patient data is handled responsibly and lawfully [142].\u003c/p\u003e \u003cp\u003eEffective cybersecurity must therefore be integrated into the design of WCE systems, balancing technical safeguards with regulatory and ethical accountability [36].\u003c/p\u003e \u003cp\u003eWith these foundations in place, the next step is to review IoT-enabled prototypes that incorporate such security features into practical WCE deployments. Section \u003cspan refid=\"Sec56\" class=\"InternalRef\"\u003e7.7\u003c/span\u003e explores these case studies.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec56\" class=\"Section2\"\u003e \u003ch2\u003e7.7 Case Studies and Recent Research in IoT-Enabled WCE\u003c/h2\u003e \u003cp\u003eRecent research has advanced the integration of IoT into WCE systems. One notable effort designed a smart capsule using BLE to stream both image and physiological data to mobile applications, enabling real-time ambulatory monitoring [123]. Another project developed a multi-sensor capsule with pH, temperature, and pressure sensors, transmitting data to the cloud via a local gateway. This approach supported remote diagnostics and reduced reliance on prolonged hospital stays [147], [18]. A 2021 study proposed an IoT-based WCE platform in which onboard machine-learning algorithms performed local anomaly detection and transmitted only flagged data to the cloud. This design reduced transmission volume and conserved battery life [36]. In a separate experiment, real-time localization was implemented using embedded inertial sensors and RF triangulation, managed through an IoT framework to improve capsule tracking [48]. Blockchain has also been explored to secure IoT-enabled WCE systems. Decentralized audit trails and secure access logs were integrated to enhance data privacy [157]. Another implementation employed NB-IoT over public cellular networks to transmit encrypted diagnostic data directly to the cloud [158],[159]. Hybrid architectures are also gaining traction. Edge units handle preprocessing, preliminary classification, and alert generation locally, while cloud resources provide long-term storage, historical trend analysis, and advanced diagnostics [142]. Together, these prototypes illustrate the convergence of IoT, edge intelligence, and real-time medical imaging in next-generation endoscopic diagnostics [26]. Despite these promising implementations, several technical and operational challenges persist, as outlined below.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec57\" class=\"Section2\"\u003e \u003ch2\u003e7.8 Challenges and Future Perspectives\u003c/h2\u003e \u003cp\u003eDespite significant progress in IoT-enabled WCE, several technical and systemic challenges remain. Power limitation is a primary constraint, as capsules must function for extended periods without external energy sources [123]. Wireless transmission also faces reliability issues in the gastrointestinal environment, where tissue attenuation and interference are particularly pronounced in deeper regions [143], [147]. A major design trade-off exists between ensuring high data fidelity and conserving energy. Capturing and transmitting high-resolution images increases both power demand and memory requirements, which conflicts with the capsule\u0026rsquo;s low-power hardware [48]. Secure and private data transfer across sensors, networks, and cloud layers is also difficult to guarantee, given the capsule\u0026rsquo;s limited processing capacity [160],[16]. System-level barriers further complicate deployment. Interoperability challenges arise from diverse communication standards, inconsistent data formats, and fragmented IoT infrastructures [142]. To address this, researchers are pursuing standardized data ontologies and flexible middleware designed for medical IoT applications [26]. Future technologies offer promising solutions. Next-generation wireless systems, such as 6G, are expected to deliver ultra-reliable, low-latency communication (see Section \u003cspan refid=\"Sec52\" class=\"InternalRef\"\u003e7.3\u003c/span\u003e) and higher spectral efficiency, thereby improving real-time telemetry in WCE [17]. Advances in AI are enabling autonomous IoT frameworks that can adjust acquisition strategies, perform onboard anomaly detection, and tailor diagnostics [36]. Energy autonomy is another frontier. Self-powered capsules that use biocompatible energy harvesters (e.g., thermoelectric or kinetic converters) are being investigated to extend operational life without increasing device size [150]. Looking ahead, the fusion of 6G wireless, edge AI, and biomedical IoT is poised to drive a new generation of intelligent diagnostic capsules [157]. This evolution marks a transition from passive data collection to proactive, connected, and personalized gastrointestinal diagnostics. Although challenges in energy efficiency, security, and standardization persist, the convergence of these technologies provides a clear pathway toward fully autonomous, secure, and clinically valuable WCE systems.\u003c/p\u003e \u003c/div\u003e"},{"header":"8. Compressed Sensing","content":"\u003cp\u003eCompressed Sensing has emerged as a disruptive paradigm for signal acquisition and image compression, enabling efficient sampling and accurate reconstruction compared to conventional methods [161]. Its key elements include sparsity-driven acquisition strategies, advanced reconstruction algorithms, and specialized analog-to-digital converters [40]. In medical imaging, particularly MRI, CS integrated with parallel imaging reduces scan duration while preserving diagnostic quality [162]. In wireless biomedical systems, CS mitigates power and bandwidth constraints by compressing signals before transmission, thereby enabling real-time monitoring on resource-limited platforms [163]. Applications also extend to photoacoustic imaging, where CS achieves high-fidelity reconstructions at reduced sampling rates [23]. Recent advances combine compressed sensing with artificial intelligence, enabling diagnostic-quality imaging even under aggressive compression [164]. Hybrid strategies, such as CS-2FFT, have further enhanced signal characterization in capsule endoscopy [165]. These developments highlight CS as a foundation for next-generation WCE; however, most approaches remain restricted to research prototypes and require further validation for clinical integration.\u003c/p\u003e \u003cdiv id=\"Sec59\" class=\"Section2\"\u003e \u003ch2\u003e8.1 Sparsity and Incoherence in Biomedical Signals\u003c/h2\u003e \u003cp\u003eSparsity is fundamental for the efficient acquisition and transmission of biomedical signals. EEG- and ECG-based systems demonstrate that physiological data can often be represented with only a few dominant coefficients in wavelet or Fourier domains [166], [41]. Multi-layer convolutional sparse coding has further advanced compression performance in biomedical imaging [10]. MRI benefits greatly from sparsity-exploiting strategies, where reduced sampling shortens acquisition times without compromising diagnostic quality [162], [163]. Similar principles apply in wearable and implantable systems, where exploiting signal sparsity accelerates both transmission and downstream processing [161], [23]. A cornerstone of Compressed Sensing (CS) is incoherence\u0026mdash;the sensing matrix must remain uncorrelated with the sparsifying basis to ensure accurate reconstruction from incomplete data [167]. Biomedical signals often meet these requirements naturally due to periodicity or structured patterns. To further enhance fidelity, techniques such as dictionary learning and transform-based sparsity models are increasingly adopted in clinical applications [42]. Having established the role of sparsity and incoherence, Section \u003cspan refid=\"Sec60\" class=\"InternalRef\"\u003e8.2\u003c/span\u003e explores the application of CS principles to image compression in WCE systems.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec60\" class=\"Section2\"\u003e \u003ch2\u003e8.2 Compressed Sensing for Image Data Compression in Wireless Capsule Endoscopy\u003c/h2\u003e \u003cp\u003eCS allows image acquisition and reconstruction with far fewer samples than conventional Nyquist-based methods, a major advantage for WCE where energy, memory, and bandwidth are highly constrained [161], [41]. By reducing the amount of data transmitted, CS improves efficiency and enables near-real-time gastrointestinal monitoring. Recent developments integrate CS with deep learning-based reconstruction frameworks, achieving high-resolution images with low power consumption and confirming feasibility for capsule-based diagnostics [164], [168]. CS has also shown strong performance in biomedical photoacoustic imaging, further highlighting its diagnostic potential [23]. Performance evaluations often rely on metrics like PSNR, with reported values consistently above 33 dB, supporting clinical reliability [169]. Hardware-oriented innovations\u0026mdash;such as modulo CS frameworks for ADCs\u0026mdash;help mitigate dynamic-range limitations [40]. At the same time, the combination of CS with parallel imaging continues to enhance image fidelity and reduce scan times [162]. Building on these compression strategies, Section \u003cspan refid=\"Sec61\" class=\"InternalRef\"\u003e8.3\u003c/span\u003e discusses how CS contributes to localization and tracking in capsule endoscopy.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec61\" class=\"Section2\"\u003e \u003ch2\u003e8.3 Compressed Sensing-Based Localization Techniques in Wireless Capsule Endoscopy\u003c/h2\u003e \u003cp\u003eCompressed sensing has been proposed for WCE to lower on-capsule power by reducing the number of transmitted/measured samples, while maintaining clinically useful reconstruction quality. Recent CS systems for wearable/implantable sensing report energy savings of 50% or more, supporting their feasibility for real-time or near-real-time WCE data reduction [170],[111]. Sparse Bayesian learning and graph-structured models further improve estimation reliability, allowing accurate trajectory recovery with fewer projection measurements [171]. Advanced methods such as compressed trajectory sensing and super-resolution combined with CS enhance angular precision and navigation fidelity [172], [37]. Compressive magnetic mapping has also shown robustness against disturbances, supporting clinical deployment [37]. Incorporating prior motion models or random projection matrices strengthens trajectory estimation under complex anatomical conditions [172], while sparse approximation enables path reconstruction even without GPS or visual markers [165]. Hybrid strategies integrate CS with Kalman filtering or compressive priors for predictive motion modelling [40]. Compressed magnetic signals have been shown to capture both positional and orientational data with minimal sensor input [167]. More recently, deep learning models trained on sparsified sensor data have improved geometry-aware localization, making systems more resilient to noise and patient variability [93].\u003c/p\u003e \u003cp\u003eIn summary, CS-based localization offers precise, energy-aware, and robust solutions for capsule tracking, addressing critical clinical and technical requirements. Section \u003cspan refid=\"Sec62\" class=\"InternalRef\"\u003e8.4\u003c/span\u003e transitions to hardware\u0026ndash;software co-design strategies for practical deployment.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec62\" class=\"Section2\"\u003e \u003ch2\u003e8.4 Hardware and Algorithm Co-Design for Compressed Sensing in Wireless Capsule Endoscopy\u003c/h2\u003e \u003cp\u003eImplementing CS in WCE is constrained by capsule size, limited energy, and low computational capacity. Hardware\u0026ndash;software co-design has therefore become essential. Ultra-low-voltage acquisition modules reduce power consumption while maintaining fidelity [169], and ASICs that embed CS reconstruction engines minimize computational load without compromising signal integrity [47]. FPGAs have been explored to exploit parallelism for real-time reconstructions, achieving sub-millisecond delays with application-specific sparsity models [42], [173]. Due to memory constraints, compressed-domain processing is often adopted, with dictionary learning and custom bases implemented directly in circuits to improve efficiency [174]. Adaptive methods, such as approximate computing and reconfigurable logic, also support dynamic adjustment to variations in resolution and sparsity [93]. At the sensor level, compressive ADCs encode signals via multiplexing, eliminating the need for raw data storage and enabling reconstruction using algorithms such as OMP and CoSaMP [47], [49]. These circuit-level innovations illustrate how CS can be embedded into compact hardware platforms while maintaining diagnostic fidelity. Overall, co-optimized pipelines that integrate CS algorithms with specialized electronics demonstrate strong potential to enable capsules capable of onboard processing and real-time in vivo diagnostics under strict resource constraints [42]. Section \u003cspan refid=\"Sec63\" class=\"InternalRef\"\u003e8.5\u003c/span\u003e next introduces evaluation metrics for assessing CS performance in WCE.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec63\" class=\"Section2\"\u003e \u003ch2\u003e8.5 Evaluation Metrics for Compressed Sensing in WCE\u003c/h2\u003e \u003cp\u003eEvaluating CS in WCE requires examining both reconstruction fidelity and system-level performance. Image quality metrics remain the most widely used: Peak Signal-to-Noise Ratio (PSNR) values above 30 dB typically indicate strong similarity to original frames [175]. In contrast, Structural Similarity Index (SSIM) values exceeding 0.85 in block-based WCE experiments confirm preserved diagnostic reliability [162], [176]. Error-based measures such as Root Mean Square Error (RMSE) and Normalized Mean Square Error (NMSE) provide additional benchmarks, with RMSE\u0026thinsp;\u0026lt;\u0026thinsp;5% generally regarded as clinically acceptable [177].\u003c/p\u003e \u003cp\u003eSystem-level indicators complement image fidelity. Studies have shown that combining CS with energy-aware modulation strategies reduces the per-bit transmission cost, and hybrid CS\u0026ndash;inertial architectures significantly reduce frame-delivery latency [42, 131, 178\u0026ndash;180]. Data compactness is often quantified by the sparsity ratio, with optimized pipelines achieving 10\u0026ndash;20% reductions in the number of coefficients [169]. Beyond quantitative results, clinical validation remains essential. Physician-based evaluations report that more than 85% of CS-reconstructed frames are judged diagnostically equivalent to originals [37]. Taken together, these metrics highlight CS as a robust and clinically reliable method for efficient image acquisition, transmission, and interpretation in WCE.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec64\" class=\"Section2\"\u003e \u003ch2\u003e8.7 Future Trends and Research Directions in Compressed Sensing for WCE\u003c/h2\u003e \u003cp\u003eThe future of CS in WCE is expected to evolve through adaptive and intelligent frameworks that address real-time constraints, energy limitations, and clinical accuracy requirements. A central direction is the shift toward edge-computing capsules, in which compressed-domain AI modules analyze sparsified data directly on board. This approach eliminates the need for full reconstruction or continuous cloud support, thereby reducing latency and energy consumption while enabling closed-loop diagnostic operation [167], [170], [32], [181]. Another emerging trajectory is multimodal integration, in which CS frameworks jointly process imaging, inertial, and physiological signals. Instead of treating each modality separately, next-generation capsules will exploit joint sparsity to achieve efficient data use while preserving diagnostic and localization accuracy [165], [37]. Looking further ahead, federated learning, quantum-inspired CS algorithms, and direct compressed-domain classification represent forward-looking directions. These innovations aim to minimize power use, enhance real-time responsiveness, and ensure robust diagnostic reliability across diverse patient populations.