FED-LIFE: Ghost LinkNet Enabled Federated Learning for Anomaly Detection in Smart Intensive Care Unit based on IOMT

FED-LIFE: Ghost LinkNet Enabled Federated Learning for Anomaly Detection in Smart Intensive Care Unit based on IOMT | 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 FED-LIFE: Ghost LinkNet Enabled Federated Learning for Anomaly Detection in Smart Intensive Care Unit based on IOMT A. Jothi Soruba Thaya, N. Karthikeyan This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6177554/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 9 You are reading this latest preprint version Abstract Intensive Care Unit (ICU) patient monitoring plays a vital role in ensuring the safety and well-being of critically ill patients by providing continuous and real-time insights into their health status. The integration of Internet of Medical Things (IoMT) devices in ICU including wearable sensors and remote monitoring tools, enables the seamless collection and transmission of patient data, allowing for real-time tracking of vital signs. Federated learning (FL) enhances this process by utilizing decentralized patient data to improve model generalization while maintaining data privacy. However, FL-based ICU monitoring faced challenges including high delays in decision-making due to centralized data processing, and significant execution time caused by the need to transfer large volumes of patient data. This research proposes a novel FEDerated learning-based LIFE saving ICU system (FED-LIFE) for effective tracking and providing timely health services to patients. The FED-LIFE system initially trains local models utilizing Ghostnet combined Enhanced LinkNet (Ghost_EliNet) which combines GhostNet and LinkNet, for tuning the Ghost_EliNet model a Red Deer Optimization (RDO) algorithm is employed for accurate service allocation. The suggested approach is implemented in Python programming. The efficacy of the developed approach is evaluated utilizing several metrics namely Precision, recall, f1-score, accuracy, delay, throughput, and execution time. The proposed method achieves the lowest delay of 22 seconds for 50 patients. Whereas the existing FEDSDM, Deep-CFL, and FL-IRL attain 45 seconds, 37 seconds, and 35 seconds for 50 patients respectively. Red Deer Optimization Federated Learning Internet of Medical Things Intensive Care Unit Patient Monitoring Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 1. Introduction ​The IoMT is revolutionizing the healthcare landscape by enabling the seamless interconnection of medical devices and applications, facilitating improved patient monitoring and care delivery [ 1 , 2 ].​ IoMT technologies play a pivotal role in increasing the efficiency and effectiveness of healthcare services specifically within the ICU where critically ill patients need constant surveillance and rapid medical intervention [ 3 , 4 ]. By connecting various medical devices namely monitors, sensors, and diagnostic tools, IoMT enables healthcare practitioners to make well-informed decisions fast [ 5 , 6 ]. ICU settings generate enormous volumes of diverse data, stemming from a multitude of monitoring devices that track vital signs, biometrics, and other critical health indicators [ 7 ]. This data includes traditional vital signs namely respiratory rate, heart rate, and blood pressure, and extends to biochemical markers, imaging results, and even data from wearable devices [ 8 , 9 ]. ICUs focus on patients recovering from major surgeries, severe trauma, or critical illnesses [ 10 ]. In addition to advanced monitoring systems, ICUs are equipped with various life-saving devices, including tools for pain management, resuscitation, respiratory assistance, and cardiac support [ 11 ]. These devices ensure that patients receive the necessary treatment and support around the clock, offering the best chance for recovery in critical situations [ 12 ]. FL is a decentralized machine learning (ML) framework that allows many participants such as hospitals and other devices to collaboratively train a model without distributing sensitive data [ 13 ]. Instead of pooling data in a central location, each participant trains a local model using their data and only shares model updates, ensuring privacy and security. In the context of ICU patient monitoring, FL allows the development of predictive models for critical care such as early detection of sepsis or organ failure, by using data from multiple ICUs while maintaining patient privacy [ 14 ]. However, traditional FL systems face several challenges, such as high latency, limited bandwidth, and potential breaches of patient privacy, which hinder their effectiveness in critical care environments [ 15 ]. To overcome these issues a novel approach has been proposed to minimize delays, lower execution time, and enhance the decision-making of ICU patients. The proposed work's major contributions are as follows, The key goal of this strategy is to develop an effective FL-based technique for the classification of ICU patient’s urgency levels by enabling timely and efficient medical interventions. Each ICU patient’s data is collected and processed locally at fog nodes, where it undergoes pre-processing using techniques namely data cleaning and normalization to ensure data quality. The proposed Ghost-LinkNet approach is utilized to train the local model and classify the health service urgency level of patients. The RDO algorithm is applied to tune the hyperparameters of the Ghost-LinkNet model. The efficacy of the developed approach is evaluated utilizing several metrics namely Precision, recall, f1-score, accuracy, delay, throughput, and execution time. The remaining portion of this research is arranged as follows, the Literature survey is summarized in section 2, and Section 3 details the suggested framework. Section 4 details the result and discussion. The future work and conclusion are included in Section 5. 2. Literature Survey This section explores the most recent advancements in FL applications, with a particular focus on their role in decision-making for smart healthcare systems. In 2022, Akter, M., et al. [ 16 ] suggested a three-tier Federated Edge Aggregator named Edge Intelligence as an FL-integrated privacy protection approach designed to safeguard Smart Healthcare applications at the edge against intrusions. This framework not only ensures enhanced privacy but also achieves an impressive 90% accuracy, outperforming the baseline technique regarding both accuracy and privacy protection. In 2023 Rajagopal, S.M., et al., [ 17 ] developed a Federated Learning-based Smart Decision-Making (FedSDM) approach for ECG data in microservice-combined IoT medical services. This framework leverages the benefits of Fog/Edge computing to enhance real-time performance in serious medical scenarios. The developed framework demonstrates that Edge-based deployment surpasses both Cloud and Fog in several key areas namely delay, energy consumption, execution time, cost, and network usage. In 2023 Gong, W., et al., [ 18 ] suggested a framework that integrates FL with inverse reinforcement learning (IRL) to develop an effective medical decision-making support tool, while ensuring the privacy of the patient. This framework was evaluated utilizing real-world medical data, and the results showed that it outperforms traditional methods, offering superior efficiency in a distributed manner. The framework enables efficient decision-making in medical applications without compromising patient confidentiality. In 2023 Alam, M.U. and Rahmani, R., [ 19 ] suggested the FedSepsis approach for the early detection of sepsis utilizing electronic health records. By leveraging multimodal frameworks with generative adversarial neural networks, FedSepsis achieves impressive results for an AUC-PR of 96.55%, an AUC-ROC of 99.35%, and an early detection time of 4.56 hours. FedSepsis demonstrates that integrating such advanced techniques, coupled with low-end computational devices, could provide significant benefits for all stakeholders in the medical sector and warrants further exploration. In 2023 Rakhmiddin, R. and Lee, K., [ 20 ] suggested an approach that combines FL with a cross-device multimodal approach for clinical event categorization based on vital signs data. The study demonstrates that FL serves as an effective tool for privacy-preserving clinical task classification and obtains 98.9% accuracy. In 2023 Di Napoli, C., et al., [ 21 ] suggested a federated learning architecture to enable distributed machine learning across healthcare institutions, ensuring that data remains securely within its original location. Experimental results reveal that knowledge sharing between nodes within the federated system enhances each node's capability to make accurate predictions, even for cases that were not previously encountered. The evaluation of the approach’s efficiency shows impressive accuracy and precision scores exceeding 0.91, highlighting the efficacy and potential of this FL system in healthcare services. In 2023 Muazu, T., et al., [ 22 ] suggested an edge-powered blockchain-combined FL approach for resource management in the IoMT. In this suggested approach, blockchain technology is utilized to enhance security features within both IoMT and edge computing environments. The results demonstrate that the system effectively reduces computational costs while maintaining robust security and privacy protections. Additionally, a security analysis confirms that the proposed framework is resilient to various types of security threats. In 2024 Nguyen, T.N., et al., [ 23 ] suggested a deep contrastive FL (Deep-CFL) framework that integrates explainable AI (XAI), CFL, and unbalanced supervised learning approaches to monitor and predict patient conditions in the ICU. Results show that Deep-CFL outperforms centralized learning-based systems achieving an average precision of 0.884, an AUC-ROC of 0.879, and an AUC-PR of 0.886. In 2024 Consul, P., et al [ 24 ] suggested a Federated Learning-based Wireless Body Area Network (FRLTO) approach for the IoMT. Numerical analysis shows that the proposed method enhances throughput by 37.06%, cuts energy consumption by approximately 69.84%, and decreases time delay by about 6.23% when contrasted to existing approaches. In 2024 Pan, W., et al., [ 25 ] introduced an adaptive FL approach for clinical risk identification using electronic health records from numerous hospitals. The framework divides the input features into domain-specific, stable, and condition-inappropriate components based on their relevance to clinical findings. The findings depicted that this approach surpasses previous FL baselines regarding prediction accuracy while also providing meaningful feature interpretations, enhancing both the performance and explainability of clinical risk predictions. From the above literature survey, several key drawbacks are identified. Many frameworks exhibit limited generalizability to diverse and heterogeneous healthcare environments, making them less adaptable to real-world scenarios. Significant delays and high execution times in certain approaches hinder real-time decision-making, which is critical in medical applications. To overcome this a novel FED-LIFE approach has been proposed to minimize delays, lower execution time, and enhance real-time decision-making in medical applications which will be covered in depth in the below section. 3. Proposed Method In this section a novel FEDerated learning-based LIFEsaving ICU system (FED-LIFE) has been proposed for effective tracking and providing timely health services to patients. FL decentralizes the training process by enabling individual devices or network segments to train local models utilizing their data. Instead of sharing raw data, these locally trained models are combined to create a global model, ensuring data privacy and security.Patient data including vital signs are collected from ICU patients through sensors and monitoring devices. The data is sent to local Fog nodes where it undergoes pre-processing utilizing a data cleaning and normalization approach. The hybrid Ghost_EliNet is developed by combining the GhostNet and enhanced Linknet, which is tuned optimally utilizing the RDO algorithm. Each fog node maintains a local model which is periodically updated to the cloud for aggregation into a global model. This global model integrates updates from all fog nodes and then synchronizes back with the local models for frequent improvement. The cloud-based global model allows doctors to remotely monitor patient statuses and make critical decisions in real-time, ensuring prompt and efficient care for ICU patients. The overall workflow of the suggested approach is demonstrated in Fig. 1 . 3.1 Data Preprocessing: Preprocessing is the process of transforming input data into a useful format by removing irrelevant data. The efficacy and accuracy of the suggested approach can only be increased by modifying and preparing the data to make it appropriate for the learning process. The subsequent tasks are included in the preprocessing phase: Data Cleaning Data cleaning implies correcting errors and detecting inconsistencies in the data to improve its quality. This process includes addressing null values using techniques such as median, interpolation, and means and handling outliers by either removing them or transforming them into a more suitable range. Data Normalization Data normalization scales features to a standard range to increase the efficiency of DL models. In this work, the Standardized Scalar normalization technique is employed to adjust the data to ensure a mean of 0 and a variance of 1. These assurances that every feature contributes equally to the learning process and enhances detection results. 3.2 Ghost Combined Enhanced Linknet Model The proposed Ghost_EliNet model is utilized to train the local model and classify the health service urgency level of patients. The Ghost_EliNet is developed by integrating the GhostNet and Linknet, wherein the hyperparameter is devised utilizing the Red Deer Optimization Algorithm. 3.2.1 GHOSTNET: GhostNet utilizes lightweight convolutional operations to minimize the model’s computational cost. This is particularly beneficial in resource-constrained environments namely edge or mobile devices in which the conventional approaches are computationally expensive. The Ghost Module in GhostNet reduces the number of parameters contrasted to traditional convolutional layers, lowering memory usage, which is vital for devices with limited resources. GhostNet enables faster inference which makes it ideal for real-time applications such as real-time classification and patient urgency identification, where latency is critical. Additionally, the GhostNet framework improves generalization, allowing it to perform well across diverse data without overfitting. Consequently, GhostNet is employed for feature mapping in the suggested FL-based patient urgency level identification module. The Ghost module first applies a convolutional layer for mapping, followed by a linear operation to derive Ghost features, as depicted in Fig. 2 . The dual paths employed by GhostNet are: Primary Path This approach employs the basic convolutional process, typically using depth-wise separable convolutions. These convolutions enhance computational efficiency by decoupling spatial filtering from channel-wise filtering, reducing the computational load compared to standard convolutions. The feature mapping that was attained using the primary route is expressed as follows $$\:{F}_{primary}=A*c+d$$ 1 In the GhostNet architecture, the feature map is represented as A, the bias is denoted as d, and the conventional filters are indicated by c. Ghost Path The ghost path is designed as a computationally efficient convolution operation that uses fewer channels, making it a low-complexity process. This path is responsible for capturing supplementary features with reduced computational overhead. By integrating the information from both the primary convolutional path and the ghost path, the network enhances its representational capacity while maintaining a high level of efficiency. The feature mapping produced by the ghost path can be expressed as \(\:{F}_{ghost}\) = A ∗ c ′ (2) where the filter employed in the ghost path is represented as c ′. The outputs from both paths are aggregated to form the final feature map: $$\:{F}_{agg}={F}_{primary}+\:{F}_{ghost}$$ 3 This lightweight effective feature extraction reduces the model's memory requirements and enhances its ability to operate in resource-constrained environments, such as edge devices. 3.2.2 Enhanced LinkNet The features extracted by GhostNet are classified by the LinkNet encoder architecture to capture long-term dependencies and spatial hierarchies. The encoder compresses the input features into compact, high-level representations, capturing critical patterns and relationships: $$\:{F}_{enc}=Encoder\left({F}_{agg}\right)$$ 4 This module employs convolutional layers interspersed with pooling operations to lower the spatial dimensions while retaining essential information. Skip connections link the encoder fine-grained features from earlier layers, which increases the overall accuracy and robustness of the classification. The output from the LinkNet encoder is passed through fully connected layers, culminating in a softmax activation function to classify the urgency level of the ICU patient: $$\:\widehat{y}=softmax(W.{F}_{enc}+b)$$ 5 Where \(\:\widehat{y}\) denotes the predicted urgency level, \(\:\:b\) indicates the bias term and \(\:W\) represents the weight matrix. 3.3 Hyperparameter Tuning via Red Deer Optimization The hyperparameter tuning process employs the Red Deer Optimization (RDO) algorithm to increase the classification accuracy of the proposed Ghost_EliNet model for ICU patient health monitoring. The RDO algorithm is inspired by the natural mating behaviors of red deer and focuses on selecting optimal hyperparameters in the model. Below is a detailed explanation of the RDO process. 3.3.1 Inspiration of RD algorithm The RDO algorithm mimics the mating behavior of red deer, with an emphasis on selecting the best-performing hyperparameters. The algorithm is designed to balance exploration (searching the solution space) and exploitation (refining existing solutions) to find optimal hyperparameter values that maximize the classification performance of the model. 3.3.2 Procedural steps of RD algorithm The RDO algorithm initiates with a randomly generated population of RDs, categorized into "male RDs" and "hinds." Male RDs roar to classify themselves as commanders or stags. Commanders and stags compete for harem ownership, with the number of hinds proportional to the commanders' abilities. Commanders mate with multiple hinds, while stags pair with the nearest hind, illustrating the algorithm's exploration and exploitation phases. These males and females interact within the solution space to optimize hyperparameter values. Figure 3 shows the flowchart of RDA. The initial population is defined by a set of hyperparameter values. These values are evaluated using a fitness function, which measures how well a particular set of hyperparameters enhances the model's performance such as accuracy. The fitness is calculated for each RD using the Eq. ( 6 ): $$\:Value=f\left(RD\right)=X1,X2,X3,\dots\:\:,{X}_{Nvar}\:$$ 6 In this phase, male RDs "roar" to establish dominance and compete for the best hyperparameters, mirroring the natural behavior of red deer. The males that achieve the best performance (highest fitness values) are designated as "commanders," while others are "stags." Commanders control a group of hinds (hyperparameters), leading them toward better solutions. The males update their positions (hyperparameter values) by exploring nearby solutions: $$\:{male}_{new}=\left\{\begin{array}{c}{male}_{old}+{a}_{1}\left(UB-LB\right)*{a}_{2}+LB),\:if\:{a}_{3}\ge\:0.5\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\\\:{male}_{old}+{a}_{1}\left(UB-LB\right)*{a}_{2}+LB),\:if\:{a}_{3}<0.5\:is\:less\:than\:0\end{array}\right.$$ 7 LB and UB define the limits of the search space to create an appropriate male neighborhood solution, representing the upper and lower boundaries. It is significant to note that the current position of the male RD is denoted as \(\:{male}_{old}\) , while its future position is \(\:{male}_{new}\) . Regard of randomization, a3, a1, and a2, represent the three phases of the roaring phase in nature, randomly drawn within a uniform range of 0 to 1. The quantity of male commanders is determined by the Eq. ( 8 ): $$\:{N}_{C}=round(\gamma\:.{N}_{male})$$ 8 where \(\:{N}_{C}\) represents the number of commanders among the male RDs, \(\:\gamma\:\) represents value chosen at random ranging from 0 to 1, and \(\:{N}_{male}\) indicates the total number of males. It is significant that \(\:\gamma\:\) serves as the initial value for the algorithm model, with a value range between zero and one. Lastly, the total stags are calculated through the below expression: $$\:{N}_{S}={N}_{male}-{N}_{C}$$ 9 In this phase, commanders and stags compete by adjusting their hyperparameters based on a combination of their values and those of their competitors. The fight produces two potential solutions, new1, and new2​, which are evaluated for fitness. The updated hyperparameters from the interaction are calculated using Eqs. ( 10 ) & ( 11 ): $$\:{new}_{1}=(C+S)/2+{b}_{1}(\left(UB-LB\right)*{b}_{2}+LB$$ 10 $$\:{new}_{2}=(C+S)/2+{b}_{1}(\left(UB-LB\right)*{b}_{2}+LB$$ 11 The two generated solutions of the fighting phase are denoted as \(\:{new}_{1}\) and \(\:{new}_{2}\) . The notations for stags and commanders are represented by S and C, respectively. The LB and UB define boundaries regarding the viability of these new solutions The LB and UB of the search space, b1, and b2, are determined through the randomization of the fighting phase, which utilizes an even distribution function ranging from zero to one. To form harem groups, hinds are distributed among commanders to create harems, in proportion to: $$\:{V}_{n}={v}_{n}-max{\:v}_{i}$$ 12 where \(\:{V}_{n}\) indicates the normalized value of the power of the nth commander and \(\:{v}_{n}\) depicts the nth commander's actual power. The below equation is utilized to evaluate the normalized power of the commanders: $$\:{P}_{n}=\left|\frac{{V}_{n}}{{\sum\:}_{i=1}^{{a}_{i}}{V}_{i}}\right|$$ 13 The number of hinds in a harem can be determined using the following equation: $$\:{N.harem}_{n}=round({P}_{n}.{N}_{hind})$$ 14 where \(\:{N}_{hind}\) represents the hinds. This mating behavior is performed by a commander, who controls a specific proportion of hinds within his group. $$\:N.{harem}_{n}^{mate}=round(\propto\:.N.{harem}_{n})$$ 15 The count of hinds in the nth harem that pairs with their leader is \(\:N.{harem}_{n}^{mate}\) . Regarding the solution space, the selection is made \(\:N.{harem}_{n}^{mate}\) of the \(\:N.{harem}_{k}\) at random. In general, the mating stage is stated as below: $$\:offs=\frac{C+Hind}{2}+\left(UB-LB\right)\:\times\:c$$ 16 A harem is chosen at random, represented by K, enabling the male leader to mate with \(\:\beta\:\) percent of the hinds in that group. To increase his domain, the leader can also initiate attacks on other harems. In this context, \(\:\beta\:\) acts as the initial parameter of the algorithm, varying between zero and one. The number of hinds that mate with the leader in the selected harem is computed utilizing Eq. ( 17 ): $$\:N.{harem}_{k}^{mate}=round(\propto\:.N.{harem}_{n})$$ 17 where \(\:N.{harem}_{k}^{mate}\) The K-th harem's count of hinds that engage in mating with the leader is represented numerically. The mating process follows the equation provided in Eq. ( 16 ). The distance between a stag and each hind in the J-dimensional is determined using the formula: $$\:{d}_{i}={\left\{\sum\:_{j\in\:J}{\left({stag}_{j}-{hind}_{j}^{i}\right)}^{2}\right\}}^{1/2}$$ 18 where \(\:{d}_{i}\) represents the distance between the i-th hind and the stag in the hyperparameter space. The selected hind corresponds to the minimum value in this distance matrix. The algorithm continues iterating, with male RDs (commanders) adjusting their positions and mating with the best hinds until convergence is achieved. The best hyperparameter set is selected based on the highest fitness value. The Red Deer Optimization algorithm effectively balances exploration and exploitation, optimizing the selection of hyperparameters for the Ghost_EliNet model. 4. Result and Discussion The FED-LIFE method’s experimental results are analyzed in this section. Performance is discussed regarding several metrics including recall, precision, accuracy, and f1score. The PC specifications used for this experiment included an i9-9820X 3.30GHz CPU, 2 TB of RAM, and an Ubuntu 20.04.1 LTS OS. The suggested approach is implemented in Python programming. The suggested approach efficacy is compared with FEDSDM, Deep-CFL and FL-IRL regarding recall, precision, accuracy, delay, and f1score. 