\u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e summarizes representative compressed sensing (CS) strategies in WCE, covering imaging, localization, AI-assisted recovery, hardware-aware implementations, graph-based models, and physiological signal compression. Each method highlights distinct design trade-offs with respect to application, advantages, and limitations.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eCS-based methods for WCE problems and their corresponding characteristics.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCS Method\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eApplication\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAdvantages\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eChallenges\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eReferences\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eImage-based CS (Image-CS)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGI image sequences from WCE\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eReduced transmission load; preserved diagnostic quality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eDegradation at high compression ratios\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e[182], [166], [168]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLocalization-based CS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMagnetic/inertial data\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eEnergy-efficient trajectory estimation; fewer sensors\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSensitive to motion models; noise susceptibility\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e[37], [172], [93]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAI-integrated CS (Deep-CS)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCompressed data with AI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHigher accuracy; learned priors; adaptability\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRequires large datasets; computational demand\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e[183], [184], [161]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHardware-aware CS (HW-CS)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFPGA / ASIC implementations\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eReal-time feasibility; low-power deployment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMemory/frequency limits; HW\u0026ndash;algorithm mismatch\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e[47], [42]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGraph-based CS (Graph-CS)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSpatially structured biomedical data\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCaptures anatomical/structural relations\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRequires graph modelling; higher complexity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e[167], [93]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePhysiological Signal CS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEEG, EMG, temperature, pulse\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMulti-parameter low-data-rate monitoring\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRisk of reduced clinical resolution at low sampling\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e[40], [165]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAs shown in Table\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e, each CS strategy offers distinct advantages and limitations, emphasizing the importance of application-specific optimization. Collectively, these approaches underline CS as a transformative enabler of future WCE platforms by aligning advances in AI, hardware co-design, and multi-modal sensing.\u003c/p\u003e \u003c/div\u003e"},{"header":"9. Overview and Objectives of the Comparative Study","content":"\u003cp\u003eWireless Capsule Endoscopy is a pivotal tool for non-invasive gastrointestinal diagnostics; however, current systems still face challenges, including limited diagnostic accuracy, high energy consumption, and transmission delays [185], [8]. To address these issues, advanced approaches in Artificial Intelligence, Internet of Things, and Compressed Sensing are being increasingly adopted. AI enhances lesion detection by improving sensitivity and specificity for bleeding, polyps, and ulcers, while also reducing the review burden on physicians [186]. IoT enables real-time data transfer and remote monitoring, ensuring timely clinical response [187]. CS mitigates power and bandwidth constraints by enabling signal reconstruction from fewer samples, making it suitable for compact, low-energy devices [188]. These technologies not only provide distinct benefits but also demonstrate potential synergy, forming the backbone of next-generation WCE [189]. The objective of this comparative study is to evaluate their respective strengths, limitations, and clinical relevance in settings such as remote monitoring, Crohn\u0026rsquo;s disease surveillance, and the detection of obscure gastrointestinal bleeding [19]. The evaluation applies key performance indicators\u0026mdash;diagnostic accuracy, energy efficiency, integration feasibility, scalability, and real-time capability\u0026mdash;alongside Technological Readiness Levels (TRLs) and regulatory perspectives [51], [190]. For example, the study contrasts AI-based image classification with CS-driven compression in resource-limited contexts and examines the scalability of IoT-enabled platforms for chronic care [53]. Ultimately, the goal is to guide the selection\u0026mdash;or integration\u0026mdash;of these technologies to support context-specific diagnostic tasks in WCE [45], [191].\u003c/p\u003e \u003cdiv id=\"Sec66\" class=\"Section2\"\u003e \u003ch2\u003e9.1 Evaluation Criteria and Methodology\u003c/h2\u003e \u003cp\u003eTo compare AI, IoT, and CS in WCE, a standardized framework was applied, focusing on diagnostic accuracy, computational demand, energy efficiency, latency, and integration feasibility [19]. For AI, diagnostic sensitivity is central, with advanced deep learning models achieving up to 95% accuracy for bleeding detection [186]. CS is assessed through compression ratio, reconstruction fidelity, and image quality; one framework reported 80% compression with preserved diagnostic value [188]. IoT evaluations emphasize energy use, reliability, and real-time data transmission, with packet loss and uptime as key indicators [187]. The methodology combines literature review, simulations, and clinical validation. Over 120 peer-reviewed studies were analyzed to compare TRL, energy demand, and clinical applicability [51], [190]. AI energy profiling used TensorFlow Lite on embedded processors [45], whereas CS frameworks were evaluated using PSNR and SSIM metrics in MATLAB [46]. The Analytical Hierarchy Process (AHP) balanced these indicators relative to system objectives [192]. Real-world cases, including BLE-enabled IoT capsules in telemedicine [193] and deep-learning polyp detection [8], were incorporated. Security and privacy were also considered, with blockchain-enhanced IoT platforms ensuring secure data transfer [194].\u003c/p\u003e \u003cp\u003eIn summary, the framework defines six evaluation criteria\u0026mdash;diagnostic accuracy, power consumption, real-time capability, integration feasibility, TRL, and data security\u0026mdash;that underpin the subsequent cross-technology analysis. Table\u0026nbsp;\u003cspan refid=\"Tab7\" class=\"InternalRef\"\u003e7\u003c/span\u003e defines the evaluation criteria applied in this comparative analysis, including diagnostic accuracy, power consumption, real-time capability, integration feasibility, technology readiness level (TRL), and security.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab7\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 7\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003edefined criteria for comparative analysis.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCriterion\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDefinition\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDiagnostic Accuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCorrect identification of abnormalities such as bleeding, ulcers, or polyps.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePower Consumption\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAverage operational power requirement, measured in milliwatts (mW).\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eReal-Time Capability\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSystem responsiveness is measured as the latency between sensing and reporting.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIntegration Feasibility\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEase of incorporating the technology into existing WCE hardware designs.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTRL (Technology Readiness Level)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eScale (1\u0026ndash;9) indicating the maturity of a technology for deployment.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSecurity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eProtection of patient data against unauthorized access or tampering.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAs shown in Table\u0026nbsp;\u003cspan refid=\"Tab7\" class=\"InternalRef\"\u003e7\u003c/span\u003e, the evaluation framework incorporates both technical and clinical considerations, ensuring that AI, IoT, and CS are compared consistently with respect to diagnostic reliability, efficiency, integration feasibility, maturity, and data protection.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec67\" class=\"Section2\"\u003e \u003ch2\u003e9.2 Diagnostic Accuracy and Clinical Effectiveness\u003c/h2\u003e \u003cp\u003eThe integration of AI into WCE has markedly improved diagnostic precision. CNN models, such as Ding et al. [8], achieved nearly 100% accuracy for bleeding and normal frame classification. In comparison, other deep learning and ResNet-based methods reported\u0026thinsp;\u0026gt;\u0026thinsp;98% specificity in polyp detection and robust multi-lesion performance [186], [189].\u003c/p\u003e \u003cp\u003eThese results demonstrate AI\u0026rsquo;s ability to enhance reliability, reduce false positives, and minimize physician workload. IoT technologies, although they do not directly improve classification accuracy, enable continuous, real-time monitoring and rapid response. Prototypes have transmitted diagnostic data to mobile devices for emergencies [193], while BLE-enabled capsules maintained uninterrupted high-resolution streaming with \u0026lt;\u0026thinsp;5% packet loss over short ranges [187]. These advances highlight IoT\u0026rsquo;s role in timely diagnosis and expanded accessibility, particularly in remote settings. CS ensures image quality under strict resource limits. Studies reported\u0026thinsp;\u0026gt;\u0026thinsp;94% fidelity retention at 70% undersampling [46], and diagnostic adequacy was validated through anonymous reviews [188]. CS-based acquisition also accelerated capsule transit without reducing accuracy [51].\u003c/p\u003e \u003cp\u003eIn summary, AI delivers state-of-the-art diagnostic accuracy, IoT enables responsive real-time connectivity, and CS optimizes image acquisition under energy and bandwidth constraints. Together, they illustrate the importance of selecting or combining technologies according to clinical needs [19].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec68\" class=\"Section2\"\u003e \u003ch2\u003e9.3 System-Level Constraints and Hardware Feasibility\u003c/h2\u003e \u003cp\u003eWireless Capsule Endoscopy systems are restricted by capsule dimensions (\u0026lt;\u0026thinsp;1 cm\u0026sup3;), limited energy storage, and thermal challenges. These constraints hinder the integration of advanced AI models, since most deep networks exceed the computational capacity of embedded microcontrollers [195], [186]. To cope, lightweight approaches such as TinyML or quantized CNNs have been adopted, though often at the cost of reduced diagnostic accuracy [45]. Energy consumption remains a central bottleneck. Standard 3 V coin-cell batteries support only 8\u0026ndash;12 hours of operation, while data transmission alone can require 50\u0026ndash;100 mW [46]. IoT modules increase demand, as continuous communication requires BLE or Wi-Fi protocols to adopt intermittent or event-driven signalling to conserve energy [193]. Excessive consumption also poses a risk of overheating, which can affect both electronics and surrounding tissue [191]. Compressed sensing provides partial relief by reducing the number of samples and lowering the transmission frequency [188]. However, reconstruction typically must be offloaded to external processors, which requires specialized analog hardware, such as random-sampling circuits and tailored ADCs. These additions complicate miniaturized designs [53]. Cost is another critical factor. BLE-based microcontrollers are relatively inexpensive, whereas AI-oriented accelerators, such as EdgeTPU, are both costly and oversized for capsule integration [147]. AI frameworks tend to increase Bill of Materials (BOM) and energy requirements [185], whereas CS shifts complexity to external computation [188]. IoT module costs vary depending on chipset, protocol, and whether energy harvesting is included [187].\u003c/p\u003e \u003cp\u003eIn summary, each enabling technology faces distinct trade-offs at the system level: AI increases diagnostic power but struggles with size and energy limits; IoT enables real-time connectivity but raises power consumption and cost; CS reduces energy and bandwidth load but relies on complex external reconstruction. Table\u0026nbsp;\u003cspan refid=\"Tab8\" class=\"InternalRef\"\u003e8\u003c/span\u003e summarizes the main system-level trade-offs of AI, IoT, and CS in WCE, highlighting their respective advantages, limitations, and overall impact on capsule feasibility.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab8\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 8\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSystem-level comparison of AI, IoT, and CS in capsule endoscopy.