4.1 Dataset description The dataset from AR Hospital in Ramanathapuram is focused on ICU patient health monitoring, containing comprehensive information such as patient demographics (age, gender, medical history), clinical records (ICU visits, symptoms, diagnoses), test results (blood tests, imaging scans), treatment details (medications, dosages), and patient outcomes (recovery, complications, mortality). Additionally, the study utilizes the MIMIC-IV dataset, which contains de-identified electronic health records from ICU patients at Beth Israel Deaconess Medical Center (BIDMC), Boston, spanning 2008 to 2019. With over 300,000 hospital admissions, it provides detailed patient data including diagnoses, demographics, medications, lab results, and vital signs, making it an essential resource for research in critical care, health results, and medical informatics. Access to MIMIC-IV needs approval through the PhysioNet Data Use Agreement (DUA). 4.2 Performance Metrics The FED-LIFE approach is assessed utilizing performance indicators, including accuracy, precision, f1score, MSE, and recall. These evaluation metrics can be derived with basic parameters such as False Positive (FalP), True Negative (TrN), False Negative (FalN), and True Positive (TrP). Accuracy A fundamental metric for measuring correct sensor measurements. In balanced sensor nodes, where False Positive (FalP) and False Negative (FalN) are nearly equal, statistical accuracy improves as it is proportional to the entire count of values. $$\:\text{A}\text{c}\text{c}\text{u}\text{r}\text{a}\text{c}\text{y}=\frac{\text{T}\text{r}\text{P}+\text{T}\text{r}\text{N}}{\text{F}\text{a}\text{l}\text{N}+\text{T}\text{r}\text{P}+\text{F}\text{a}\text{l}\text{P}+\text{T}\text{r}\text{N}}$$ 19 Precision It is stated as the ratio of correctly anticipated favorable findings to the entirety of favorable findings. $$\:\text{P}\text{r}\text{e}\text{c}\text{i}\text{s}\text{i}\text{o}\text{n}=\frac{\text{T}\text{r}\text{P}}{\text{T}\text{r}\text{P}+\text{F}\text{a}\text{l}\text{N}}$$ 20 Recall It is a ratio of positive comments that was accurately predicted based on every real observation made in class. $$\:\text{R}\text{e}\text{c}\text{a}\text{l}\text{l}=\frac{\text{T}\text{r}\text{P}}{\text{T}\text{r}\text{P}+\text{F}\text{a}\text{l}\text{N}}$$ 21 F1 score Recall and precision are averaged and weighted. This score therefore takes into consideration both FalN and FalP. $$\:\mathbf{F}1\:\mathbf{s}\mathbf{c}\mathbf{o}\mathbf{r}\mathbf{e}=2\times\:\frac{\text{P}\text{R}.\text{R}\text{C}}{\text{P}\text{R}+\text{R}\text{C}}$$ 22 4.2 Performance Comparison The suggested method performance is compared with existing FEDSDM, Deep-CFL, and FL-IRL regarding recall, precision, accuracy, and f1score. The efficacy of the suggested approach is measured using the AR, and MIMIC-IV datasets. Figures 4 show the accuracy comparison of suggested and existing frameworks. The accuracy of the suggested framework for the AR dataset is 99.01% over existing FEDSDM, Deep-CFL, and FL-IRL methods achieving a low accuracy of 97.21%, 95%, and 92.5% respectively. For the MIMIC-IV dataset, the suggested framework achieves an accuracy of 98.45% whereas the existing FEDSDM, Deep-CFL, and FL-IRL methods achieve a low accuracy of 96.02%, 96.56%, and 91.58% respectively. The suggested framework has attained an overall accuracy of 99.01% on the AR dataset. The proposed method of validation and testing is illustrated through accuracy and loss plots in Figs. 6 (a) and 6(b). These plots demonstrate the model's performance highlighting its effectiveness in patient health monitoring. The low loss values also reflect successful learning with minimal overfitting during the training process. The suggested framework has attained an overall accuracy of 98.45% on the MIMIC-IV dataset. The classification of the validation and testing is illustrated through accuracy and loss plots of the proposed method in Figs. 6 (a) and 6(b). These plots depict the model's performance highlighting its effectiveness in predicting health issues. The low loss values also reflect successful learning with minimal overfitting during the training process. Figures 7 (a), and 7(b) indicate the performance evaluation of the FED-LIFE and existing approaches across AR, and MIMIC-IV datasets regards to precision, recall, and f1score. The recall, f1score, and precision of the suggested framework for the AR dataset are 97.1%, 98.01%, and 96.21% respectively. For the MIMIC-IV dataset, the suggested method achieves recall, f1score, and precision of 98.17%, 97.01%, and 95.33%. Overall, all the suggested approach outperforms the existing approaches. Figure 8 illustrates the delay in seconds for four methods namely FEDSDM, Deep-CFL, FL-IRL, and the proposed method across a patient count ranging from 10 to 50. The proposed method achieves the lowest delay of 22 seconds for 50 patients. Whereas the existing FEDSDM, Deep-CFL, and FL-IRL attain 45 seconds, 37 seconds, and 35 seconds for 50 patients. Across all patient counts the proposed method consistently achieves the lowest delay indicating more efficient data transmission. Figure 9 illustrates the performance based on throughput with existing and suggested frameworks evaluated across varying numbers of patients ranging from 10 to 50. The throughput for FEDSDM is 45 Mbps for 10 patients and 50 Mbps for 50 patients which indicates slow performance. Whereas Deep-CFL and FL-IRL achieve 48 Mbps and 50 Mbps for 10 patients. The Proposed method consistently achieves the highest throughput beginning at 53 Mbps for 10 patients and increasing linearly to 69 Mbps for 50 patients. Overall, the Proposed method demonstrates a clear advantage in scalability and efficiency, outperforming the other methods across all patient counts. The number of patients and the execution time in milliseconds are displayed in Fig. 10 . to provide a visual representation of the results. As the number of patients rises the execution time decreases for the proposed method. The FED-LIFE framework achieves a less execution time of 2.1 ms for 10 patients whereas, the FEDSDM, Deep-CFL, and FL-IRL techniques achieve 5.6 ms, 7.8 ms, and 9.8 ms respectively. Figure 11 illustrates the bandwidth consumption (in Kbps) of four methods, FEDSDM, Deep-CFL, FL-IRL, and the proposed method, which was analyzed across varying numbers of patients ranging from 100 to 500. For 100 patients the FEDSDM achieves 89.6 Kbps and Deep-CFL attains 90 Kbps. FL-IRL achieves 82 Kbps for 500 patients. The Proposed method consistently outperforms the other techniques, starting at 96.5 Kbps for 100 patients and 90 Kbps for 500 patients. Overall, the Proposed method demonstrates superior bandwidth efficiency, maintaining higher values across all patient counts than the other approaches. 5. Conclusion In this section, a novel FED-LIFE framework has been proposed for effective tracking and providing timely health services to patients. The suggested system integrates the Ghost_EliNet model for efficient and accurate classification of patient conditions, enabling precise categorization into urgency levels. The integration of Ghost_EliNet improves the system’s ability to detect and classify patient deterioration swiftly, thus minimizing the risk of delayed medical responses. By combining Fog computing, Red Deer Optimization, and Ghost_EliNet, the framework ensures continuous, reliable patient monitoring and timely interventions by addressing critical needs in ICU settings and improving overall patient outcomes. The accuracy of the suggested framework for the AR dataset is 99.01% over existing FEDSDM, Deep-CFL, and FL-IRL methods achieving a low accuracy of 97.21%, 95%, and 92.5%. For the MIMIC-IV dataset, the suggested approach achieves an accuracy of 98.45% whereas the existing FEDSDM, Deep-CFL, and FL-IRL methods achieve a low accuracy of 96.02%, 96.56%, and 91.58% respectively. The proposed method demonstrates the lowest delay of 8 seconds for 10 patients and 22 seconds for 50 patients. In future work, the system can be expanded to increase data security and privacy using blockchain technology. Declarations Ethical approval: My research guide reviewed and ethically approved this manuscript for publishing in this Journal. Author Contributions statement: The authors confirm contribution to the paper as follows:Study conception and design: A. Jothi Soruba Thaya and N. Karthikeyan; Data collection: A. Jothi Soruba Thaya and N. Karthikeyan; Analysis and interpretation of results: A. Jothi Soruba Thaya and N. Karthikeyan; Draft manuscript preparation: A. Jothi Soruba Thaya and N. Karthikeyan. All authors reviewed the results and approved the final version of the manuscript. Competing interests: This paper has no conflict of interest for publishing Research funding: No Financial support Availability of data and material: Data sharing is not applicable to this article as no new data were created or analyzed in this Research. Human and Animal Rights: This article does not contain any studies with human or animal subjects performed by any of the authors. Informed consent: I certify that I have explained the nature and purpose of this study to the above-named individual, and I have discussed the potential benefits of this study participation. The questions the individual had about this study have been answered, and we will always be available to address future questions. Acknowledgments : The author would like to express his heartfelt gratitude to the supervisor for his guidance and unwavering support during this research for his guidance and support References Wu, Q., Chen, X., Zhou, Z., Zhang, J.: Fedhome: Cloud-edge based personalized federated learning for in-home health monitoring. IEEE Trans. Mob. Comput. 21 (8), 2818–2832 (2020) Nguyen, D.C., Pham, Q.V., Pathirana, P.N., Ding, M., Seneviratne, A., Lin, Z., Dobre, O., Hwang, W.J.: learning for smart healthcare: A survey. ACM Comput. Surv. (Csur). 55 (3), 1–37 (2022) Singh, C., Mishra, R., Gupta, H.P., Banga, G.: A federated learning-based patient monitoring system in internet of medical things. IEEE Trans. Comput. Social Syst. 10 (4), 1622–1628 (2022) Berghout, T., Benbouzid, M., Bentrcia, T., Lim, W.H., Amirat, Y.: Federated learning for condition monitoring of industrial processes: a review on fault diagnosis methods, challenges, and prospects. Electronics. 12 (1), 158 (2022) Alawadi, S., Kebande, V.R., Dong, Y., Bugeja, J., Persson, J.A., Olsson, C.M.: A federated interactive learning iot-based health monitoring platform. In European Conference on Advances in Databases and Information Systems, 235–246 Cham: Springer International Publishing. (2021) Shaik, T., Tao, X., Higgins, N., Gururajan, R., Li, Y., Zhou, X., Acharya, U.R.: FedStack: Personalized activity monitoring using stacked federated learning. Knowl. Based Syst. 257 , 109929 (2022) Rahman, A., Hossain, M.S., Muhammad, G., Kundu, D., Debnath, T., Rahman, M., Khan, M.S.I., Tiwari, P., Band, S.S.: Federated learning-based AI approaches in smart healthcare: concepts, taxonomies, challenges and open issues. Cluster Comput. 26 (4), 2271–2311 (2023) Arunan, A., Qin, Y., Li, X., Yuen, C.: A federated learning-based industrial health prognostics for heterogeneous edge devices using matched feature extraction. IEEE Trans. Autom. Sci. Eng. (2023) Mosaiyebzadeh, F., Pouriyeh, S., Parizi, R.M., Sheng, Q.Z., Han, M., Zhao, L., Sannino, G., Ranieri, C.M., Ueyama, J., Batista, D.M.: Privacy-enhancing technologies in federated learning for the internet of healthcare things: a survey. Electronics. 12 (12), 2703 (2023) Choi, G., Cha, W.C., Lee, S.U., Shin, S.Y.: Survey of Medical Applications of Federated Learning. Healthc. Inf. Res. 30 (1), 3–15 (2024) Zhong, C., Sarkar, A., Manna, S., Khan, M.Z., Noorwali, A., Das, A., Chakraborty, K.: Federated learning-guided intrusion detection and neural key exchange for safeguarding patient data on the internet of medical things. Int. J. Mach. Learn. Cybernet., 1–31 (2024) Pais, V., Rao, S., Muniyal, B., Yun, S.: FedICU: a federated learning model for reducing the medication prescription errors in intensive care units. Cogent Eng. 11 (1), 2301150 (2024) Ali, M., Naeem, F., Tariq, M., Kaddoum, G.: Federated learning for privacy preservation in smart healthcare systems: A comprehensive survey. IEEE J. biomedical health Inf. 27 (2), 778–789 (2022) Li, J., Meng, Y., Ma, L., Du, S., Zhu, H., Pei, Q., Shen, X.: A federated learning based privacy-preserving smart healthcare system. IEEE Trans. Industr. Inf., 18 (3) (2021) Lim, W.Y.B., Garg, S., Xiong, Z., Niyato, D., Leung, C., Miao, C., Guizani, M.: Dynamic contract design for federated learning in smart healthcare applications. IEEE Internet Things J. 8 (23), 16853–16862 (2020) Akter, M., Moustafa, N., Lynar, T., Razzak, I.: Edge intelligence: Federated learning-based privacy protection framework for smart healthcare systems. IEEE J. Biomedical Health Inf. 26 (12), 5805–5816 (2022) Rajagopal, S.M., Supriya, M., Buyya, R.: FedSDM: Federated learning based smart decision making module for ECG data in IoT integrated Edge–Fog–Cloud computing environments. Internet Things. 22 , 100784 (2023) Gong, W., Cao, L., Zhu, Y., Zuo, F., He, X., Zhou, H.: Federated inverse reinforcement learning for smart icus with differential privacy. IEEE Internet Things J. 10 (21), 19117–19124 (2023) Alam, M.U., Rahmani, R.: Fedsepsis: A federated multi-modal deep learning-based internet of medical things application for early detection of sepsis from electronic health records using raspberry pi and jetson nano devices. Sensors. 23 (2), 970 (2023) Rakhmiddin, R., Lee, K.: Federated Learning for Clinical Event Classification Using Vital Signs Data. Multimodal Technol. Interact. 7 (7), 67 (2023) Di Napoli, C., Paragliola, G., Ribino, P., Serino, L.: Balancing Uneven Knowledge of Hospital Nodes for ICU Patients Diagnosis through Federated Learning (2023) Muazu, T., Yingchi, M., Muhammad, A.U., Ibrahim, M., Samuel, O., Tiwari, P.: Iomt: A medical resource management system using edge empowered blockchain federated learning. IEEE Trans. Netw. Serv. Manage. (2023) Nguyen, T.N., Yang, H.J., Kho, B.G., Kang, S.R., Kim, S.H.: Explainable Deep Contrastive Federated Learning System for Early Prediction of Clinical Status in-Intensive Care Unit. IEEE Access. (2024) Consul, P., Budhiraja, I., Arora, R., Garg, S., Choi, B.J., Hossain, M.S.: Federated reinforcement learning based task offloading approach for MEC-assisted WBAN-enabled IoMT. Alexandria Eng. J. 86 , 56–66 (2024) Pan, W., Xu, Z., Rajendran, S., Wang, F.: An adaptive federated learning framework for clinical risk prediction with electronic health records from multiple hospitals. Patterns, 5 (1) (2024) Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 10 Sep, 2025 Reviews received at journal 04 Jul, 2025 Reviewers agreed at journal 01 Jul, 2025 Reviews received at journal 13 Jun, 2025 Reviewers agreed at journal 12 Jun, 2025 Reviewers invited by journal 06 May, 2025 Editor assigned by journal 27 Mar, 2025 Submission checks completed at journal 27 Mar, 2025 First submitted to journal 07 Mar, 2025 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-6177554","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":452431234,"identity":"738c8fd7-af88-40b2-8dc0-f95ece584975","order_by":0,"name":"A. Jothi Soruba Thaya","email":"data:image/png;base64,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","orcid":"","institution":"Syed Ammal Engineering College (Autonomous)","correspondingAuthor":true,"prefix":"","firstName":"A.","middleName":"Jothi Soruba","lastName":"Thaya","suffix":""},{"id":452431235,"identity":"404c71f7-2065-4fd0-9c82-eb3e7f362b2f","order_by":1,"name":"N. Karthikeyan","email":"","orcid":"","institution":"Syed Ammal Engineering College (Autonomous)","correspondingAuthor":false,"prefix":"","firstName":"N.","middleName":"","lastName":"Karthikeyan","suffix":""}],"badges":[],"createdAt":"2025-03-07 10:53:12","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6177554/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6177554/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":82561100,"identity":"af50e280-99d7-42c9-a6d2-5aae8386e9e8","added_by":"auto","created_at":"2025-05-13 01:38:08","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":220681,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eOverall workflow of the FED-LIFE\u003c/strong\u003e \u003cstrong\u003emethodology\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6177554/v1/2f3646368f99d12e9c79ec53.png"},{"id":82561102,"identity":"f3894295-d9a8-4f43-bed1-31869e14dc1f","added_by":"auto","created_at":"2025-05-13 01:38:08","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":143490,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eArchitecture of Ghost Enhanced LinkNet\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-6177554/v1/2dbae841e5bcb9fcb9c55c57.png"},{"id":82561098,"identity":"011f6c68-5d77-477e-adae-0c40ea711404","added_by":"auto","created_at":"2025-05-13 01:38:08","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":82231,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFlowchart of Red Deer Optimization Algorithm\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-6177554/v1/6ed340d8a8d8c934360fe8c1.png"},{"id":82561104,"identity":"5e2875bf-8391-4df2-8121-399637f21c55","added_by":"auto","created_at":"2025-05-13 01:38:08","extension":"jpeg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":153393,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eAccuracy comparison (a) AR dataset (b) MIMIC-IV dataset\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6177554/v1/bd715748afc799faed19dd1c.jpeg"},{"id":82561097,"identity":"3ba89ffc-160b-464f-831a-bf3c5e2439da","added_by":"auto","created_at":"2025-05-13 01:38:08","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":92935,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eAccuracy and loss curve of the AR dataset\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-6177554/v1/dfc02d25d5821125e8d6497f.png"},{"id":82561099,"identity":"f88c8be9-f1f5-4418-a906-c21c61423c80","added_by":"auto","created_at":"2025-05-13 01:38:08","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":72632,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eAccuracy and loss curve of the MIMIC-IV dataset\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-6177554/v1/f690d58a6ded1e60eed97b8c.png"},{"id":82561111,"identity":"73ad13f9-d1f8-4c70-9e9d-f94dd2154bc8","added_by":"auto","created_at":"2025-05-13 01:38:08","extension":"jpeg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":191814,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePerformance Comparison for datasets (a)AR (b) MIMIC-IV\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6177554/v1/b9289a521a547ca229d32f3a.jpeg"},{"id":82561103,"identity":"7949805b-7064-4867-a74b-e307c93b65be","added_by":"auto","created_at":"2025-05-13 01:38:08","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":85131,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDelay comparison with the proposed and existing method\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-6177554/v1/12e92cd99db5309f9f31734f.png"},{"id":82561105,"identity":"d83800eb-2378-484b-8d09-e545a6792bd4","added_by":"auto","created_at":"2025-05-13 01:38:08","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":84050,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThroughput comparison with the proposed and existing method\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-6177554/v1/9f74c4a2a20bbdbfc833024e.png"},{"id":82561113,"identity":"4d7ab8c1-360b-4640-b266-450f52ca5c8c","added_by":"auto","created_at":"2025-05-13 01:38:09","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":25975,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eExecution time comparison\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-6177554/v1/072cdf56d75feb5ad4f967b3.png"},{"id":82561115,"identity":"36cd3243-da86-49a1-9fa3-4351179ab79e","added_by":"auto","created_at":"2025-05-13 01:38:09","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":43847,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eBandwidth Comparison\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage11.png","url":"https://assets-eu.researchsquare.com/files/rs-6177554/v1/9635a44866cfa30913b6db2d.png"},{"id":82563914,"identity":"b223d233-3abf-416a-b7e8-4236bf4103a5","added_by":"auto","created_at":"2025-05-13 02:10:11","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1934800,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6177554/v1/c9a4988c-994f-4781-8d39-e34d60c4c226.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"FED-LIFE: Ghost LinkNet Enabled Federated Learning for Anomaly Detection in Smart Intensive Care Unit based on IOMT","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003e​The IoMT is revolutionizing the healthcare landscape by enabling the seamless interconnection of medical devices and applications, facilitating improved patient monitoring and care delivery [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e].​ IoMT technologies play a pivotal role in increasing the efficiency and effectiveness of healthcare services specifically within the ICU where critically ill patients need constant surveillance and rapid medical intervention [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. By connecting various medical devices namely monitors, sensors, and diagnostic tools, IoMT enables healthcare practitioners to make well-informed decisions fast [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eICU settings generate enormous volumes of diverse data, stemming from a multitude of monitoring devices that track vital signs, biometrics, and other critical health indicators [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. This data includes traditional vital signs namely respiratory rate, heart rate, and blood pressure, and extends to biochemical markers, imaging results, and even data from wearable devices [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. ICUs focus on patients recovering from major surgeries, severe trauma, or critical illnesses [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. In addition to advanced monitoring systems, ICUs are equipped with various life-saving devices, including tools for pain management, resuscitation, respiratory assistance, and cardiac support [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. These devices ensure that patients receive the necessary treatment and support around the clock, offering the best chance for recovery in critical situations [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eFL is a decentralized machine learning (ML) framework that allows many participants such as hospitals and other devices to collaboratively train a model without distributing sensitive data [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Instead of pooling data in a central location, each participant trains a local model using their data and only shares model updates, ensuring privacy and security. In the context of ICU patient monitoring, FL allows the development of predictive models for critical care such as early detection of sepsis or organ failure, by using data from multiple ICUs while maintaining patient privacy [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. However, traditional FL systems face several challenges, such as high latency, limited bandwidth, and potential breaches of patient privacy, which hinder their effectiveness in critical care environments [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. To overcome these issues a novel approach has been proposed to minimize delays, lower execution time, and enhance the decision-making of ICU patients. The proposed work's major contributions are as follows,\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eThe key goal of this strategy is to develop an effective FL-based technique for the classification of ICU patient\u0026rsquo;s urgency levels by enabling timely and efficient medical interventions.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eEach ICU patient\u0026rsquo;s data is collected and processed locally at fog nodes, where it undergoes pre-processing using techniques namely data cleaning and normalization to ensure data quality.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eThe proposed Ghost-LinkNet approach is utilized to train the local model and classify the health service urgency level of patients. The RDO algorithm is applied to tune the hyperparameters of the Ghost-LinkNet model.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eThe efficacy of the developed approach is evaluated utilizing several metrics namely Precision, recall, f1-score, accuracy, delay, throughput, and execution time.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eThe remaining portion of this research is arranged as follows, the Literature survey is summarized in section 2, and Section 3 details the suggested framework. Section 4 details the result and discussion. The future work and conclusion are included in Section 5.\u003c/p\u003e"},{"header":"2. Literature Survey","content":"\u003cp\u003eThis section explores the most recent advancements in FL applications, with a particular focus on their role in decision-making for smart healthcare systems.