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTech\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAdvantages\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLimitations\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSystem-Level Impact\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAI\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHigh diagnostic accuracy; supports automated lesion detection; potential for real-time decision support\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHigh computational demand; poor fit for capsule microcontrollers; increases BOM cost and power usage [186], [45]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eImproves diagnostic reliability but is constrained by size, energy, and thermal limits\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eIoT\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEnables real-time telemetry; supports remote monitoring and integration with EHRs; event-driven signalling reduces delays [187], [193]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHigh continuous energy demand; risk of overheating; variable cost depending on protocols and chipsets [191]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eEnhances connectivity and accessibility, but drains limited capsule resources\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCS\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eReduces data transmission load; lowers energy and bandwidth usage; offloads computation externally [188]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eReconstruction heavy; requires specialized analog hardware; increases design complexity [53]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eEfficient at managing resource limits, but depends on external processing for clinical usability\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAs shown in Table\u0026nbsp;\u003cspan refid=\"Tab8\" class=\"InternalRef\"\u003e8\u003c/span\u003e, AI offers superior diagnostic accuracy but is constrained by computational and energy demands; IoT enables real-time connectivity but significantly increases power consumption and cost; and CS reduces energy and bandwidth usage but relies heavily on external reconstruction. These trade-offs emphasize the need for hybrid approaches that balance diagnostic performance with the strict physical and energy constraints of capsule systems.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec69\" class=\"Section2\"\u003e \u003ch2\u003e9.4 Data Handling and Transmission Dynamics\u003c/h2\u003e \u003cp\u003eEfficient data handling remains a major challenge in WCE, as a single session may generate 50,000\u0026ndash;60,000 images (several gigabytes) that must be stored or transmitted [185].\u003c/p\u003e \u003cp\u003eThis scale strains bandwidth, memory, and latency. IoT-based capsules use wireless protocols such as BLE and Wi-Fi, achieving rates up to 2 Mbps in line-of-sight conditions [193]. Within the body, however, packet loss can reach 7% [187], jeopardizing diagnostic continuity unless mitigated by error correction or retransmission. Compressed Sensing reduces this burden by reducing the number of samples by up to 80%, thereby lowering payload and latency [188]. Reconstructions with only 25% of data still achieve diagnostic quality (\u0026gt;\u0026thinsp;30 dB PSNR) [46]. By eliminating heavy encoding, CS suits power-limited capsules [53], though reliance on external reconstruction and sparsity assumptions raises concerns for real-time workflows. AI brings different trade-offs. Lightweight models such as MobileNet and SqueezeNet operate on compressed inputs [186] but still produce large intermediate feature maps, thereby increasing memory and communication requirements [45]. Saliency-based filtering addresses this by transmitting only the most relevant frames [19], balancing efficiency with clinical utility.\u003c/p\u003e \u003cp\u003eIn summary, IoT enables real-time connectivity but faces bandwidth constraints; CS reduces transmission load while relying on robust reconstruction; and AI supports intelligent prioritization while increasing computational demand. Hybrid systems\u0026mdash;combining CS for compression, AI for selective transmission, and IoT for real-time links\u0026mdash;emerge as the most promising solution for scalable WCE [195].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec70\" class=\"Section2\"\u003e \u003ch2\u003e9.5 Synergistic Integration and Hybrid Architectures\u003c/h2\u003e \u003cp\u003eRecent developments in Wireless Capsule Endoscopy emphasize the integration of AI, CS, and IoT into hybrid systems that leverage complementary strengths. For example, a capsule integrating AI-based lesion detection, CS-driven compression, and BLE-enabled IoT reduced power consumption by 45% while maintaining\u0026thinsp;\u0026gt;\u0026thinsp;90% diagnostic accuracy [189]. The key advantage lies in task distribution: CS reduces image volume during acquisition, enabling lightweight AI models to run locally or with minimal transmission [188]. At the same time, IoT edge devices handle compressed data without relying heavily on cloud processing [187]. One prototype employed CS-based sampling, onboard CNN classification, and BLE transmission, achieving\u0026thinsp;\u0026lt;\u0026thinsp;300 ms latency and more than 10 hours of operation. AI-driven control logic further optimized transmission modes based on anatomical context [186]. Other studies confirm this trend. In [67], TinyML-optimized CNNs combined with CS achieved a 43% power reduction compared with full-frame streaming while maintaining\u0026thinsp;\u0026gt;\u0026thinsp;90% sensitivity. Another design [193] employed BLE telemetry and localized AI logic to activate modules only upon anomaly detection, conserving energy and supporting real-time alerts. Despite the promise, challenges remain. CS requires analog front-end hardware and specialized ADCs, complicating mass production [192]. Embedding AI inference engines within capsule constraints demands highly optimized, energy-efficient designs [45]. IoT modules must also ensure secure, real-time wireless communication, particularly in cloud-assisted workflows [194]. Overall, hybrid integration offers the most practical path forward. By integrating AI, CS, and IoT, capsules can evolve into adaptive closed-loop diagnostic platforms that detect abnormalities, adjust imaging strategies, and tailor communication in real time [67]. Such systems align with the broader vision of personalized, intelligent, and autonomous gastrointestinal diagnostics [190].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec71\" class=\"Section2\"\u003e \u003ch2\u003e9.6 Clinical Scenarios and Technology Readiness\u003c/h2\u003e \u003cp\u003eEvaluating the deployment of AI, IoT, and CS in WCE requires both clinical context and Technology Readiness Level (TRL) assessment. Among them, AI is the most mature, reaching TRL 7\u0026ndash;8, with clinical trials consistently showing\u0026thinsp;\u0026gt;\u0026thinsp;90% sensitivity for detecting bleeding, ulcers, and polyps [186]. IoT-enabled platforms using BLE or Wi-Fi are at TRL 6\u0026ndash;7, having been validated in prototypes and pilot studies but not yet widely adopted [193]. CS remains at an earlier TRL (4\u0026ndash;5) due to limited in vivo testing and hardware integration challenges, although its efficiency in energy and data reduction is promising [188]. Clinically, AI has demonstrated clear benefits. In a 200-patient study of obscure gastrointestinal bleeding, CNN-based detection reduced localization time by 35% compared with manual review [8]. IoT-enabled capsules have supported chronic disease monitoring, such as Crohn\u0026rsquo;s, by transmitting near real-time alerts during capsule progression [187]. These illustrate the impact of AI and IoT in enhancing precision, responsiveness, and accessibility, particularly for remote or emergency care. CS has demonstrated experimental utility, as in a pilot study of celiac disease in which sparsely sampled images (70% reduction) reconstructed via dictionary learning yielded 82% diagnostic acceptability [189]. However, inconsistent clarity limited physician confidence, underscoring the need for higher reconstruction fidelity. Hybrid AI\u0026ndash;IoT capsules have also proven effective in emergency response; one prototype transmitted AI-based anomaly alerts via BLE with \u0026lt;\u0026thinsp;5-second latency, improving triage speed and diagnostic throughput [193]. Regulatory pathways remain decisive. AI-based imaging requires extensive validation, clinical trials, and certifications (FDA, CE), often delaying adoption [190]. IoT systems must comply with strict privacy and security rules (HIPAA, GDPR). CS, being post-processing oriented, faces fewer regulatory barriers but still depends on clinical validation and physician acceptance.\u003c/p\u003e \u003cp\u003eIn summary, readiness varies: AI is the most advanced\u003c/p\u003e \u003cp\u003eand clinically proven, IoT is rapidly advancing in remote diagnostics, whereas CS remains experimental but shows strong potential for energy-efficient, miniaturized WCE systems.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec72\" class=\"Section2\"\u003e \u003ch2\u003e9.7 Comparative Summary Table and Visualizations\u003c/h2\u003e \u003cp\u003eThis section synthesizes insights from the previous analysis by comparing AI, IoT, and CS across key evaluation criteria, including diagnostic accuracy, power consumption, real-time capability, integration feasibility, security, hardware requirements, and technology readiness level (TRL). Table\u0026nbsp;9 provides a comparative summary of these technologies, highlighting their relative performance and clinical applicability in WCE.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab9\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparative performance metrics of AI, IoT, and CS in wireless capsule endoscopy systems.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDimension\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAI\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eIoT\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCS\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDiagnostic Accuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHigh (90\u0026ndash;95%) [186]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMedium [187]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eModerate\u0026ndash;High [188]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePower Consumption\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHigh (100\u0026ndash;250 mW) [45]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eModerate\u0026ndash;High (50\u0026ndash;100 mW) [193]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLow (20\u0026ndash;50 mW) [188]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eReal-Time Capability\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLimited (offboard preferred) [186]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eStrong (BLE/Wi-Fi) [193]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePost-processed preferred [189]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIntegration Feasibility\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMedium\u0026ndash;Low [195], [147]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHigh [193]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLow\u0026ndash;Medium [53], [192]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSecurity \u0026amp; Privacy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMedium (cloud dependency) [194]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eVariable (BLE vulnerabilities) [194]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHigh (local, compressed data) [188]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHardware Demand\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHigh (EdgeTPU / CNN) [45]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eModerate (MCU\u0026thinsp;+\u0026thinsp;BLE) [147]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLow (sensor\u0026thinsp;+\u0026thinsp;ADC) [53]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTRL\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTRL 7\u0026ndash;8 [186]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTRL 6\u0026ndash;7 [193]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTRL 4\u0026ndash;5 [188]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKey Clinical Use Cases\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOGIB, polyp and bleeding detection [8]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTelemedicine, Crohn\u0026rsquo;s monitoring [187]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSparse imaging, celiac detection [189]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eTo complement the tabular comparison, Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e5\u003c/span\u003e provides a radar visualization that highlights the relative strengths and weaknesses of AI, IoT, and CS across the same evaluation dimensions. As shown in the figure. 5, each technology exhibits unique advantages\u0026mdash;AI excels in diagnostic accuracy, IoT in real-time connectivity, and CS in efficiency. Hybrid architectures that integrate these strengths represent a strategic direction for next-generation WCE platforms.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"10. Discussion","content":"\u003cp\u003eThe structured analysis presented in this review reveals a decisive paradigm shift in wireless capsule endoscopy (WCE) research\u0026mdash;from isolated hardware-centric localization mechanisms toward intelligent, system-level architectures that integrate sensing, computation, and communication. Early-stage developments were predominantly grounded in physics-based localization strategies, including magnetic-field modelling, inverse electromagnetic formulations, and state-space estimation frameworks [18\u0026ndash;20], [28]. These approaches provided mathematical interpretability and deterministic guarantees but were inherently sensitive to anatomical variability, magnetic distortion, and modelling inaccuracies in complex gastrointestinal environments.\u003c/p\u003e \u003cp\u003eIn contrast, recent research trajectories demonstrate a strong migration toward data-driven intelligence. AI-based lesion detection frameworks, particularly convolutional neural networks (CNNs), have substantially redefined diagnostic performance benchmarks. Reported accuracies of up to 95% for bleeding detection [186] and near-perfect classification performance in controlled settings [8] significantly exceed those of traditional feature-engineering pipelines. Furthermore, deep residual architectures have demonstrated\u0026thinsp;\u0026gt;\u0026thinsp;98% specificity in polyp detection tasks [186], [189], confirming the robustness of learned hierarchical feature representations in high-dimensional perceptual domains. These findings collectively indicate that diagnostic intelligence in WCE is increasingly software-defined rather than sensor-limited.\u003c/p\u003e \u003cp\u003eHowever, performance superiority alone does not equate to deployability. As highlighted within the comparative evaluation framework [19], learning-based systems remain heavily dependent on dataset composition, annotation quality, and domain consistency. Model generalization across diverse patient populations, imaging protocols, and device manufacturers remains insufficiently validated. Moreover, the limited interpretability of deep neural architectures raises regulatory and clinical trust concerns, particularly in safety-critical diagnostic workflows. Without standardized cross-institutional validation and explainability-aware model design, large-scale translational integration may remain constrained.\u003c/p\u003e \u003cp\u003eBeyond perception, IoT-enabled architectures expand WCE from a passive imaging modality to an active node within smart healthcare ecosystems. BLE-enabled capsule platforms have demonstrated reliable real-time telemetry with packet loss rates below 5% in short-range environments [187], while telemedicine-oriented IoT infrastructures enable remote supervision and emergency intervention pathways [193]. Nevertheless, these capabilities introduce systemic trade-offs. Embedded AI execution increases computational demand and energy consumption [45], while wireless communication infrastructures introduce vulnerabilities related to data interception and protocol-level security weaknesses [194]. Thus, scalable IoT-WCE integration requires not only communication reliability but also co-optimization of power budgeting, cybersecurity, and embedded hardware design.