\u003c/p\u003e \u003cp\u003eIn 2022, Akter, M., et al. [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e] suggested a three-tier Federated Edge Aggregator named Edge Intelligence as an FL-integrated privacy protection approach designed to safeguard Smart Healthcare applications at the edge against intrusions. This framework not only ensures enhanced privacy but also achieves an impressive 90% accuracy, outperforming the baseline technique regarding both accuracy and privacy protection.\u003c/p\u003e \u003cp\u003eIn 2023 Rajagopal, S.M., et al., [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e] developed a Federated Learning-based Smart Decision-Making (FedSDM) approach for ECG data in microservice-combined IoT medical services. This framework leverages the benefits of Fog/Edge computing to enhance real-time performance in serious medical scenarios. The developed framework demonstrates that Edge-based deployment surpasses both Cloud and Fog in several key areas namely delay, energy consumption, execution time, cost, and network usage.\u003c/p\u003e \u003cp\u003eIn 2023 Gong, W., et al., [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e] suggested a framework that integrates FL with inverse reinforcement learning (IRL) to develop an effective medical decision-making support tool, while ensuring the privacy of the patient. This framework was evaluated utilizing real-world medical data, and the results showed that it outperforms traditional methods, offering superior efficiency in a distributed manner. The framework enables efficient decision-making in medical applications without compromising patient confidentiality.\u003c/p\u003e \u003cp\u003eIn 2023 Alam, M.U. and Rahmani, R., [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e] suggested the FedSepsis approach for the early detection of sepsis utilizing electronic health records. By leveraging multimodal frameworks with generative adversarial neural networks, FedSepsis achieves impressive results for an AUC-PR of 96.55%, an AUC-ROC of 99.35%, and an early detection time of 4.56 hours. FedSepsis demonstrates that integrating such advanced techniques, coupled with low-end computational devices, could provide significant benefits for all stakeholders in the medical sector and warrants further exploration.\u003c/p\u003e \u003cp\u003eIn 2023 Rakhmiddin, R. and Lee, K., [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] suggested an approach that combines FL with a cross-device multimodal approach for clinical event categorization based on vital signs data. The study demonstrates that FL serves as an effective tool for privacy-preserving clinical task classification and obtains 98.9% accuracy.\u003c/p\u003e \u003cp\u003eIn 2023 Di Napoli, C., et al., [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e] suggested a federated learning architecture to enable distributed machine learning across healthcare institutions, ensuring that data remains securely within its original location. Experimental results reveal that knowledge sharing between nodes within the federated system enhances each node's capability to make accurate predictions, even for cases that were not previously encountered. The evaluation of the approach\u0026rsquo;s efficiency shows impressive accuracy and precision scores exceeding 0.91, highlighting the efficacy and potential of this FL system in healthcare services.\u003c/p\u003e \u003cp\u003eIn 2023 Muazu, T., et al., [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e] suggested an edge-powered blockchain-combined FL approach for resource management in the IoMT. In this suggested approach, blockchain technology is utilized to enhance security features within both IoMT and edge computing environments. The results demonstrate that the system effectively reduces computational costs while maintaining robust security and privacy protections. Additionally, a security analysis confirms that the proposed framework is resilient to various types of security threats.\u003c/p\u003e \u003cp\u003eIn 2024 Nguyen, T.N., et al., [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e] suggested a deep contrastive FL (Deep-CFL) framework that integrates explainable AI (XAI), CFL, and unbalanced supervised learning approaches to monitor and predict patient conditions in the ICU. Results show that Deep-CFL outperforms centralized learning-based systems achieving an average precision of 0.884, an AUC-ROC of 0.879, and an AUC-PR of 0.886.\u003c/p\u003e \u003cp\u003eIn 2024 Consul, P., et al [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e] suggested a Federated Learning-based Wireless Body Area Network (FRLTO) approach for the IoMT. Numerical analysis shows that the proposed method enhances throughput by 37.06%, cuts energy consumption by approximately 69.84%, and decreases time delay by about 6.23% when contrasted to existing approaches.\u003c/p\u003e \u003cp\u003eIn 2024 Pan, W., et al., [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] introduced an adaptive FL approach for clinical risk identification using electronic health records from numerous hospitals. The framework divides the input features into domain-specific, stable, and condition-inappropriate components based on their relevance to clinical findings. The findings depicted that this approach surpasses previous FL baselines regarding prediction accuracy while also providing meaningful feature interpretations, enhancing both the performance and explainability of clinical risk predictions.\u003c/p\u003e \u003cp\u003eFrom the above literature survey, several key drawbacks are identified. Many frameworks exhibit limited generalizability to diverse and heterogeneous healthcare environments, making them less adaptable to real-world scenarios. Significant delays and high execution times in certain approaches hinder real-time decision-making, which is critical in medical applications. To overcome this a novel FED-LIFE approach has been proposed to minimize delays, lower execution time, and enhance real-time decision-making in medical applications which will be covered in depth in the below section.\u003c/p\u003e"},{"header":"3. Proposed Method","content":"\u003cp\u003eIn this section a novel FEDerated learning-based LIFEsaving ICU system (FED-LIFE) has been proposed for effective tracking and providing timely health services to patients. FL decentralizes the training process by enabling individual devices or network segments to train local models utilizing their data. Instead of sharing raw data, these locally trained models are combined to create a global model, ensuring data privacy and security.Patient data including vital signs are collected from ICU patients through sensors and monitoring devices. The data is sent to local Fog nodes where it undergoes pre-processing utilizing a data cleaning and normalization approach. The hybrid Ghost_EliNet is developed by combining the GhostNet and enhanced Linknet, which is tuned optimally utilizing the RDO algorithm. Each fog node maintains a local model which is periodically updated to the cloud for aggregation into a global model. This global model integrates updates from all fog nodes and then synchronizes back with the local models for frequent improvement. The cloud-based global model allows doctors to remotely monitor patient statuses and make critical decisions in real-time, ensuring prompt and efficient care for ICU patients. The overall workflow of the suggested approach is demonstrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Data Preprocessing:\u003c/h2\u003e \u003cp\u003ePreprocessing is the process of transforming input\u003c/p\u003e \u003cp\u003edata into a useful format by removing irrelevant data. The efficacy and accuracy of the suggested approach can only be increased by modifying and preparing the data to make it appropriate for the learning process. The subsequent tasks are included in the preprocessing phase:\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eData Cleaning\u003c/strong\u003e \u003cp\u003eData cleaning implies correcting errors and detecting inconsistencies in the data to improve its quality. This process includes addressing null values using techniques such as median, interpolation, and means and handling outliers by either removing them or transforming them into a more suitable range.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eData Normalization\u003c/strong\u003e \u003cp\u003eData normalization scales features to a standard range to increase the efficiency of DL models. In this work, the Standardized Scalar normalization technique is employed to adjust the data to ensure a mean of 0 and a variance of 1. These assurances that every feature contributes equally to the learning process and enhances detection results.\u003c/p\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Ghost Combined Enhanced Linknet Model\u003c/h2\u003e \u003cp\u003eThe proposed Ghost_EliNet model is utilized to train the local model and classify the health service urgency level of patients. The Ghost_EliNet is developed by integrating the GhostNet and Linknet, wherein the hyperparameter is devised utilizing the Red Deer Optimization Algorithm.\u003c/p\u003e \u003cdiv id=\"Sec6\" class=\"Section3\"\u003e \u003ch2\u003e3.2.1 GHOSTNET:\u003c/h2\u003e \u003cp\u003eGhostNet utilizes lightweight convolutional operations to minimize the model\u0026rsquo;s computational cost. This is particularly beneficial in resource-constrained environments namely edge or mobile devices in which the conventional approaches are computationally expensive. The Ghost Module in GhostNet reduces the number of parameters contrasted to traditional convolutional layers, lowering memory usage, which is vital for devices with limited resources. GhostNet enables faster inference which makes it ideal for real-time applications such as real-time classification and patient urgency identification, where latency is critical. Additionally, the GhostNet framework improves generalization, allowing it to perform well across diverse data without overfitting. Consequently, GhostNet is employed for feature mapping in the suggested FL-based patient urgency level identification module. The Ghost module first applies a convolutional layer for mapping, followed by a linear operation to derive Ghost features, as depicted in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. The dual paths employed by GhostNet are:\u003c/p\u003e \u003cp\u003e \u003cstrong\u003ePrimary Path\u003c/strong\u003e \u003cp\u003eThis approach employs the basic convolutional process, typically using depth-wise separable convolutions. These convolutions enhance computational efficiency by decoupling spatial filtering from channel-wise filtering, reducing the computational load compared to standard convolutions. The feature mapping that was attained using the primary route is expressed as follows\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv id=\"Equ1\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ1\" name=\"EquationSource\"\u003e\n$$\\:{F}_{primary}=A*c+d$$\u003c/div\u003e \u003cdiv class=\"EquationNumber\"\u003e1\u003c/div\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eIn the GhostNet architecture, the feature map is represented as A, the bias is denoted as d, and the conventional filters are indicated by c.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eGhost Path\u003c/strong\u003e \u003cp\u003eThe ghost path is designed as a computationally efficient convolution operation that uses fewer channels, making it a low-complexity process. This path is responsible for capturing supplementary features with reduced computational overhead. By integrating the information from both the primary convolutional path and the ghost path, the network enhances its representational capacity while maintaining a high level of efficiency. The feature mapping produced by the ghost path can be expressed as\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cspan class=\"InlineEquation\"\u003e \u003cspan class=\"mathinline\"\u003e\\(\\:{F}_{ghost}\\)\u003c/span\u003e \u003c/span\u003e = A \u0026lowast; c \u0026prime; (2)\u003c/p\u003e \u003cp\u003ewhere the filter employed in the ghost path is represented as c \u0026prime;. The outputs from both paths are aggregated to form the final feature map:\u003cdiv id=\"Equ2\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ2\" name=\"EquationSource\"\u003e\n$$\\:{F}_{agg}={F}_{primary}+\\:{F}_{ghost}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e3\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eThis lightweight effective feature extraction reduces the model's memory requirements and enhances its ability to operate in resource-constrained environments, such as edge devices.