\u003c/p\u003e \u003cp\u003eCompressed sensing (CS) methodologies provide a complementary pathway by explicitly addressing acquisition and transmission constraints. By modelling signal recovery as a sparsity-constrained inverse problem [39\u0026ndash;41], CS-based frameworks have achieved compression ratios approaching 80% while preserving diagnostic utility [188]. This capacity directly mitigates bandwidth and energy bottlenecks in resource-constrained ingestible platforms. However, reconstruction fidelity under biologically induced noise, motion artifacts, and non-ideal sampling remains a critical limitation, particularly under real-time processing constraints [46]. Importantly, the literature reveals limited exploration of tightly coupled CS\u0026ndash;AI pipelines in which sparse acquisition, adaptive reconstruction, and diagnostic inference operate as a unified, latency-aware system. This represents a substantial opportunity for architectural innovation.\u003c/p\u003e \u003cp\u003eA broader cross-domain insight emerging from this review is the fragmentation of technological maturity. While AI-based diagnostic modules and IoT communication layers have reached moderate-to-high Technology Readiness Levels (TRL 6\u0026ndash;8) [186], [193], CS-based acquisition frameworks remain at comparatively earlier stages of maturity (TRL 4\u0026ndash;5) [188]. More critically, fully integrated end-to-end architectures\u0026mdash;where AI inference, IoT telemetry, and CS-driven data optimization are co-designed under shared energy and regulatory constraints\u0026mdash;are rarely demonstrated in clinically validated environments. This translational gap underscores a systemic disconnect between algorithmic innovation and deployable medical-grade platforms.\u003c/p\u003e \u003cp\u003eFrom a systems-engineering perspective, future progress in smart WCE will depend less on incremental performance gains within individual components and more on holistic architectural convergence. Energy-aware co-design, explainable AI, secure IoT communication layers, and mathematically grounded sparse acquisition must be jointly optimized rather than independently advanced. Only through such integrative frameworks can next-generation smart capsules transition from promising prototypes to clinically scalable, regulatory-compliant, and interoperable healthcare devices.\u003c/p\u003e \u003cdiv id=\"Sec74\" class=\"Section2\"\u003e \u003ch2\u003e10.1 Limitations of This Review\u003c/h2\u003e \u003cp\u003eThis review adopts a structured narrative synthesis approach based on publicly available peer-reviewed studies. Although more than 120 publications were systematically analyzed within a standardized comparative framework [51], [190], the heterogeneity of reported performance metrics, datasets, and evaluation methodologies precluded quantitative meta-analysis. Consequently, conclusions are interpretative and trend-oriented rather than statistically aggregated. Nevertheless, the comparative synthesis provides a coherent systems-level perspective that clarifies technological maturity, integration feasibility, and translational readiness across AI, IoT, and CS domains.\u003c/p\u003e \u003c/div\u003e"},{"header":"11. Future Research Opportunities and Open Challenges","content":"\n\u003ch3\u003e11.1 Summary of Unresolved Issues\u003c/h3\u003e\n\u003cp\u003eDespite significant progress in WCE enabled by AI, the IoT, and CS, several barriers still limit full-scale clinical adoption. For AI, the most pressing concern is limited generalizability. Current models are typically trained on small, curated datasets that do not capture the full diversity of gastrointestinal conditions, which increases the risk of misclassification in rare or atypical cases [35]. Another critical challenge is interpretability: deep learning models often function as black boxes, reducing physician confidence and delaying regulatory approval [144]. IoT integration continues to face three persistent issues: power consumption, latency, and data security. Continuous real-time transmission via BLE or Wi-Fi can rapidly deplete capsule batteries, sometimes within only a few hours [167]. Latency also fluctuates due to patient movement and tissue interference, undermining reliability in time-sensitive scenarios such as bleeding detection [143]. Furthermore, cloud-based data sharing exposes patient information to cybersecurity threats if strong encryption and authentication measures are not implemented [194]. Although CS effectively reduces data acquisition and transmission loads, it also introduces drawbacks. High undersampling levels (e.g., 70% reduction) may distort image structures, limiting lesion detectability in edge cases [93]. At the hardware level, integrating CS remains immature: random-sampling circuits and miniaturized ADCs remain difficult to design within the strict spatial and thermal constraints of capsule devices [53]. Taken together, these unresolved issues emphasize the need for cross-domain strategies that combine algorithmic innovation, power optimization, enhanced security, and hardware\u0026ndash;software co-design. Importantly, no single technology can address all these limitations; future progress will rely on the complementary strengths of AI for intelligent analysis, IoT for continuous connectivity, and CS for efficient data handling.\u003c/p\u003e\n\u003ch3\u003e11.2 Open Technical Challenges\u003c/h3\u003e\n\u003cp\u003eAs WCE systems advance toward greater autonomy and diagnostic precision, several unresolved technical barriers remain. One of the central challenges is the real-time execution of AI models on the capsule\u0026rsquo;s constrained hardware. Although CNNs deliver high diagnostic accuracy, their computational and memory requirements exceed the capabilities of capsule-grade microcontrollers [52]. Efforts to adopt lightweight variants, such as MobileNet or quantized CNNs, have shown promise; however, they still strain the limited power budget and may cause overheating during extended operation [186]. Ensuring safe, energy-efficient real-time inference, therefore, remains a major obstacle [17]. Wireless communication is another critical bottleneck, particularly for latency-sensitive tasks like gastrointestinal bleeding detection. BLE and Wi-Fi modules often experience significant degradation due to tissue attenuation and patient movement, leading to transmission interruptions [169]. Reliable low-latency connectivity without frequent retransmissions or buffering is essential [187]. Moreover, AI-triggered anomaly alerts can introduce sudden power spikes, underscoring the need for intelligent power management [123]. The implementation of CS at the hardware level also remains underdeveloped. While simulations demonstrate strong performance, analog front-end components\u0026mdash;such as pseudo-random samplers, low-noise amplifiers, and CS-aware ADCs\u0026mdash;must be miniaturized to operate within \u0026lt;\u0026thinsp;1 cm\u0026sup3; while remaining reliable in varying biological and electromagnetic environments [53], [23]. Thermal regulation adds another layer of complexity. High-frequency sensing, wireless bursts, and on-chip AI computations generate localized heating that may irritate surrounding tissue or impair sensor performance [191]. Existing strategies, including duty cycling and passive dissipation, may not suffice for future hybrid AI\u0026ndash;CS\u0026ndash;IoT systems operating continuously in real time.\u003c/p\u003e \u003cp\u003eIn summary, the evolution of WCE requires multi-layered innovation in hardware\u0026ndash;software co-design, adaptive communication protocols, energy-aware AI algorithms, and robust thermal management. Addressing these technical bottlenecks is essential to enabling safe, reliable, and intelligent capsule diagnostics.\u003c/p\u003e\n\u003ch3\u003e11.3 Research Opportunities\u003c/h3\u003e\n\u003cp\u003eThe convergence of AI, IoT, and CS in WCE creates extensive research opportunities to overcome current limitations and move toward autonomous, personalized diagnostics. A major direction is the development of lightweight, energy-efficient AI architectures tailored to capsule hardware. Techniques such as TinyML, network pruning, weight quantization, and knowledge distillation have been shown to reduce memory and power consumption while maintaining diagnostic accuracy significantly [33]. Emerging neuromorphic accelerators and spiking neural networks could further improve performance, enabling real-time inference in the micro-Watt regime [45]. Such innovations would enhance onboard lesion detection while reducing dependence on external servers [109]. Hybrid pipelines that combine AI, CS, and IoT also represent a promising avenue. CS can reduce the data volume during acquisition, after which compact AI classifiers running on-chip determine whether full transmission is necessary [188]. This adaptive approach ensures that only diagnostically relevant information is transmitted, conserving energy and minimizing network congestion [171]. These cascading frameworks underpin closed-loop WCE systems, in which sensing, inference, and communication adapt dynamically to evolving diagnostic insights [194]. Privacy and data security remain critical areas for research. Federated learning enables AI models to be trained collaboratively without transferring raw patient data [194]. At the same time, blockchain mechanisms create immutable audit trails that support regulatory compliance, such as HIPAA and GDPR [143]. Another emerging frontier is patient-specific personalization. Capsules may eventually adjust imaging frequency, AI sensitivity thresholds, or transmission modes based on individual patient histories and clinical objectives [32]. For this vision to materialize, AI systems must not only be accurate but also interpretable, adaptable, and ethically aligned with medical practice. Taken together, these research opportunities highlight the potential for next-generation capsule platforms that are intelligent, energy-efficient, context-aware, and secure. However, realizing these advancements will ultimately require rigorous clinical validation and regulatory approval to ensure safety and trustworthiness in real-world deployments.\u003c/p\u003e\n\u003ch3\u003e11.4 Clinical and Regulatory Considerations\u003c/h3\u003e\n\u003cp\u003eThe translation of WCE systems from prototypes to routine clinical use depends not only on technological readiness but also on rigorous validation and regulatory compliance. For AI-powered capsules, agencies such as the FDA and CE require multi-stage clinical trials that demonstrate safety, efficacy, reproducibility, and superiority to existing diagnostic standards [32]. Adaptive learning models further complicate this process, as their evolution after deployment may necessitate continuous monitoring and periodic re-certification. IoT-enabled WCE platforms face additional requirements arising from privacy and data protection laws, such as HIPAA in North America and the GDPR in Europe. Real-time telemetry, remote alerts, and cloud synchronization must operate within secure communication frameworks that incorporate encryption and authentication to prevent unauthorized access [143]. Particularly in remote care applications, where data is transmitted via mobile or cloud networks, interoperability with hospital systems and cybersecurity auditing remain critical deployment concerns [194]. Compressed sensing, while less directly regulated, still demands validation to ensure diagnostic reliability. Even minor reconstruction artifacts can influence physician decision-making, underscoring the need for standardized image-quality benchmarks, such as PSNR and SSIM, combined with clinical scoring methods [167]. Furthermore, transparent documentation of reconstruction workflows and their limitations is vital to support informed consent practices [188]. Another pressing issue is the generalizability of performance across diverse populations and disease types. Many AI models are trained on region-specific datasets, which limits diversity and introduces risks of bias [57]. To mitigate this, large-scale multicenter trials across diverse demographics and pathologies are necessary to build trust in AI-assisted diagnostics [35]. Finally, explainability is increasingly central to both ethical and legal acceptance. Clinicians and regulators require transparency regarding how AI systems reach diagnostic decisions. Tools such as saliency maps, confidence scores, and traceable decision logs are gaining traction, aligning with explainable AI (XAI) initiatives in medicine [45].\u003c/p\u003e \u003cp\u003eIn summary, the clinical deployment of smart WCE systems is not solely a technical achievement but also a regulatory and ethical undertaking. Closing validation gaps, ensuring data security, and embedding explainability into diagnostic pipelines will be essential for safe, scalable, and trustworthy adoption in healthcare practice.\u003c/p\u003e\n\u003ch3\u003e11.5 Future Vision\u003c/h3\u003e\n\u003cp\u003eThe convergence of AI, IoT, and CS in WCE signals a paradigm shift toward personalized, autonomous gastrointestinal diagnostics. The vision extends beyond incremental improvements, aiming for self-directed, context-aware, and therapeutically capable capsules. A central trajectory is the development of autonomous diagnostic capsules that adapt dynamically in real time\u0026mdash;adjusting frame rates, lighting, or transmission protocols according to lesion detection or gastrointestinal motility [67]. Such systems enable closed-loop control, where AI actively governs data acquisition and telemetry strategies based on pathological findings [148]. The synergy between CS and AI offers a promising path to achieving real-time intelligence under strict hardware and energy constraints. By performing sparse sampling at the sensor level and reconstructing only diagnostically relevant features, computational demands are reduced while maintaining high-fidelity classification [167], [188]. This approach could extend battery life beyond 10 hours, enabling coverage of the entire gastrointestinal tract without compromising accuracy [30]. Advances in neuromorphic processors, which mimic brain-inspired, event-driven computation, may enable CNNs and RNNs to operate at micro-Watt energy profiles. Embedding such processors into WCE systems could deliver true edge-level intelligence while meeting thermal and spatial limitations [8]. From the IoT perspective, future capsules are expected to support bidirectional communication and adaptive alerts. This would enable clinicians to query or reprioritize data streams in real time [186]. In resource-limited regions, Low-Power Wide-Area Networks (LPWAN) could facilitate remote diagnostics, expanding access to care [147]. Personalized diagnostic models also represent a major frontier. By incorporating patient-specific factors such as prior medical history, imaging patterns, or microbiome data, WCE platforms could tailor their acquisition and diagnostic routines to individual risk profiles [43]. Looking ahead, capsules may evolve into theranostic systems \u0026mdash;capable of localized drug delivery, thermal ablation, or biosample collection\u0026mdash;supported by breakthroughs in power harvesting, actuation, and miniaturization [195],[51],[177].\u003c/p\u003e \u003cp\u003eIn summary, the future of WCE lies in intelligent autonomy, energy-aware operation, and patient-centred personalization. Achieving this vision requires addressing unresolved issues, technical barriers, and regulatory demands while leveraging AI for accuracy, IoT for connectivity, and CS for efficiency. No single approach is sufficient; only through their integration and sustained interdisciplinary collaboration across engineering, clinical medicine, regulatory science, and ethics can smart capsules evolve into reliable, autonomous, and intelligent diagnostic platforms. The convergence of AI, IoT, and CS is not merely additive but catalytic, accelerating the transition of capsule endoscopy from an experimental technology into routine clinical practice.\u003c/p\u003e"},{"header":"12. Conclusion and Future Outlook","content":"\u003cp\u003eThis review provides a comprehensive, structured synthesis of the transformative roles that AI, IoT, and CS play in enabling the next generation of Wireless Capsule Endoscopy. By systematically examining their respective and combined contributions to lesion detection, telemetry, localization, and intelligent control, we have shown that these technologies offer not only incremental improvements but also the potential for a fundamental shift in how gastrointestinal diseases are diagnosed and managed. AI offers powerful tools for automated image interpretation, and IoT ensures real-time connectivity and remote supervision. At the same time, CS addresses the longstanding issue of data overload under strict bandwidth and energy limitations. Despite significant progress, several core limitations remain. As discussed in the AI section, AI systems continue to struggle with data scarcity and limited generalization across diverse patient populations and imaging conditions. Moreover, due to the capsule's limited space and energy budget, high-performance AI models must often be executed offline, thereby introducing diagnostic latency. Current power sources cannot sustain continuous operation, and wireless power transfer methods are still in early experimental stages. In addition, while compressed sensing offers promising data reduction, it must strike a delicate balance between compression efficiency and diagnostic interpretability. Equally concerning is the absence of active navigation and therapeutic functionalities, which restrict the capsule\u0026rsquo;s use to passive observation rather than actionable clinical intervention. These unresolved challenges underscore the urgent need for innovative architectures that integrate intelligence, connectivity, and efficiency in a balanced manner.\u003c/p\u003e \u003cp\u003eBy organizing the literature around unified methodological, mathematical, and computational frameworks, this review provides a coherent foundation for interpreting recent advances in smart wireless capsule endoscopy. This structured perspective clarifies how diverse AI-, IoT-, and compressive sensing\u0026ndash;based approaches relate to one another and highlights their respective assumptions, constraints, and trade-offs.\u003c/p\u003e \u003cp\u003eLooking ahead, the integration of ultra-low-power AI models, custom-designed embedded processors (e.g., ASICs, FPGAs), and hybrid AI\u0026ndash;CS pipelines may enable on-board real-time intelligence without compromising energy efficiency. New paradigms like neuromorphic computing and bio-inspired sensors may further enhance embedded processing in next-generation capsules. Clinical translation will also require extensive human-centred design practices, multi-site trials, and adherence to emerging medical AI standards. Ethical considerations\u0026mdash;such as patient data privacy, algorithm transparency, and risk mitigation in autonomous decision-making\u0026mdash;must be built into the design pipeline from the outset.\u003c/p\u003e \u003cp\u003eThe advancement of smart capsule endoscopy is inherently interdisciplinary. Effective progress demands tight collaboration among biomedical engineers, gastroenterologists, AI researchers, hardware designers, clinical ethicists, and regulatory authorities. Initiatives to establish interoperable platforms, common benchmarking datasets, and shared validation protocols will play a key role in accelerating innovation while maintaining clinical rigour and safety. With persistent research, regulatory foresight, and collaborative innovation, smart capsule endoscopy has the potential to evolve from a passive imaging modality into a fully integrated diagnostic and therapeutic platform. This evolution promises not just a refinement of existing tools but a profound rethinking of minimally invasive medicine\u0026mdash;where real-time sensing, targeted intervention, and patient-centred design converge in a single swallowable device. Ultimately, the convergence of AI, IoT, and CS provides not only incremental advances but also the foundation for a paradigm shift toward intelligent, connected, and energy-efficient gastrointestinal diagnostics.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003ctable border=\"0\" cellspacing=\"3\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eAbbreviation\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eFull Term\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eAI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eArtificial Intelligence\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eASIC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eApplication-Specific Integrated Circuit\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eBLE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eBluetooth Low Energy\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eCAM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eClass Activation Map\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eCNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eConvolutional Neural Network\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eCS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eCompressed Sensing\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eEMR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eEndoscopic Mucosal Resection\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eESD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eEndoscopic Submucosal Dissection\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eFL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eFederated Learning\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eGAN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eGenerative Adversarial Network\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eGI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eGastrointestinal\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eIoT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eInternet of Things\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eLSTM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eLong Short-Term Memory\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eMEMS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMicroelectromechanical Systems\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eNB-IoT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eNarrowband Internet of Things\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eRadio Frequency\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eSAR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eSpecific Absorption Rate\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eSoC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eSystem-on-Chip\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eUWB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eUltra-Wideband\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eWCE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eWireless Capsule Endoscopy\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eWPT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eWireless Power Transfer\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAvailability of data and materials.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData sharing not applicable to this article as no datasets were generated or analysed during the current study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests.\u003c/strong\u003e The authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding.\u0026nbsp;\u003c/strong\u003eNo funding was received for this study\u003cstrong\u003e.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors’contributions.\u0026nbsp;\u003c/strong\u003eZeinab Javid conceptualized the study, conducted the literature review, structured the analysis, and drafted the manuscript. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eMichel Kadoch supervised the research, contributed to critical revision of the manuscript, and approved the final version. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAll authors read and approved the final manuscript\u003cstrong\u003e.\u003c/strong\u003e\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003e Wang Y, Huang Y, Chase RC, et al (2023) Global Burden of Digestive Diseases: A Systematic Analysis of the Global Burden of Diseases Study, 1990 to 2019. Gastroenterology 165:773\u0026ndash;783.e15. https://doi.org/10.1053/j.gastro.2023.05.050\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Endoscopy C, Size M, Region B, et al (2025) Capsule Endoscopy Market Summary Market Concentration \u0026amp; Characteristics. 2\u0026ndash;11\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Enns RA, Hookey L, Armstrong D, et al (2017) Clinical Practice Guidelines for the Use of Video Capsule Endoscopy. Gastroenterology 152:497\u0026ndash;514. https://doi.org/10.1053/j.gastro.2016.12.032\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Enns C, Galorport C, Ou G, Enns R (2021) Assessment of Capsule Endoscopy Utilizing Capsocam Plus in Patients with Suspected Small Bowel Disease, Including Pilot Study with Remote Access Patients during Pandemic. J Can Assoc Gastroenterol 4:269\u0026ndash;273. https://doi.org/10.1093/jcag/gwaa042\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Khattab Y, Pott PP (2025) Active/robotic capsule endoscopy - A review. Alexandria Eng J 127:431\u0026ndash;451. https://doi.org/10.1016/j.aej.2025.05.032\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Nomura S, Shimura T, Katano T, et al (2020) A multicenter, single-blind randomized controlled trial of endoscopic clipping closure for preventing coagulation syndrome after colorectal endoscopic submucosal dissection. Gastrointest Endosc 91:859\u0026ndash;867.e1. https://doi.org/10.1016/j.gie.2019.11.030\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Ali MA, Tom N, Alsunaydih FN, et al (2024). Recent Advancements in Localization Technologies for Wireless Capsule Endoscopy: A Technical Review. Sensors 15:1\u0026ndash;21. https://doi.org/10.3390/bioengineering12060613\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Dhali A, Kipkorir V, Maity R, et al (2025) Artificial Intelligence\u0026ndash;Assisted Capsule Endoscopy Versus Conventional Capsule Endoscopy for Detection of Small Bowel Lesions: A Systematic Review and Meta-Analysis. J Gastroenterol Hepatol 40:1105\u0026ndash;1118. https://doi.org/10.1111/jgh.16931\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Wang F, Hu D, Sun H, et al (2023) Global, regional, and national burden of digestive diseases: findings from the global burden of disease study 2019. Front Public Heal 11:. https://doi.org/10.3389/fpubh.2023.1202980\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Chen Q, Wang X, Peng J, et al (2025) The burden of digestive diseases in Asian countries and territories from 1990 to 2019: an analysis for the Global Burden of Disease 2019 study. npj Gut Liver 2:1\u0026ndash;12. https://doi.org/10.1038/s44355-025-00019-x\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Chen W, Neely MJ, Mitra U (2008) Energy-Efficient Transmissions With Individual Packet Delay Constraints. IEEE Trans Inf Theory 54:2090\u0026ndash;2109. https://doi.org/10.1109/TIT.2008.920344\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Yung DE, Plevris JN, Koulaouzidis A (2017) Short article: Aspiration of capsule endoscopes: a comprehensive review of the existing literature. Eur J Gastroenterol Hepatol 29:428\u0026ndash;434. https://doi.org/10.1097/MEG.0000000000000821\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Mehmood T, Naeem N, Parveen S (2025) Survey on Integrated Power Optimization with Battery Friendly Algorithms Survey in Wireless Capsule Endoscopy. IJCSNS Int J Comput Sci Netw Secur 25:\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Chen X, Zhang X, Zhang L, et al (2009) A Wireless Capsule Endoscope System With Low-Power Controlling and Processing ASIC. IEEE Trans Biomed Circuits Syst 3:11\u0026ndash;22. https://doi.org/10.1109/TBCAS.2008.2006493\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Bouyaya D, Benierbah S, Khamadja M (2021) An intelligent compression system for wireless capsule endoscopy images. Biomed Signal Process Control 70:102929. https://doi.org/10.1016/j.bspc.2021.102929\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Manolescu VD, AlZu\u0026rsquo;bi H, Secco EL (2024) Review on endoscopic capsules. Explor Digit Heal Technol 346\u0026ndash;359. https://doi.org/10.37349/edht.2024.00033\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Liu Y, Wang B (2025) Advanced applications in chronic disease monitoring using IoT mobile sensing device data, machine learning algorithms and frame theory: a systematic review. Front Public Heal 13:. https://doi.org/10.3389/fpubh.2025.1510456\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Huang K, Qiu H, Deng Y, et al (2024) Capsule Endoscopy Technology: A New Era in Digestive Tract Examination. J Dig Endosc 15:243\u0026ndash;249. https://doi.org/10.1055/s-0044-1800916\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Chen Q, Wang XX, Peng J, et al (2025) Performance evaluation and future prospects of capsule robot localization technology. Front Public Heal 15:1\u0026ndash;31. https://doi.org/10.1080/10095020.2024.2354239\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Piccirelli S, Salvi D, Pugliano CL, et al (2025) Unmet Needs of Artificial Intelligence in Small Bowel Capsule Endoscopy. Diagnostics 15:1\u0026ndash;11. https://doi.org/10.3390/diagnostics15091092\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Wang X, Xu H, Ren Y, et al (2025) Advances in implantable capsule robots for monitoring and treatment of gastrointestinal diseases. Mater Sci Eng R Reports 163:. https://doi.org/10.1016/j.mser.2025.100943\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Rehan M, Al-Bahadly I, Thomas DG, et al (2023) Smart capsules for sensing and sampling the gut: status, challenges and prospects. Gut 73:186\u0026ndash;202. https://doi.org/10.1136/gutjnl-2023-329614\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Wang Y, Chen Y, Zhao Y, Liu S (2024) Compressed Sensing for Biomedical Photoacoustic Imaging: A Review\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Krumb HJ, Mukhopadhyay A (2025) eNCApsulate: NCA for Precision Diagnosis on Capsule Endoscopes. Int J Comput Assist Radiol Surg. https://doi.org/10.1007/s11548-025-03425-x\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Tsuboi A, Oka S, Tanaka S (2024) Capsule endoscopy: clinical insights, challenges, and evolving perspectives in the 21st century. Mini-invasive Surg 8:. https://doi.org/10.20517/2574-1225.2023.94\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Goel S, Guleria K, Panda SN (2025). Comprehensive Review on Routing Protocols for Wireless Body Area Networks. Int J Math Eng Manag Sci 10:797\u0026ndash;855. https://doi.org/10.33889/IJMEMS.2025.10.4.040\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Wei X, Xi P, Chen M, et al (2024). Capsule robots for the monitoring, diagnosis, and treatment of intestinal diseases. Mater Today Bio 29:101294. https://doi.org/10.1016/j.mtbio.2024.101294\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Drzisga D, K\u0026ouml;ppl T, Pohl U, et al (2016) Numerical modelling of compensation mechanisms for peripheral arterial stenoses. Comput Biol Med 70:190\u0026ndash;201. https://doi.org/10.1016/j.compbiomed.2016.01.015\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Kim HE (2023) Enteroscopy versus Video Capsule Endoscopy for Automatic Diagnosis of Small Bowel Disorders\u0026mdash;A Comparative Analysis of Artificial Intelligence Applications. Biomedicines 11:. https://doi.org/10.3390/biomedicines11112991\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Kim HE, Cosa-Linan A, Santhanam N, et al (2022) Transfer learning for medical image classification: a literature review. BMC Med Imaging 22:69. https://doi.org/10.1186/s12880-022-00793-7\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Gupta M, Mishra A (2024) A systematic review of deep learning based image segmentation to detect polyps. Artif Intell Rev 57:7. https://doi.org/10.1007/s10462-023-10621-1\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Sahafi A, Wang Y, Rasmussen CLM, et al (2022) Edge artificial intelligence wireless video capsule endoscopy. Sci Rep 12:13723. https://doi.org/10.1038/s41598-022-17502-7\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Habe TT, Haataja K, Toivanen P (2024) Review of Deep Learning Performance in Wireless Capsule Endoscopy Images for GI Disease Classification. F1000Research 13:201. https://doi.org/10.12688/f1000research.145950.2\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Naskar S, Sharma S, Kuotsu K, et al (2025). The biomedical applications of artificial intelligence: an overview of decades of research. J Drug Target 33:717\u0026ndash;748. https://doi.org/10.1080/1061186X.2024.2448711\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Kim HE, Cosa-Linan A, Maros ME, et al (2021) A review of transfer learning for medical image classification. Int. J. Wirel. Inf. Networks 27:30\u0026ndash;44\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Munir A, Noor K, Shams MU, et al (2025). The Research of Medical Science Review. The Impact of AI TECHNOLOGIES IN MODERN HEALTHCARE : A Critical Analysis of Challenges, OPPORTUNITIES OF FUTURE PROSPECTS. The Research of Medical Science Review. 