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section3\"\u003e \u003ch2\u003e3.2.2 Enhanced LinkNet\u003c/h2\u003e \u003cp\u003eThe features extracted by GhostNet are classified by the LinkNet encoder architecture to capture long-term dependencies and spatial hierarchies. The encoder compresses the input features into compact, high-level representations, capturing critical patterns and relationships:\u003cdiv id=\"Equ3\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ3\" name=\"EquationSource\"\u003e\n$$\\:{F}_{enc}=Encoder\\left({F}_{agg}\\right)$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e4\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eThis module employs convolutional layers interspersed with pooling operations to lower the spatial dimensions while retaining essential information. Skip connections link the encoder fine-grained features from earlier layers, which increases the overall accuracy and robustness of the classification. The output from the LinkNet encoder is passed through fully connected layers, culminating in a softmax activation function to classify the urgency level of the ICU patient:\u003cdiv id=\"Equ4\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ4\" name=\"EquationSource\"\u003e\n$$\\:\\widehat{y}=softmax(W.{F}_{enc}+b)$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e5\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eWhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\widehat{y}\\)\u003c/span\u003e\u003c/span\u003e denotes the predicted urgency level,\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\:b\\)\u003c/span\u003e\u003c/span\u003e indicates the bias term and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:W\\)\u003c/span\u003e\u003c/span\u003e represents the weight matrix.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Hyperparameter Tuning via Red Deer Optimization\u003c/h2\u003e \u003cp\u003eThe hyperparameter tuning process employs the Red Deer Optimization (RDO) algorithm to increase the classification accuracy of the proposed Ghost_EliNet model for ICU patient health monitoring. The RDO algorithm is inspired by the natural mating behaviors of red deer and focuses on selecting optimal hyperparameters in the model. Below is a detailed explanation of the RDO process.\u003c/p\u003e \u003cdiv id=\"Sec9\" class=\"Section3\"\u003e \u003ch2\u003e3.3.1 Inspiration of RD algorithm\u003c/h2\u003e \u003cp\u003eThe RDO algorithm mimics the mating behavior of red deer, with an emphasis on selecting the best-performing hyperparameters. The algorithm is designed to balance exploration (searching the solution space) and exploitation (refining existing solutions) to find optimal hyperparameter values that maximize the classification performance of the model.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section3\"\u003e \u003ch2\u003e3.3.2 Procedural steps of RD algorithm\u003c/h2\u003e \u003cp\u003eThe RDO algorithm initiates with a randomly generated population of RDs, categorized into \"male RDs\" and \"hinds.\" Male RDs roar to classify themselves as commanders or stags. Commanders and stags compete for harem ownership, with the number of hinds proportional to the commanders' abilities. Commanders mate with multiple hinds, while stags pair with the nearest hind, illustrating the algorithm's exploration and exploitation phases. These males and females interact within the solution space to optimize hyperparameter values.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e shows the flowchart of RDA. The initial population is defined by a set of hyperparameter values. These values are evaluated using a fitness function, which measures how well a particular set of hyperparameters enhances the model's performance such as accuracy. The fitness is calculated for each RD using the Eq.\u0026nbsp;(\u003cspan refid=\"Equ5\" class=\"InternalRef\"\u003e6\u003c/span\u003e):\u003cdiv id=\"Equ5\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ5\" name=\"EquationSource\"\u003e\n$$\\:Value=f\\left(RD\\right)=X1,X2,X3,\\dots\\:\\:,{X}_{Nvar}\\:$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e6\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eIn this phase, male RDs \"roar\" to establish dominance and compete for the best hyperparameters, mirroring the natural behavior of red deer. The males that achieve the best performance (highest fitness values) are designated as \"commanders,\" while others are \"stags.\" Commanders control a group of hinds (hyperparameters), leading them toward better solutions. The males update their positions (hyperparameter values) by exploring nearby solutions:\u003cdiv id=\"Equ6\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ6\" name=\"EquationSource\"\u003e\n$$\\:{male}_{new}=\\left\\{\\begin{array}{c}{male}_{old}+{a}_{1}\\left(UB-LB\\right)*{a}_{2}+LB),\\:if\\:{a}_{3}\\ge\\:0.5\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\\\\\:{male}_{old}+{a}_{1}\\left(UB-LB\\right)*{a}_{2}+LB),\\:if\\:{a}_{3}\u0026lt;0.5\\:is\\:less\\:than\\:0\\end{array}\\right.$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e7\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eLB and UB define the limits of the search space to create an appropriate male neighborhood solution, representing the upper and lower boundaries. It is significant to note that the current position of the male RD is denoted as \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{male}_{old}\\)\u003c/span\u003e\u003c/span\u003e, while its future position is \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{male}_{new}\\)\u003c/span\u003e\u003c/span\u003e. Regard of randomization, a3, a1, and a2, represent the three phases of the roaring phase in nature, randomly drawn within a uniform range of 0 to 1. The quantity of male commanders is determined by the Eq.\u0026nbsp;(\u003cspan refid=\"Equ7\" class=\"InternalRef\"\u003e8\u003c/span\u003e):\u003cdiv id=\"Equ7\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ7\" name=\"EquationSource\"\u003e\n$$\\:{N}_{C}=round(\\gamma\\:.{N}_{male})$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e8\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{N}_{C}\\)\u003c/span\u003e\u003c/span\u003e represents the number of commanders among the male RDs, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\gamma\\:\\)\u003c/span\u003e\u003c/span\u003e represents value chosen at random ranging from 0 to 1, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{N}_{male}\\)\u003c/span\u003e\u003c/span\u003e indicates the total number of males. It is significant that \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\gamma\\:\\)\u003c/span\u003e\u003c/span\u003e serves as the initial value for the algorithm model, with a value range between zero and one. Lastly, the total stags are calculated through the below expression:\u003cdiv id=\"Equ8\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ8\" name=\"EquationSource\"\u003e\n$$\\:{N}_{S}={N}_{male}-{N}_{C}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e9\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eIn this phase, commanders and stags compete by adjusting their hyperparameters based on a combination of their values and those of their competitors. The fight produces two potential solutions, new1, and new2​, which are evaluated for fitness. The updated hyperparameters from the interaction are calculated using Eqs.\u0026nbsp;(\u003cspan refid=\"Equ9\" class=\"InternalRef\"\u003e10\u003c/span\u003e) \u0026amp; (\u003cspan refid=\"Equ10\" class=\"InternalRef\"\u003e11\u003c/span\u003e):\u003cdiv id=\"Equ9\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ9\" name=\"EquationSource\"\u003e\n$$\\:{new}_{1}=(C+S)/2+{b}_{1}(\\left(UB-LB\\right)*{b}_{2}+LB$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e10\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equ10\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ10\" name=\"EquationSource\"\u003e\n$$\\:{new}_{2}=(C+S)/2+{b}_{1}(\\left(UB-LB\\right)*{b}_{2}+LB$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e11\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eThe two generated solutions of the fighting phase are denoted as \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{new}_{1}\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{new}_{2}\\)\u003c/span\u003e\u003c/span\u003e. The notations for stags and commanders are represented by S and C, respectively. The LB and UB define boundaries regarding the viability of these new solutions The LB and UB of the search space, b1, and b2, are determined through the randomization of the fighting phase, which utilizes an even distribution function ranging from zero to one. To form harem groups, hinds are distributed among commanders to create harems, in proportion to:\u003cdiv id=\"Equ11\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ11\" name=\"EquationSource\"\u003e\n$$\\:{V}_{n}={v}_{n}-max{\\:v}_{i}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e12\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{V}_{n}\\)\u003c/span\u003e\u003c/span\u003e indicates the normalized value of the power of the nth commander and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{v}_{n}\\)\u003c/span\u003e\u003c/span\u003e depicts the nth commander's actual power. The below equation is utilized to evaluate the normalized power of the commanders:\u003cdiv id=\"Equ12\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ12\" name=\"EquationSource\"\u003e\n$$\\:{P}_{n}=\\left|\\frac{{V}_{n}}{{\\sum\\:}_{i=1}^{{a}_{i}}{V}_{i}}\\right|$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e13\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eThe number of hinds in a harem can be determined using the following equation:\u003cdiv id=\"Equ13\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ13\" name=\"EquationSource\"\u003e\n$$\\:{N.harem}_{n}=round({P}_{n}.{N}_{hind})$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e14\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{N}_{hind}\\)\u003c/span\u003e\u003c/span\u003e represents the hinds. This mating behavior is performed by a commander, who controls a specific proportion of hinds within his group.\u003cdiv id=\"Equ14\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ14\" name=\"EquationSource\"\u003e\n$$\\:N.{harem}_{n}^{mate}=round(\\propto\\:.N.{harem}_{n})$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e15\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eThe count of hinds in the nth harem that pairs with their leader is \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:N.{harem}_{n}^{mate}\\)\u003c/span\u003e\u003c/span\u003e. Regarding the solution space, the selection is made \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:N.{harem}_{n}^{mate}\\)\u003c/span\u003e\u003c/span\u003e of the \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:N.{harem}_{k}\\)\u003c/span\u003e\u003c/span\u003e at random. In general, the mating stage is stated as below:\u003cdiv id=\"Equ15\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ15\" name=\"EquationSource\"\u003e\n$$\\:offs=\\frac{C+Hind}{2}+\\left(UB-LB\\right)\\:\\times\\:c$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e16\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eA harem is chosen at random, represented by K, enabling the male leader to mate with \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\beta\\:\\)\u003c/span\u003e\u003c/span\u003e percent of the hinds in that group. To increase his domain, the leader can also initiate attacks on other harems. In this context, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\beta\\:\\)\u003c/span\u003e\u003c/span\u003e acts as the initial parameter of the algorithm, varying between zero and one. The number of hinds that mate with the leader in the selected harem is computed utilizing Eq.\u0026nbsp;(\u003cspan refid=\"Equ16\" class=\"InternalRef\"\u003e17\u003c/span\u003e):\u003cdiv id=\"Equ16\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ16\" name=\"EquationSource\"\u003e\n$$\\:N.{harem}_{k}^{mate}=round(\\propto\\:.N.{harem}_{n})$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e17\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:N.{harem}_{k}^{mate}\\)\u003c/span\u003e\u003c/span\u003e The K-th harem's count of hinds that engage in mating with the leader is represented numerically. The mating process follows the equation provided in Eq.