3:352\u0026ndash;366\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Miao L, Zhang T, Zuo C, et al (2024) A Rapid Localization Method Based on Super Resolution Magnetic Array Information for an Unknown Number of Magnetic Sources. Sensors 24:. https://doi.org/10.3390/s24103226\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Vedaei SS, Wahid KA (2021) A localization method for wireless capsule endoscopy using side wall cameras and IMU sensor. Sci Rep 11:11204. https://doi.org/10.1038/s41598-021-90523-w\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Schulz A, Silva EAB (2009) Compressive Sensing. 1\u0026ndash;41\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Zhu J, Ma J, Liu Z, et al (2025) A Modulo Sampling Hardware Prototype and Reconstruction Algorithms Evaluation. IEEE Trans Instrum Meas 74:. https://doi.org/10.1109/TIM.2025.3551491\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Li B, Wang Y, Zhao J, Shi J (2024) Ultra-Wideband Antennas for Wireless Capsule Endoscope System: A Review. IEEE Open J Antennas Propag 5:241\u0026ndash;255. https://doi.org/10.1109/OJAP.2024.3355217\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Jiang D, Zhu L, Tong S, et al (2023) Photoacoustic imaging plus X: a review. J Biomed Opt 29:1\u0026ndash;27. https://doi.org/10.1117/1.jbo.29.s1.s11513\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Eregowda N, Pruthviraja D (2023) Feature Extraction and Classification Techniques for Wireless Endoscopy Images: A Review. Rev d\u0026rsquo;Intelligence Artif 37:171\u0026ndash;178. https://doi.org/10.18280/ria.370121\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Kim HE (2022) Convolution neural network for the diagnosis of wireless capsule endoscopy: a systematic review and meta-analysis. Surg Endosc 36:16\u0026ndash;31. https://doi.org/10.1007/s00464-021-08689-3\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Yu G, Zhao S (2020) A new feature descriptor for multimodal image registration using phase congruency. Sensors (Switzerland) 20:1\u0026ndash;21. https://doi.org/10.3390/s20185105\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Xu B, Yu C (2025) Wireless, Battery-free, Implantable Inductor-Capacitor Based Sensors. Adv Electron Mater 2500184:1\u0026ndash;14. https://doi.org/10.1002/aelm.202500184\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Han S, Wang N, Guo Y, et al (2021) Application of Sparse Representation in Bioinformatics. Front Genet 12:1\u0026ndash;12. https://doi.org/10.3389/fgene.2021.810875\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Munagandla VB, Pochu S, Nersu SRK, Kathram SR (2024) Real-Time Data Integration for Emergency Response in Healthcare Systems. J AI-Powered Med Innov (International online ISSN 3078\u0026thinsp;\u0026minus;\u0026thinsp;1930) 3:25\u0026ndash;38. https://doi.org/10.60087/japmi.vol.03.issue.01.id.002\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Mart C, Weinreich W, Czernohorsky M, et al (2018) CMOS-compatible pyroelectric applications enabled by doped HfO2 films on deep-trench structures. Eur Solid-State Device Res Conf 2018-Septe:130\u0026ndash;133. https://doi.org/10.1109/ESSDERC.2018.8486864\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Cao Q, Deng R, Pan Y, et al (2024) Robotic wireless capsule endoscopy: recent advances and upcoming technologies. Nat Commun 15:4597. https://doi.org/10.1038/s41467-024-49019-0\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Su C-C, Chou C-K, Mukundan A, et al (2025) Capsule Endoscopy: Current Trends, Technological Advancements, and Future Perspectives in Gastrointestinal Diagnostics. Bioengineering 12:613. https://doi.org/10.3390/bioengineering12060613\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Kiran B, Thomas D, Parakkal R (2018) An Overview of Deep Learning Based Methods for Unsupervised and Semi-Supervised Anomaly Detection in Videos. J Imaging 4:36. https://doi.org/10.3390/jimaging4020036\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Wang X, Hu X, Xu Y, et al (2023) A systematic review on diagnosis and treatment of gastrointestinal diseases by magnetically controlled capsule endoscopy and artificial intelligence. Therap Adv Gastroenterol 16:1\u0026ndash;13. https://doi.org/10.1177/17562848231206991\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Liu L, Towfighian S, Hila A (2015) A Review of Locomotion Systems for Capsule Endoscopy. IEEE Rev Biomed Eng 8:138\u0026ndash;151. https://doi.org/10.1109/RBME.2015.2451031\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Alam MW, Hasan MM, Mohammed SK, et al (2017) Are Current Advances of Compression Algorithms for Capsule Endoscopy Enough? A Technical Review. IEEE Rev Biomed Eng 10:26\u0026ndash;43. https://doi.org/10.1109/RBME.2017.2757013\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e SWAIN P (2010) At a watershed? Technical developments in wireless capsule endoscopy. J Dig Dis 11:259\u0026ndash;265. https://doi.org/10.1111/j.1751-2980.2010.00448.x\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Edet AN, Ben IAO, Imoh U (2020). Past, Present and Future of Robotic Capsule Endoscopy. In: International Journal of Scientific Research and Engineering Development. pp 935\u0026ndash;941\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Carpi F, Kastelein N, Talcott M, Pappone C (2011) Magnetically Controllable Gastrointestinal Steering of Video Capsules. IEEE Trans Biomed Eng 58:231\u0026ndash;234. https://doi.org/10.1109/TBME.2010.2087332\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Pan G, Wang L (2012) Swallowable Wireless Capsule Endoscopy: Progress and Technical Challenges. Gastroenterol Res Pract 2012:1\u0026ndash;9. https://doi.org/10.1155/2012/841691\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Nakamura T, Terano A (2008) Capsule endoscopy: past, present, and future. J Gastroenterol 43:93\u0026ndash;99. https://doi.org/10.1007/s00535-007-2153-6\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Adler SN, Metzger YC (2011) PillCam COLON capsule endoscopy: recent advances and new insights. Therap Adv Gastroenterol 4:265\u0026ndash;268. https://doi.org/10.1177/1756283X11401645\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Dupont PE, Nelson BJ, Goldfarb M, et al (2021) A decade retrospective of medical robotics research from 2010 to 2020. Sci Robot 6:. https://doi.org/10.1126/scirobotics.abi8017\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Hakimian S, Raines D, Reed G, et al (2021) Assessment of Video Capsule Endoscopy in the Management of Acute Gastrointestinal Bleeding During the COVID-19 Pandemic. JAMA Netw Open 4:e2118796. https://doi.org/10.1001/jamanetworkopen.2021.18796\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Iddan G, Meron G, Glukhovsky A, Swain P (2000) Wireless capsule endoscopy. Nature 405:417\u0026ndash;417. https://doi.org/10.1038/35013140\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Ding Z, Shi H, Zhang H, et al (2019) Gastroenterologist-Level Identification of Small-Bowel Diseases and Normal Variants by Capsule Endoscopy Using a Deep-Learning Model. Gastroenterology 157:1044\u0026ndash;1054.e5. https://doi.org/10.1053/j.gastro.2019.06.025\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Luo Y-Y, Pan J, Chen Y-Z, et al (2019) Magnetic Steering of Capsule Endoscopy Improves Small Bowel Capsule Endoscopy Completion Rate. Dig Dis Sci 64:1908\u0026ndash;1915. https://doi.org/10.1007/s10620-019-5479-z\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Cao Q, Deng R, Pan Y, et al (2024) Robotic wireless capsule endoscopy: recent advances and upcoming technologies. Nat Commun 15:1\u0026ndash;21. https://doi.org/10.1038/s41467-024-49019-0\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Swain P (2008) The future of wireless capsule endoscopy. World J Gastroenterol 14:4142. https://doi.org/10.3748/wjg.14.4142\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Ciuti G, Menciassi A, Dario P (2011) Capsule Endoscopy: From Current Achievements to Open Challenges. IEEE Rev Biomed Eng 4:59\u0026ndash;72. https://doi.org/10.1109/RBME.2011.2171182\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Pikul JH, Gang Zhang H, Cho J, et al (2013) High-power lithium ion microbatteries from interdigitated three-dimensional bicontinuous nanoporous electrodes. Nat Commun 4:1732. https://doi.org/10.1038/ncomms2747\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Ma S, Jiang M, Tao P, et al (2018) Temperature effect and thermal impact in lithium-ion batteries: A review. Prog Nat Sci Mater Int 28:653\u0026ndash;666. https://doi.org/10.1016/j.pnsc.2018.11.002\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Mostafalu P, Sonkusale S (2014). Flexible and transparent gastric battery: Energy harvesting from gastric acid for endoscopy application. Biosens Bioelectron 54:292\u0026ndash;296. https://doi.org/10.1016/j.bios.2013.10.040\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Nadeau P, El-Damak D, Glettig D, et al (2017) Prolonged energy harvesting for ingestible devices. Nat Biomed Eng 1:0022. https://doi.org/10.1038/s41551-016-0022\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Sharova AS, Melloni F, Lanzani G, et al (2021) Edible Electronics: The Vision and the Challenge. Adv Mater Technol 6:. https://doi.org/10.1002/admt.202000757\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Ciuti G, Cali\u0026ograve; R, Camboni D, et al (2016) Frontiers of robotic endoscopic capsules: a review. J Micro-Bio Robot 11:1\u0026ndash;18. https://doi.org/10.1007/s12213-016-0087-x\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Ilic IK, Galli V, Lamanna L, et al (2023) An Edible Rechargeable Battery. Adv Mater 35:. https://doi.org/10.1002/adma.202211400\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Wang F, Wu X, Yuan X, et al (2017). Latest advances in supercapacitors: from new electrode materials to novel device designs. Chem Soc Rev 46:6816\u0026ndash;6854. https://doi.org/10.1039/C7CS00205J\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Chen K, Yan L, Sheng Y, et al (2022) An Edible and Nutritive Zinc-Ion Micro-supercapacitor in the Stomach with Ultrahigh Energy Density. ACS Nano 16:15261\u0026ndash;15272. https://doi.org/10.1021/acsnano.2c06656\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Zhang M, Du H, Liu K, et al (2021) Fabrication and applications of cellulose-based nanogenerators. Adv Compos Hybrid Mater 4:865\u0026ndash;884. https://doi.org/10.1007/s42114-021-00312-2\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Labib M, Grabowski L, Br\u0026uuml;sseler C, et al (2022) Toward the Sustainable Production of the Active Pharmaceutical Ingredient Metaraminol. ACS Sustain Chem Eng 10:5117\u0026ndash;5128. https://doi.org/10.1021/acssuschemeng.1c08275\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Yang J, Zhang Y, Cao W, et al (2022). Analysis of voltage vectors and the realization method of torque ripple reduction for BLDCM. IET Power Electron 15:865\u0026ndash;876. https://doi.org/10.1049/pel2.12274\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Jia Q, Song T, Li Y, et al (2019) OAR Dose Distribution Prediction and gEUD-Based Automatic Treatment Planning Optimization for Intensity Modulated Radiotherapy. IEEE Access 7:141426\u0026ndash;141437. https://doi.org/10.1109/ACCESS.2019.2942393\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Gao J, Zhang Z, Yan G (2019). Development of a Capsule Robot for Exploring the Colon. Micromachines 10:456. https://doi.org/10.3390/mi10070456\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Egger J, Tokuda J, Chauvin L, et al (2012) Integration of the OpenIGTLink Network Protocol for image-guided therapy with the medical platform MeVisLab. Int J Med Robot Comput Assist Surg 8:282\u0026ndash;290. https://doi.org/10.1002/rcs.1415\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e H\u0026ouml;\u0026ouml;g CM, Bark L-\u0026Aring;, Arkani J, et al (2012) Capsule Retentions and Incomplete Capsule Endoscopy Examinations: An Analysis of 2300 Examinations. Gastroenterol Res Pract 2012:1\u0026ndash;7. https://doi.org/10.1155/2012/518718\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Lo W, Liu Y, Elhajj IH, et al (2004) Cooperative Teleoperation of a Multirobot System With Force Reflection via Internet. IEEE/ASME Trans Mechatronics 9:661\u0026ndash;670. https://doi.org/10.1109/TMECH.2004.839040\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Chao-Chieh Lan, Kok-Meng Lee. Dynamic model of a compliant link with large deflection and shear deformation. In: Proceedings, 2005 IEEE/ASME International Conference on Advanced Intelligent Mechatronics. IEEE, pp 729\u0026ndash;734\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Johnson MJ, Xin Feng, Johnson LM, et al Robotic Systems that Rehabilitate as well as Motivate: Three Strategies for Motivating Impaired Arm Use. In: The First IEEE/RAS-EMBS International Conference on Biomedical Robotics and Biomechatronics, 2006. BioRob 2006. IEEE, pp 254\u0026ndash;259\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Gosselin B, Faniel L, Sawan M (2006) A high data rate telemetry system for multi-channel biosignal recording. In: 2006 IEEE Biomedical Circuits and Systems Conference. IEEE, pp 170\u0026ndash;173\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e (2012) IEEE Standard for Local and metropolitan area networks - Part 15.6: Wireless Body Area Networks\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Won