\u0026nbsp;(\u003cspan refid=\"Equ15\" class=\"InternalRef\"\u003e16\u003c/span\u003e). The distance between a stag and each hind in the J-dimensional is determined using the formula:\u003cdiv id=\"Equ17\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ17\" name=\"EquationSource\"\u003e\n$$\\:{d}_{i}={\\left\\{\\sum\\:_{j\\in\\:J}{\\left({stag}_{j}-{hind}_{j}^{i}\\right)}^{2}\\right\\}}^{1/2}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e18\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{d}_{i}\\)\u003c/span\u003e\u003c/span\u003e represents the distance between the i-th hind and the stag in the hyperparameter space. The selected hind corresponds to the minimum value in this distance matrix. The algorithm continues iterating, with male RDs (commanders) adjusting their positions and mating with the best hinds until convergence is achieved. The best hyperparameter set is selected based on the highest fitness value. The Red Deer Optimization algorithm effectively balances exploration and exploitation, optimizing the selection of hyperparameters for the Ghost_EliNet model.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"4. Result and Discussion","content":"\u003cp\u003eThe FED-LIFE method\u0026rsquo;s experimental results are analyzed in this section. Performance is discussed regarding several metrics including recall, precision, accuracy, and f1score. The PC specifications used for this experiment included an i9-9820X 3.30GHz CPU, 2 TB of RAM, and an Ubuntu 20.04.1 LTS OS. The suggested approach is implemented in Python programming. The suggested approach efficacy is compared with FEDSDM, Deep-CFL and FL-IRL regarding recall, precision, accuracy, delay, and f1score.\u003c/p\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Dataset description\u003c/h2\u003e \u003cp\u003eThe dataset from AR Hospital in Ramanathapuram is focused on ICU patient health monitoring, containing comprehensive information such as patient demographics (age, gender, medical history), clinical records (ICU visits, symptoms, diagnoses), test results (blood tests, imaging scans), treatment details (medications, dosages), and patient outcomes (recovery, complications, mortality). Additionally, the study utilizes the MIMIC-IV dataset, which contains de-identified electronic health records from ICU patients at Beth Israel Deaconess Medical Center (BIDMC), Boston, spanning 2008 to 2019. With over 300,000 hospital admissions, it provides detailed patient data including diagnoses, demographics, medications, lab results, and vital signs, making it an essential resource for research in critical care, health results, and medical informatics. Access to MIMIC-IV needs approval through the PhysioNet Data Use Agreement (DUA).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Performance Metrics\u003c/h2\u003e \u003cp\u003eThe FED-LIFE approach is assessed utilizing performance indicators, including accuracy, precision, f1score, MSE, and recall. These evaluation metrics can be derived with basic parameters such as False Positive (FalP), True Negative (TrN), False Negative (FalN), and True Positive (TrP).\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eAccuracy\u003c/strong\u003e \u003cp\u003eA fundamental metric for measuring correct sensor measurements. In balanced sensor nodes, where False Positive (FalP) and False Negative (FalN) are nearly equal, statistical accuracy improves as it is proportional to the entire count of values.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv id=\"Equ18\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ18\" name=\"EquationSource\"\u003e\n$$\\:\\text{A}\\text{c}\\text{c}\\text{u}\\text{r}\\text{a}\\text{c}\\text{y}=\\frac{\\text{T}\\text{r}\\text{P}+\\text{T}\\text{r}\\text{N}}{\\text{F}\\text{a}\\text{l}\\text{N}+\\text{T}\\text{r}\\text{P}+\\text{F}\\text{a}\\text{l}\\text{P}+\\text{T}\\text{r}\\text{N}}$$\u003c/div\u003e \u003cdiv class=\"EquationNumber\"\u003e19\u003c/div\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003ePrecision\u003c/strong\u003e \u003cp\u003eIt is stated as the ratio of correctly anticipated favorable findings to the entirety of favorable findings.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv id=\"Equ19\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ19\" name=\"EquationSource\"\u003e\n$$\\:\\text{P}\\text{r}\\text{e}\\text{c}\\text{i}\\text{s}\\text{i}\\text{o}\\text{n}=\\frac{\\text{T}\\text{r}\\text{P}}{\\text{T}\\text{r}\\text{P}+\\text{F}\\text{a}\\text{l}\\text{N}}$$\u003c/div\u003e \u003cdiv class=\"EquationNumber\"\u003e20\u003c/div\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eRecall\u003c/strong\u003e \u003cp\u003eIt is a ratio of positive comments that was accurately predicted based on every real observation made in class.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv id=\"Equ20\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ20\" name=\"EquationSource\"\u003e\n$$\\:\\text{R}\\text{e}\\text{c}\\text{a}\\text{l}\\text{l}=\\frac{\\text{T}\\text{r}\\text{P}}{\\text{T}\\text{r}\\text{P}+\\text{F}\\text{a}\\text{l}\\text{N}}$$\u003c/div\u003e \u003cdiv class=\"EquationNumber\"\u003e21\u003c/div\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eF1 score\u003c/strong\u003e \u003cp\u003eRecall and precision are averaged and weighted. This score therefore takes into consideration both FalN and FalP.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv id=\"Equ21\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ21\" name=\"EquationSource\"\u003e\n$$\\:\\mathbf{F}1\\:\\mathbf{s}\\mathbf{c}\\mathbf{o}\\mathbf{r}\\mathbf{e}=2\\times\\:\\frac{\\text{P}\\text{R}.\\text{R}\\text{C}}{\\text{P}\\text{R}+\\text{R}\\text{C}}$$\u003c/div\u003e \u003cdiv class=\"EquationNumber\"\u003e22\u003c/div\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Performance Comparison\u003c/h2\u003e \u003cp\u003eThe suggested method performance is compared with existing FEDSDM, Deep-CFL, and FL-IRL regarding recall, precision, accuracy, and f1score. The efficacy of the suggested approach is measured using the AR, and MIMIC-IV datasets.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigures \u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e show the accuracy comparison of suggested and existing frameworks. The accuracy of the suggested framework for the AR dataset is 99.01% over existing FEDSDM, Deep-CFL, and FL-IRL methods achieving a low accuracy of 97.21%, 95%, and 92.5% respectively. For the MIMIC-IV dataset, the suggested framework achieves an accuracy of 98.45% whereas the existing FEDSDM, Deep-CFL, and FL-IRL methods achieve a low accuracy of 96.02%, 96.56%, and 91.58% respectively.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe suggested framework has attained an overall accuracy of 99.01% on the AR dataset. The proposed method of validation and testing is illustrated through accuracy and loss plots in Figs.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e (a) and 6(b). These plots demonstrate the model's performance highlighting its effectiveness in patient health monitoring. The low loss values also reflect successful learning with minimal overfitting during the training process.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe suggested framework has attained an overall accuracy of 98.45% on the MIMIC-IV dataset. The classification of the validation and testing is illustrated through accuracy and loss plots of the proposed method in Figs.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e (a) and 6(b). These plots depict the model's performance highlighting its effectiveness in predicting health issues. The low loss values also reflect successful learning with minimal overfitting during the training process.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigures \u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e(a), and 7(b) indicate the performance evaluation of the FED-LIFE and existing approaches across AR, and MIMIC-IV datasets regards to precision, recall, and f1score. The recall, f1score, and precision of the suggested framework for the AR dataset are 97.1%, 98.01%, and 96.21% respectively. For the MIMIC-IV dataset, the suggested method achieves recall, f1score, and precision of 98.17%, 97.01%, and 95.33%. Overall, all the suggested approach outperforms the existing approaches.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e illustrates the delay in seconds for four methods namely FEDSDM, Deep-CFL, FL-IRL, and the proposed method across a patient count ranging from 10 to 50. The proposed method achieves the lowest delay of 22 seconds for 50 patients. Whereas the existing FEDSDM, Deep-CFL, and FL-IRL attain 45 seconds, 37 seconds, and 35 seconds for 50 patients. Across all patient counts the proposed method consistently achieves the lowest delay indicating more efficient data transmission.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e illustrates the performance based on throughput with existing and suggested frameworks evaluated across varying numbers of patients ranging from 10 to 50. The throughput for FEDSDM is 45 Mbps for 10 patients and 50 Mbps for 50 patients which indicates slow performance. Whereas Deep-CFL and FL-IRL achieve 48 Mbps and 50 Mbps for 10 patients. The Proposed method consistently achieves the highest throughput beginning at 53 Mbps for 10 patients and increasing linearly to 69 Mbps for 50 patients. Overall, the Proposed method demonstrates a clear advantage in scalability and efficiency, outperforming the other methods across all patient counts.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe number of patients and the execution time in milliseconds are displayed in Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003e. to provide a visual representation of the results. As the number of patients rises the execution time decreases for the proposed method. The FED-LIFE framework achieves a less execution time of 2.1 ms for 10 patients whereas, the FEDSDM, Deep-CFL, and FL-IRL techniques achieve 5.6 ms, 7.8 ms, and 9.8 ms respectively.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e illustrates the bandwidth consumption (in Kbps) of four methods, FEDSDM, Deep-CFL, FL-IRL, and the proposed method, which was analyzed across varying numbers of patients ranging from 100 to 500. For 100 patients the FEDSDM achieves 89.6 Kbps and Deep-CFL attains 90 Kbps. FL-IRL achieves 82 Kbps for 500 patients. The Proposed method consistently outperforms the other techniques, starting at 96.5 Kbps for 100 patients and 90 Kbps for 500 patients. Overall, the Proposed method demonstrates superior bandwidth efficiency, maintaining higher values across all patient counts than the other approaches.\u003c/p\u003e \u003c/div\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eIn this section, a novel FED-LIFE framework has been proposed for effective tracking and providing timely health services to patients. The suggested system integrates the Ghost_EliNet model for efficient and accurate classification of patient conditions, enabling precise categorization into urgency levels. The integration of Ghost_EliNet improves the system\u0026rsquo;s ability to detect and classify patient deterioration swiftly, thus minimizing the risk of delayed medical responses. By combining Fog computing, Red Deer Optimization, and Ghost_EliNet, the framework ensures continuous, reliable patient monitoring and timely interventions by addressing critical needs in ICU settings and improving overall patient outcomes. The accuracy of the suggested framework for the AR dataset is 99.01% over existing FEDSDM, Deep-CFL, and FL-IRL methods achieving a low accuracy of 97.21%, 95%, and 92.5%. For the MIMIC-IV dataset, the suggested approach achieves an accuracy of 98.45% whereas the existing FEDSDM, Deep-CFL, and FL-IRL methods achieve a low accuracy of 96.02%, 96.56%, and 91.58% respectively. The proposed method demonstrates the lowest delay of 8 seconds for 10 patients and 22 seconds for 50 patients. In future work, the system can be expanded to increase data security and privacy using blockchain technology.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthical approval:\u0026nbsp;\u003c/strong\u003eMy research guide reviewed and ethically approved this manuscript for publishing in this Journal.