K, Choi H-J (2011) A novel diversity transmission technique using cooperative relay. In: The 17th Asia Pacific Conference on Communications. IEEE, pp 649\u0026ndash;653\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Sashiyama H, Hamahata Y, Matsuo K, et al (2008) Rectal burn caused by hot-water coffee enema. Gastrointest Endosc 68:1008\u0026ndash;1009. https://doi.org/10.1016/j.gie.2008.04.017\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Ali MA, Tom N, Alsunaydih FN, Yuce MR (2025) Recent Advancements in Localization Technologies for Wireless Capsule Endoscopy: A Technical Review. Sensors 25:1\u0026ndash;31. https://doi.org/10.3390/s25010253\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Taniuchi K, Ohba Y, Fajardo V, et al (2009) IEEE 802.21: Media independent handover: Features, applicability, and realization. IEEE Commun Mag 47:112\u0026ndash;120. https://doi.org/10.1109/MCOM.2009.4752687\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Slawinski PR (2015) Capsule endoscopy of the future: What\u0026rsquo;s on the horizon? World J Gastroenterol 21:10528. https://doi.org/10.3748/wjg.v21.i37.10528\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Campisano F, Gramuglia F, Dawson IR, et al (2017) Gastric Cancer Screening in Low-Income Countries: System Design, Fabrication, and Analysis for an Ultralow-Cost Endoscopy Procedure. IEEE Robot Autom Mag 24:73\u0026ndash;81. https://doi.org/10.1109/MRA.2017.2673852\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Koulaouzidis A (2015) Wireless endoscopy in 2020: Will it still be a capsule? World J Gastroenterol 21:5119. https://doi.org/10.3748/wjg.v21.i17.5119\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Borsdorf A, Raupach R, Flohr T, Hornegger J (2008) Wavelet-Based Noise Reduction in CT-Images Using Correlation Analysis. IEEE Trans Med Imaging 27:1685\u0026ndash;1703. https://doi.org/10.1109/TMI.2008.923983\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Wang C, Luo Z, Liu X, et al (2018) Organic Boundary Location Based on Colour-Texture of Visual Perception in Wireless Capsule Endoscopy Video. J Healthc Eng 2018:1\u0026ndash;11. https://doi.org/10.1155/2018/3090341\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Nister D, Naroditsky O, Bergen J Visual odometry. In: Proceedings of the 2004 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2004. CVPR 2004. IEEE, pp 652\u0026ndash;659\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Maggi N, Arrigo P, Ruggiero C (2013). Toll-like receptor structural determinants: Variability analysis by digital signal processing methods. In: 13th IEEE International Conference on BioInformatics and BioEngineering. IEEE, pp 1\u0026ndash;4\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Wahab H, Mehmood I, Ugail H, et al (2023). Machine learning based small bowel video capsule endoscopy analysis: Challenges and opportunities. Futur Gener Comput Syst 143:191\u0026ndash;214. https://doi.org/10.1016/j.future.2023.01.011\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Trasolini R, Byrne MF (2021) Artificial intelligence and deep learning for small bowel capsule endoscopy. Dig Endosc 33:290\u0026ndash;297. https://doi.org/10.1111/den.13896\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Muruganantham P, Balakrishnan SM (2022). Attention Aware Deep Learning Model for Wireless Capsule Endoscopy Lesion Classification and Localization. J Med Biol Eng 42:157\u0026ndash;168. https://doi.org/10.1007/s40846-022-00686-8\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Ingle SB, Alexander JA (2007). Recurrent obscure gastrointestinal bleeding: time for provocative thinking? Gastroenterol Hepatol (N Y) 3:571\u0026ndash;3\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Li X, Ren J, Jiang H (2017) Experimental Investigation of Endwall Heat Transfer With Film and Impingement Cooling. J Eng Gas Turbines Power 139:. https://doi.org/10.1115/1.4036361\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Cummins G (2021) Smart pills for gastrointestinal diagnostics and therapy. Adv Drug Deliv Rev 177:113931. https://doi.org/10.1016/j.addr.2021.113931\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Kurbanov B (2012) Apoptose und Melanom: Neue therapeutische Zielstrukturen. Aktuelle Derm 38:91\u0026ndash;94. https://doi.org/10.1055/s-0031-1291547\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Choudhury D (2017) Challenges for 5G?The Future of Wireless Communications [From the Guest Editor\u0026rsquo;s Desk]. IEEE Microw Mag 18:16\u0026ndash;16. https://doi.org/10.1109/MMM.2017.2696838\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Koo TH, Tee V, Lee YY, et al (2024) Green Endoscopy and Sustainable Practices: A Scoping Review. J Dig Endosc 15:184\u0026ndash;191. https://doi.org/10.1055/s-0044-1790203\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Intzes I, Meng H, Cosmas J (2020) An Ingenious Design of a High Performance-Low Complexity Image Compressor for Wireless Capsule Endoscopy. Sensors 20:1617. https://doi.org/10.3390/s20061617\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Mimee M, Nadeau P, Hayward A, et al (2018). An ingestible bacterial-electronic system to monitor gastrointestinal health. Science (80- ) 360:915\u0026ndash;918. https://doi.org/10.1126/science.aas9315\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Kalantar-Zadeh K, Berean KJ, Ha N, et al (2018) A human pilot trial of ingestible electronic capsules capable of sensing different gases in the gut. Nat Electron 1:79\u0026ndash;87. https://doi.org/10.1038/s41928-017-0004-x\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Ke Q, Luo W, Yan G, Yang K (2016). Analytical Model and Optimized Design of Power Transmitting Coil for Inductively Coupled Endoscope Robot. IEEE Trans Biomed Eng 63:694\u0026ndash;706. https://doi.org/10.1109/TBME.2015.2469137\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Sekiya N, Oshimoto N, Ebihara K, et al (2023) Wireless Power Transfer System Using High-Quality Factor Superconducting Transmitting Coil for Biomedical Capsule Endoscopy. IEEE Trans Appl Supercond 33:1\u0026ndash;5. https://doi.org/10.1109/TASC.2023.3256346\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Miah MS, Jayathurathnage P, Icheln C, et al (2019) High-Efficiency Wireless Power Transfer System for Capsule Endoscope. In: 2019 13th International Symposium on Medical Information and Communication Technology (ISMICT). IEEE, pp 1\u0026ndash;5\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Sumiyama K, Futakuchi T, Kamba S, et al (2021) Artificial intelligence in endoscopy: Present and future perspectives. Dig Endosc 33:218\u0026ndash;230. https://doi.org/10.1111/den.13837\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Xu Y, Li K, Zhao Z, Meng MQ-H (2021) On Reciprocally Rotating Magnetic Actuation of a Robotic Capsule in Unknown Tubular Environments. IEEE Trans Med Robot Bionics 3:919\u0026ndash;927. https://doi.org/10.1109/TMRB.2021.3123407\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Sliker L, Ciuti G, Rentschler M, Menciassi A (2015) Magnetically driven medical devices: a review. Expert Rev Med Devices 12:737\u0026ndash;752. https://doi.org/10.1586/17434440.2015.1080120\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Wang Z, Guo S, Fu Q, Guo J (2019) Characteristic evaluation of a magnetic-actuated microrobot in a pipe with screw jet motion. Microsyst Technol 25:719\u0026ndash;727. https://doi.org/10.1007/s00542-018-4000-5\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Zhang Y, Fang M, Feng X, et al (2025). Forging trust in AI-assisted disease diagnosis. Life. https://doi.org/10.1016/j.hlife.2025.05.011\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Smith H (2021) Clinical AI: opacity, accountability, responsibility and liability. AI Soc 36:535\u0026ndash;545. https://doi.org/10.1007/s00146-020-01019-6\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Abdigazy A, Arfan M, Lazzi G, et al (2024). End-to-end design of ingestible electronics. Nat Electron 7:102\u0026ndash;118. https://doi.org/10.1038/s41928-024-01122-2\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Xu Y, Li K, Zhao Z, Meng MQ-H (2021) A Novel System for Closed-Loop Simultaneous Magnetic Actuation and Localization of WCE Based on External Sensors and Rotating Actuation. IEEE Trans Autom Sci Eng 18:1640\u0026ndash;1652. https://doi.org/10.1109/TASE.2020.3013954\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Garbay T, Chuquimia O, Pinna A, et al (2019). Distilling the knowledge in CNN for the WCE screening tool. In: 2019 Conference on Design and Architectures for Signal and Image Processing (DASIP). IEEE, pp 19\u0026ndash;22\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Wang Y, Yoo S, Braun J-M, Nadimi ES (2021) A locally-processed light-weight deep neural network for detecting colorectal polyps in wireless capsule endoscopes. J Real-Time Image Process 18:1183\u0026ndash;1194. https://doi.org/10.1007/s11554-021-01126-7\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Oo WM, Linklater JM, Bennell KL, et al (2020) Superb Microvascular Imaging in Low-Grade Inflammation of Knee Osteoarthritis Compared With Power Doppler: Clinical, Radiographic and MRI Relationship. Ultrasound Med Biol 46:566\u0026ndash;574. https://doi.org/10.1016/j.ultrasmedbio.2019.11.017\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Samel NS, Mashimo H (2019) Application of OCT in the Gastrointestinal Tract. Appl Sci 9:2991. https://doi.org/10.3390/app9152991\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Kang CM, Lee S-H, Chung CC (2018) Discrete-Time LPV Observer With Nonlinear Bounded Varying Parameter and Its Application to the Vehicle State Observer. IEEE Trans Ind Electron 65:8768\u0026ndash;8777. https://doi.org/10.1109/TIE.2018.2813961\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Khan SR, Desmulliez MPY (2019) Towards a Miniaturized 3D Receiver WPT System for Capsule Endoscopy. Micromachines 10:545. https://doi.org/10.3390/mi10080545\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Vedaei SS, Wahid KA (2021) A localization method for wireless capsule endoscopy using side wall cameras and IMU sensor. Sci Rep 11:1\u0026ndash;16. https://doi.org/10.1038/s41598-021-90523-w\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Narmatha P, Thangavel V, Vidhya DS (2022) A Hybrid RF and Vision Aware Fusion Scheme for Multi-Sensor Wireless Capsule Endoscopic Localization. Wirel Pers Commun 123:1593\u0026ndash;1624. https://doi.org/10.1007/s11277-021-09205-5\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Vedaei SS, Wahid KA (2022) MagnetOFuse: A Hybrid Tracking Algorithm for Wireless Capsule Endoscopy Within the GI Track. IEEE Trans Instrum Meas 71:1\u0026ndash;11. https://doi.org/10.1109/TIM.2022.3204103\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Rochmawati N, Fatichah C, Amaliah B, et al (2025) Deep Learning-Based Lesion Detection in Endoscopy: A Systematic Literature Review. IEEE Access 13:43532\u0026ndash;43556. https://doi.org/10.1109/ACCESS.2025.3548167\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Rathnamala DS, Rajagopal A, Arunachalam M, Dhanasekaran V (2025) GI Bleeding Detection in WCE Images Using E-ORB and ME-DEEP CAPSNET. Int J Environ Sci 11:147\u0026ndash;160. https://doi.org/10.64252/358k2389\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Zhang Y, Zhang Y, Huang X (2021) Development and Application of Magnetically Controlled Capsule Endoscopy in Detecting Gastric Lesions. Gastroenterol Res Pract 2021:1\u0026ndash;7. https://doi.org/10.1155/2021/2716559\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Giordano A, Romero-Mascarell C, Gonz\u0026aacute;lez-Su\u0026aacute;rez B, Guarner-Argente C (2025) Integration of Artificial Intelligence-Enhanced Capsule Endoscopy in Clinical Practice: A Review of Market-Available Tools for Clinical Practice. Dig Dis Sci. https://doi.org/10.1007/s10620-025-09099-4\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Putra KT, Arrayyan AZ, Hayati N, et al (2024) A Review on the Application of Internet of Medical Things in Wearable Personal Health Monitoring: A Cloud-Edge Artificial Intelligence Approach. IEEE Access 12:21437\u0026ndash;21452. https://doi.org/10.1109/ACCESS.2024.3358827\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Gupta J, Pathak S, Kumar G (2022) Deep Learning (CNN) and Transfer Learning: A Review. J Phys Conf Ser 2273:012029. https://doi.org/10.1088/1742-6596/2273/1/012029\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Cheng Z, Liao B, He Z, et al (2019) Joint Design of the Transmit and Receive Beamforming in MIMO Radar Systems. IEEE Trans Veh Technol 68:7919\u0026ndash;7930. https://doi.org/10.1109/TVT.2019.2927045\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Gubbi J, Buyya R, Marusic S, Palaniswami M (2013) Internet of Things (IoT): A vision, architectural elements, and future directions. Futur Gener Comput Syst 29:1645\u0026ndash;1660. https://doi.org/10.1016/j.future.2013.01.010\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Osama M, Ateya AA, Sayed MS, et al (2023) Internet of Medical Things and Healthcare 4.0: Trends, Requirements, Challenges, and Research Directions. Sensors 23:. https://doi.org/10.3390/s23177435\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Yang LL, Wang XF, Ke YH, et al (2022). Dielectric Patch Antenna Self-Decoupling by Proper Structural Parameters. IEEE Antennas Wirel Propag Lett 21:1447\u0026ndash;1451. https://doi.org/10.1109/LAWP.2022.3171198\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Bollineni C, Sharma M, Hazra A, et al (2025) IoT for Next-Generation Smart Healthcare: A Comprehensive Survey. IEEE Internet Things J PP:1. https://doi.org/10.1109/JIOT.2025.3570188\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Zhao G, Ban Y, Zhang Z, et al (2023) Phase Demodulation Strategy Based on Kalman Filter for Sinusoidal Encoders. IEEE Sens J 23:10625\u0026ndash;10632. https://doi.org/10.1109/JSEN.2023.3264846\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Anderson K (2024) \u0026lsquo;The role of edge computing and hybrid clouds in next-generation healthcare. Univ Cambridge, Cambridge, UK, Tech Rep\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Fontana S, D\u0026rsquo;Agostino S, Paffi A, et al (2024) State of the Art on Advancements in Wireless Capsule Endoscopy Telemetry: A Systematic Approach. IEEE Open J Antennas Propag 5:1282\u0026ndash;1294. https://doi.org/10.1109/OJAP.2024.3409827\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Huhulea EN, Huang L, Eng S, et al (2025) Artificial Intelligence Advancements in Oncology: A Review of Current Trends and Future Directions. Biomedicines 13:1\u0026ndash;18. https://doi.org/10.3390/biomedicines13040951\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Pitt J, Dryzek J, Ober J (2020) Algorithmic reflexive governance for socio-techno-ecological systems. IEEE Technol Soc Mag 39:52\u0026ndash;59. https://doi.org/10.1109/MTS.2020.2991500\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Nan J, Xu L (2022) Designing Interoperable Healthcare Services Based on HL7 FHIR : A Literature Review towards Motivations, Techniques, and Applications Table of Contents\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Mohseni M, Amirghafouri F, Pourghebleh B (2023) CEDAR: A cluster-based energy-aware data aggregation routing protocol in the internet of things using capuchin search algorithm and fuzzy logic. Peer-to-Peer Netw Appl 16:189\u0026ndash;209. https://doi.org/10.1007/s12083-022-01388-3\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Sankar S, Ramasubbareddy S, Luhach AK, et al (2022) NCCLA: new caledonian crow learning algorithm based cluster head selection for Internet of Things in smart cities. J Ambient Intell Humaniz Comput 13:4651\u0026ndash;4661. https://doi.org/10.1007/s12652-021-03503-3\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Srinivasulu M, Shivamurthy G, Venkataramana B (2023) Quality of service aware energy efficient multipath routing protocol for internet of things using hybrid optimization algorithm. Multimed Tools Appl 82:26829\u0026ndash;26858. https://doi.org/10.1007/s11042-022-14285-x\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Lei C (2024) An energy-aware cluster-based routing in the Internet of things using particle swarm optimization algorithm and fuzzy clustering. J Eng Appl Sci 71:135. https://doi.org/10.1186/s44147-024-00464-0\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Wahab H, Mehmood I, Ugail H, et al (2024) Federated Deep Learning for Wireless Capsule Endoscopy Analysis: Enabling Collaboration Across Multiple Data Centers for Robust Learning of Diverse Pathologies. Futur Gener Comput Syst 152:361\u0026ndash;371. https://doi.org/10.1016/j.future.2023.10.007\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Irshad RR, Sohail SS, Hussain S, et al (2023) Towards enhancing security of IoT-Enabled healthcare system. Heliyon 9:e22336. https://doi.org/10.1016/j.heliyon.2023.e22336\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Flores Fern\u0026aacute;ndez A, S\u0026aacute;nchez Morales E, Botsch M, et al (2023) Generation of Correction Data for Autonomous Driving by Means of Machine Learning and On-Board Diagnostics. Sensors 23:1\u0026ndash;28. https://doi.org/10.3390/s23010159\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Saif S, Das P, Biswas S, et al (2024) A secure data transmission framework for IoT enabled healthcare. Heliyon 10:e36269. https://doi.org/10.1016/j.heliyon.2024.e36269\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Li C, Wang J, Wang S, Zhang Y (2024) A review of IoT applications in healthcare. Neurocomputing 565:127017. https://doi.org/10.1016/j.neucom.2023.127017\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Kaur K, Kaur A, Gulzar Y, Gandhi V (2024) Unveiling the core of IoT: comprehensive review on data security challenges and mitigation strategies. Front Comput Sci 6:. https://doi.org/10.3389/fcomp.2024.1420680\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Velho L (2009) Compressive Sensing In Image Compression\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Guermazi A, Omoumi P, Tordjman M, et al (2024) How AI May Transform Musculoskeletal Imaging. Radiology 310:. https://doi.org/10.1148/radiol.230764\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Chen F, Chandrakasan AP, Stojanovi\u0026aelig; VM (2012) Design and analysis of a hardware-efficient compressed sensing architecture for data compression in wireless sensors. IEEE J Solid-State Circuits 47:744\u0026ndash;756. https://doi.org/10.1109/JSSC.2011.2179451\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Ziegelmayer S, Marka AW, Strenzke M, et al (2025) Speed and efficiency: evaluating pulmonary nodule detection with AI-enhanced 3D gradient echo imaging. Eur Radiol 35:2237\u0026ndash;2244. https://doi.org/10.1007/s00330-024-11027-5\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Banitalebi-Dehkordi M, Abouei J, Plataniotis KN (2014) Compressive-sampling-based positioning in wireless body area networks. IEEE J Biomed Heal Informatics 18:335\u0026ndash;344. https://doi.org/10.1109/JBHI.2013.2261997\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Duart X, Quiles E, Suay F, et al (2021) Evaluating the effect of stimuli color and frequency on SSVEP. Sensors (Switzerland) 21:1\u0026ndash;19. https://doi.org/10.3390/s21010117\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Ali HA, Rashed EA, Kudo H (2025) Compressed sensing-based image reconstruction for discrete tomography with sparse view and limited angle geometries. PLoS One 20:1\u0026ndash;26. https://doi.org/10.1371/journal.pone.0327666\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Ozgurun B, Lafci B, Razansky D (2025) Sparse optoacoustic sensing with convolutional dictionary learning. XX:1\u0026ndash;11. https://doi.org/10.1109/TBME.2025.3589329\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Park Y, Jeon B (2020) An acquisition method for visible and near infrared images from single CMYG color filter array-based sensor. Sensors (Switzerland) 20:1\u0026ndash;18. https://doi.org/10.3390/s20195578\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Saputra OD, Murti FW, Irfan M, et al (2018) Reducing power consumption of wireless capsule endoscopy utilizing compressive sensing under channel constraint. J Inf Commun Converg Eng 16:130\u0026ndash;134. https://doi.org/10.6109/jicce.2018.16.2.130\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Krishnan S, Abdel-Hafez M, H\u0026auml;m\u0026auml;l\u0026auml;inen M (2025) Location estimation of UWB-based wireless capsule endoscopy using TDoA in various gastrointestinal simulation models. PLoS One 20:1\u0026ndash;32. https://doi.org/10.1371/journal.pone.0319167\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Suveren M, Akay R, Kanaan M (2022) Localization of an ultra wide band wireless endoscopy capsule inside the human body using received signal strength and centroid algorithm. Int J Optim Control Theor Appl 12:151\u0026ndash;159. https://doi.org/10.11121/ijocta.2022.1146\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Stoica BA, Sahoo SK, Larus JR, Adve VS (2019) Wok: Statistical program slicing in production. Proc\u0026thinsp;\u0026minus;\u0026thinsp;2019 IEEE/ACM 41st Int Conf Softw Eng Companion, ICSE-Companion 2019 324\u0026ndash;325. https://doi.org/10.1109/ICSE-Companion.2019.00136\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Drakoulogkonas P, Apostolou D (2021) On the selection of process mining tools. Electron 10:1\u0026ndash;24. https://doi.org/10.3390/electronics10040451\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Yao Y, Tan J, Wu J, Zhang X (2022) A Unified Fuzzy Control Approach for Stochastic High-Order Nonlinear Systems With or Without State Constraints. IEEE Trans Fuzzy Syst 30:4530\u0026ndash;4540. https://doi.org/10.1109/TFUZZ.2022.3155297\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Liu G, Zhao P, Zhao M, et al (2020). Electromagnetic disturbed mechanism of electronic current transformer acquisition card under high frequency electromagnetic interference. Electron 9:1\u0026ndash;18. https://doi.org/10.3390/electronics9081293\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Salman D (2025). Optimized Image Compression Using Sparse Representations and Fourier Transform. East J Eng 1:62\u0026ndash;75. https://doi.org/10.63496/eje.vol1.iss1.36\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Razzaque M, Dobson S (2014) Energy-Efficient Sensing in Wireless Sensor Networks Using Compressed Sensing. Sensors 14:2822\u0026ndash;2859. https://doi.org/10.3390/s140202822\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Miao M, Ye C, Xu Z, et al (2025). Optimization Scheme for Modulation of Data Transmission Module in Endoscopic Capsule. Sensors 25:4738. https://doi.org/10.3390/s25154738\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Barrett HH, Myers KJ, Hoeschen C, et al (2015). Task-based measures of image quality and their relation to radiation dose and patient risk. Phys Med Biol 60:R1\u0026ndash;R75. https://doi.org/10.1088/0031-9155/60/2/R1\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Ali B H B, Ramachandran P (2022) Compressive Domain Deep CNN for Image Classification and Performance Improvement Using Genetic Algorithm-Based Sensing Mask Learning. Appl Sci 12:6881. https://doi.org/10.3390/app12146881\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Anand H, Mathur S (2019) Sustainable Communication Networks and Application\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Li Y, Li C, Tian M, et al (2020) Two-Step Thresholds TBD Algorithm for Time Sensitive Target Based on Dynamic Programming. IEEE Access 8:209267\u0026ndash;209277. https://doi.org/10.1109/ACCESS.2020.3038190\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Zanetti F, Bergamaschi L (2020). Scalable block preconditioners for linearized Navier-Stokes equations at high Reynolds number. Algorithms 13:1\u0026ndash;27. https://doi.org/10.3390/A13080199\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e O\u0026rsquo;Hara FJ, Mc Namara D (2023) Capsule endoscopy with artificial intelligence-assisted technology: Real-world usage of a validated AI model for capsule image review. Endosc Int Open 11:E970\u0026ndash;E975. https://doi.org/10.1055/a-2161-1816\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e El-Gammal EM, El-Shafai W, Taha TE, et al (2025) A survey of artificial intelligence models for wireless capsule endoscopy videos for superior automatic diagnosis: problems and solutions. Springer US\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Jameil AK, Al-Raweshidy H (2025) A digital twin framework for real-time healthcare monitoring: leveraging AI and secure systems for enhanced patient outcomes. Discover Internet Things 5: https://doi.org/10.1007/s43926-025-00135-3\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Monika R, Dhanalakshmi S, Rajamanickam N, et al (2024) Coefficient-Shuffled Variable Block Compressed Sensing for Medical Image Compression in Telemedicine Systems. Bioengineering 11:1\u0026ndash;15. https://doi.org/10.3390/bioengineering11111101\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Chen X, Huang L, Ren L (2025) Multitasking Smart Intestinal Capsule Robot : A Cutting-Edge Platform for Sampling, Diagnosis, and Therapy. 1\u0026ndash;6. https://doi.org/10.1002/adrr.202500056\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e McDermott O, Stam DL, Duarte S, Sony M (2025) Readiness for Industry 4.0 in a Medical Device Manufacturer as an Enabler for Sustainability, a Case Study. Sustain 17:1\u0026ndash;20. https://doi.org/10.3390/su17010357\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Hajjar A El, Rey JF (2020) Artificial intelligence in gastrointestinal endoscopy: General overview. Chin Med J (Engl) 133:326\u0026ndash;334. https://doi.org/10.1097/CM9.0000000000000623\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Hanscom M, Cave DR (2022) Endoscopic capsule robot-based diagnosis, navigation and localization in the gastrointestinal tract. Front Robot AI 9:1\u0026ndash;16. https://doi.org/10.3389/frobt.2022.896028\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e McCausland C, Biglarbeigi P, Bond R, et al (2022) Time-Frequency Ridge Analysis of Sleep Stage Transitions. In: 2022 IEEE Signal Processing in Medicine and Biology Symposium (SPMB). IEEE, pp 1\u0026ndash;5\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Ren W, Dimarogonas D V. (2020) Symbolic Abstractions for Periodic Event-triggered Linear Control Systems. In: 2020 European Control Conference (ECC). IEEE, pp 2062\u0026ndash;2067\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e Chen J, Xia K, Zhang Z, et al (2024) Establishing an AI model and application for automated capsule endoscopy recognition based on convolutional neural networks (with video). BMC Gastroenterol 24:. https://doi.org/10.1186/s12876-024-03482-7\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"eurasip-journal-on-wireless-communications-and-networking","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"jwcn","sideBox":"Learn more about [EURASIP Journal on Wireless Communications and Networking](http://jwcn-eurasipjournals.springeropen.com)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/jwcn/default.aspx","title":"EURASIP Journal on Wireless Communications and Networking","twitterHandle":"@SpringerEng","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Wireless Capsule Endoscopy, Artificial Intelligence, Internet of Things, Compressed Sensing, Indoor Localization","lastPublishedDoi":"10.21203/rs.3.rs-8896919/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8896919/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eWireless Capsule Endoscopy (WCE) has emerged as a minimally invasive diagnostic modality for gastrointestinal disorders, offering distinct advantages over conventional endoscopy. Despite clinical adoption, current WCE systems remain limited in energy efficiency, data transmission, lesion-detection accuracy, and reliable localization. These constraints hinder its full potential as a smart, autonomous diagnostic platform. This review synthesizes recent advancements at the intersection of artificial intelligence (AI), the Internet of Things (IoT), and compressed sensing (CS) as key enablers for next-generation WCE. AI-driven methods, particularly convolutional neural networks, have demonstrated near-human accuracy in lesion detection while significantly reducing diagnostic review times. IoT-enabled frameworks enhance connectivity, support remote monitoring, and integrate capsule data into broader digital health infrastructures. Meanwhile, compressed sensing techniques and application-specific integrated circuits (ASICs) improve transmission efficiency and extend battery life without compromising image quality. The paper further examines ongoing challenges in localization, where visual odometry, magnetic tracking, and hybrid multisensory fusion continue to evolve but remain short of clinical reliability. By consolidating these perspectives, the review highlights how cross-disciplinary integration is reshaping WCE from a passive imaging tool into an intelligent, multifunctional platform. However, significant translational gaps remain between experimental prototypes and clinical practice, underscoring the need for multidisciplinary collaboration. The convergence of AI, IoT, and CS not only addresses current bottlenecks but also paves the way for capsule systems capable of autonomous navigation, advanced diagnostics, and therapeutic intervention.\u003c/p\u003e","manuscriptTitle":"A Review of Novel Approaches to Indoor Localization Based on Wireless Capsule Endoscopy","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-05-15 02:36:19","doi":"10.21203/rs.3.rs-8896919/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewerAgreed","content":"","date":"2026-05-17T10:57:14+00:00","index":0,"fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-05-06T06:12:33+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-02-26T02:25:08+00:00","index":"","fulltext":""},{"type":"submitted","content":"EURASIP Journal on Wireless Communications and Networking","date":"2026-02-20T10:58:05+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"eurasip-journal-on-wireless-communications-and-networking","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"jwcn","sideBox":"Learn more about [EURASIP Journal on Wireless Communications and Networking](http://jwcn-eurasipjournals.springeropen.com)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/jwcn/default.aspx","title":"EURASIP Journal on Wireless Communications and Networking","twitterHandle":"@SpringerEng","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"1f957101-c601-4a7f-b37b-9149c0045c0e","owner":[],"postedDate":"May 15th, 2026","published":true,"recentEditorialEvents":[{"type":"reviewerAgreed","content":"","date":"2026-05-17T10:57:14+00:00","index":0,"fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-05-06T06:12:33+00:00","index":"","fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-15T02:36:20+00:00","versionOfRecord":[],"versionCreatedAt":"2026-05-15 02:36:19","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8896919","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8896919","identity":"rs-8896919","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}