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions statement:\u0026nbsp;\u003c/strong\u003eThe authors confirm contribution to the paper as follows:Study conception and design: A. Jothi Soruba Thaya and N. Karthikeyan; Data collection: A. Jothi Soruba Thaya and N. Karthikeyan; Analysis and interpretation of results: A. Jothi Soruba Thaya and N. Karthikeyan; Draft manuscript preparation: A. Jothi Soruba Thaya and N. Karthikeyan. All authors reviewed the results and approved the final version of the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests:\u0026nbsp;\u003c/strong\u003eThis paper has no conflict of interest for publishing\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResearch funding:\u0026nbsp;\u003c/strong\u003eNo Financial support\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and material:\u003c/strong\u003e Data sharing is not applicable to this article as no new data were created or analyzed in this Research.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eHuman and Animal Rights:\u003c/strong\u003e This article does not contain any studies with human or animal subjects performed by any of the authors.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eInformed consent:\u0026nbsp;\u003c/strong\u003eI certify that I have explained the nature and purpose of this study to the above-named individual, and I have discussed the potential benefits of this study participation. The questions the individual had about this study have been answered, and we will always be available to address future questions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e: The author would like to express his heartfelt gratitude to the supervisor for his guidance and unwavering support during this research for his guidance and support\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eWu, Q., Chen, X., Zhou, Z., Zhang, J.: Fedhome: Cloud-edge based personalized federated learning for in-home health monitoring. IEEE Trans. Mob. Comput. \u003cb\u003e21\u003c/b\u003e(8), 2818\u0026ndash;2832 (2020)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNguyen, D.C., Pham, Q.V., Pathirana, P.N., Ding, M., Seneviratne, A., Lin, Z., Dobre, O., Hwang, W.J.: learning for smart healthcare: A survey. ACM Comput. Surv. (Csur). \u003cb\u003e55\u003c/b\u003e(3), 1\u0026ndash;37 (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSingh, C., Mishra, R., Gupta, H.P., Banga, G.: A federated learning-based patient monitoring system in internet of medical things. IEEE Trans. Comput. Social Syst. \u003cb\u003e10\u003c/b\u003e(4), 1622\u0026ndash;1628 (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBerghout, T., Benbouzid, M., Bentrcia, T., Lim, W.H., Amirat, Y.: Federated learning for condition monitoring of industrial processes: a review on fault diagnosis methods, challenges, and prospects. Electronics. \u003cb\u003e12\u003c/b\u003e(1), 158 (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlawadi, S., Kebande, V.R., Dong, Y., Bugeja, J., Persson, J.A., Olsson, C.M.: A federated interactive learning iot-based health monitoring platform. In European Conference on Advances in Databases and Information Systems, 235\u0026ndash;246 Cham: Springer International Publishing. (2021)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShaik, T., Tao, X., Higgins, N., Gururajan, R., Li, Y., Zhou, X., Acharya, U.R.: FedStack: Personalized activity monitoring using stacked federated learning. Knowl. Based Syst. \u003cb\u003e257\u003c/b\u003e, 109929 (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRahman, A., Hossain, M.S., Muhammad, G., Kundu, D., Debnath, T., Rahman, M., Khan, M.S.I., Tiwari, P., Band, S.S.: Federated learning-based AI approaches in smart healthcare: concepts, taxonomies, challenges and open issues. Cluster Comput. \u003cb\u003e26\u003c/b\u003e(4), 2271\u0026ndash;2311 (2023)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eArunan, A., Qin, Y., Li, X., Yuen, C.: A federated learning-based industrial health prognostics for heterogeneous edge devices using matched feature extraction. IEEE Trans. Autom. Sci. Eng. (2023)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMosaiyebzadeh, F., Pouriyeh, S., Parizi, R.M., Sheng, Q.Z., Han, M., Zhao, L., Sannino, G., Ranieri, C.M., Ueyama, J., Batista, D.M.: Privacy-enhancing technologies in federated learning for the internet of healthcare things: a survey. Electronics. \u003cb\u003e12\u003c/b\u003e(12), 2703 (2023)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChoi, G., Cha, W.C., Lee, S.U., Shin, S.Y.: Survey of Medical Applications of Federated Learning. Healthc. Inf. Res. \u003cb\u003e30\u003c/b\u003e(1), 3\u0026ndash;15 (2024)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhong, C., Sarkar, A., Manna, S., Khan, M.Z., Noorwali, A., Das, A., Chakraborty, K.: Federated learning-guided intrusion detection and neural key exchange for safeguarding patient data on the internet of medical things. Int. J. Mach. Learn. Cybernet., 1\u0026ndash;31 (2024)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePais, V., Rao, S., Muniyal, B., Yun, S.: FedICU: a federated learning model for reducing the medication prescription errors in intensive care units. Cogent Eng. \u003cb\u003e11\u003c/b\u003e(1), 2301150 (2024)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAli, M., Naeem, F., Tariq, M., Kaddoum, G.: Federated learning for privacy preservation in smart healthcare systems: A comprehensive survey. IEEE J. biomedical health Inf. \u003cb\u003e27\u003c/b\u003e(2), 778\u0026ndash;789 (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, J., Meng, Y., Ma, L., Du, S., Zhu, H., Pei, Q., Shen, X.: A federated learning based privacy-preserving smart healthcare system. IEEE Trans. Industr. Inf., \u003cb\u003e18\u003c/b\u003e(3) (2021)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLim, W.Y.B., Garg, S., Xiong, Z., Niyato, D., Leung, C., Miao, C., Guizani, M.: Dynamic contract design for federated learning in smart healthcare applications. IEEE Internet Things J. \u003cb\u003e8\u003c/b\u003e(23), 16853\u0026ndash;16862 (2020)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAkter, M., Moustafa, N., Lynar, T., Razzak, I.: Edge intelligence: Federated learning-based privacy protection framework for smart healthcare systems. IEEE J. Biomedical Health Inf. \u003cb\u003e26\u003c/b\u003e(12), 5805\u0026ndash;5816 (2022)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRajagopal, S.M., Supriya, M., Buyya, R.: FedSDM: Federated learning based smart decision making module for ECG data in IoT integrated Edge\u0026ndash;Fog\u0026ndash;Cloud computing environments. Internet Things. \u003cb\u003e22\u003c/b\u003e, 100784 (2023)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGong, W., Cao, L., Zhu, Y., Zuo, F., He, X., Zhou, H.: Federated inverse reinforcement learning for smart icus with differential privacy. IEEE Internet Things J. \u003cb\u003e10\u003c/b\u003e(21), 19117\u0026ndash;19124 (2023)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlam, M.U., Rahmani, R.: Fedsepsis: A federated multi-modal deep learning-based internet of medical things application for early detection of sepsis from electronic health records using raspberry pi and jetson nano devices. Sensors. \u003cb\u003e23\u003c/b\u003e(2), 970 (2023)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRakhmiddin, R., Lee, K.: Federated Learning for Clinical Event Classification Using Vital Signs Data. Multimodal Technol. Interact. \u003cb\u003e7\u003c/b\u003e(7), 67 (2023)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDi Napoli, C., Paragliola, G., Ribino, P., Serino, L.: Balancing Uneven Knowledge of Hospital Nodes for ICU Patients Diagnosis through Federated Learning (2023)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMuazu, T., Yingchi, M., Muhammad, A.U., Ibrahim, M., Samuel, O., Tiwari, P.: Iomt: A medical resource management system using edge empowered blockchain federated learning. IEEE Trans. Netw. Serv. Manage. (2023)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNguyen, T.N., Yang, H.J., Kho, B.G., Kang, S.R., Kim, S.H.: Explainable Deep Contrastive Federated Learning System for Early Prediction of Clinical Status in-Intensive Care Unit. IEEE Access. (2024)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eConsul, P., Budhiraja, I., Arora, R., Garg, S., Choi, B.J., Hossain, M.S.: Federated reinforcement learning based task offloading approach for MEC-assisted WBAN-enabled IoMT. Alexandria Eng. J. \u003cb\u003e86\u003c/b\u003e, 56\u0026ndash;66 (2024)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePan, W., Xu, Z., Rajendran, S., Wang, F.: An adaptive federated learning framework for clinical risk prediction with electronic health records from multiple hospitals. Patterns, \u003cb\u003e5\u003c/b\u003e(1) (2024)\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":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"cluster-computing","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Cluster Computing](https://www.springer.com/journal/10586)","snPcode":"10586","submissionUrl":"https://submission.nature.com/new-submission/10586/3","title":"Cluster Computing","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Red Deer Optimization, Federated Learning, Internet of Medical Things, Intensive Care Unit, Patient Monitoring","lastPublishedDoi":"10.21203/rs.3.rs-6177554/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6177554/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eIntensive Care Unit (ICU) patient monitoring plays a vital role in ensuring the safety and well-being of critically ill patients by providing continuous and real-time insights into their health status. The integration of Internet of Medical Things (IoMT) devices in ICU including wearable sensors and remote monitoring tools, enables the seamless collection and transmission of patient data, allowing for real-time tracking of vital signs. Federated learning (FL) enhances this process by utilizing decentralized patient data to improve model generalization while maintaining data privacy. However, FL-based ICU monitoring faced challenges including high delays in decision-making due to centralized data processing, and significant execution time caused by the need to transfer large volumes of patient data. This research proposes a novel FEDerated learning-based LIFE saving ICU system (FED-LIFE) for effective tracking and providing timely health services to patients. The FED-LIFE system initially trains local models utilizing Ghostnet combined Enhanced LinkNet (Ghost_EliNet) which combines GhostNet and LinkNet, for tuning the Ghost_EliNet model a Red Deer Optimization (RDO) algorithm is employed for accurate service allocation. The suggested approach is implemented in Python programming. The efficacy of the developed approach is evaluated utilizing several metrics namely Precision, recall, f1-score, accuracy, delay, throughput, and execution time. The proposed method achieves the lowest delay of 22 seconds for 50 patients. Whereas the existing FEDSDM, Deep-CFL, and FL-IRL attain 45 seconds, 37 seconds, and 35 seconds for 50 patients respectively.\u003c/p\u003e","manuscriptTitle":"FED-LIFE: Ghost LinkNet Enabled Federated Learning for Anomaly Detection in Smart Intensive Care Unit based on IOMT","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-05-13 01:38:03","doi":"10.21203/rs.3.rs-6177554/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-09-11T03:05:12+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-07-04T07:14:15+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"302001774893844852794643598853198551392","date":"2025-07-01T09:14:29+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-06-13T04:28:43+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"285959321877047211336897116012541659617","date":"2025-06-12T07:46:34+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-05-06T06:01:31+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-03-27T09:52:41+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-03-27T06:42:40+00:00","index":"","fulltext":""},{"type":"submitted","content":"Cluster Computing","date":"2025-03-07T10:40:46+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"cluster-computing","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Cluster Computing](https://www.springer.com/journal/10586)","snPcode":"10586","submissionUrl":"https://submission.nature.com/new-submission/10586/3","title":"Cluster Computing","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"c9ad2d9c-f1c4-41f8-ba5f-d446586d9845","owner":[],"postedDate":"May 13th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-02-08T14:53:08+00:00","versionOfRecord":[],"versionCreatedAt":"2025-05-13 01:38:03","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6177554","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6177554","identity":"rs-6177554","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}