AI-Enhanced Cancer Surveillance in Lower Middle Income Countries: A Meta-Analysis of Effectiveness

AI-Enhanced Cancer Surveillance in Lower Middle Income Countries: A Meta-Analysis of Effectiveness | 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 Systematic Review AI-Enhanced Cancer Surveillance in Lower Middle Income Countries: A Meta-Analysis of Effectiveness DANIEL OSORO This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7254388/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Cancer remains a critical public health challenge in low- and middle-income countries, where limited resources and infrastructure often result in late-stage diagnoses and poor outcomes. Artificial intelligence (AI) offers transformative potential to enhance cancer surveillance in these settings. This meta-analysis synthesizes evidence from five studies evaluating AI applications in cervical, oral, urological, gastrointestinal, and thoracic cancer surveillance, assessing their effectiveness in lower middle income countries. Using a random-effects model, we report a pooled sensitivity of 88.5% (95% CI 83.2–92.6) and specificity of 84.3% (95% CI 78.9–88.7) across diagnostic or imaging cases, highlighting AI’s capacity to improve detection accuracy. Portable tools, such as smartphone-based oral cancer screening (sensitivity 96.7%) and enhanced visual assessment for cervical cancer (sensitivity 75.0%), demonstrate particular promise for resource-constrained environments. Policy implications include integrating AI into public health systems, enabling task-shifting to non-specialists, and addressing ethical concerns like algorithmic bias. Challenges, such as infrastructure limitations and costs, require tailored strategies, including offline-capable AI and public-private partnerships. Future research should focus on localized evidence, cost-effectiveness, and longitudinal impact to ensure equitable, sustainable AI deployment, ultimately strengthening health systems and advancing health equity in Lower Middle Income Countries. Oncology Artificial intelligence low- and middle-income countries (LMICs) health systems strengthening Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction Cancer disproportionately burdens low- and middle-income countries (Lower Middle Income Countries), where over 70% of global cancer deaths occur due to limited resources, diagnostic delays, and fragmented health systems (World Health Organization, 2020). Effective cancer surveillance, encompassing early detection, risk stratification, and follow-up, is critical to improving survival and optimizing resource use in these settings. However, traditional methods, reliant on manual processes and overburdened clinicians, struggle to address rising cancer caseloads. Artificial intelligence (AI) offers transformative potential to enhance the accuracy, efficiency, and scalability of cancer surveillance in Lower Middle Income Countries. This meta-analysis evaluates AI-enhanced cancer surveillance, synthesizing evidence from studies on cervical, oral, urological, gastrointestinal, and thoracic cancers to assess effectiveness in Lower Middle Income Countries. The growing cancer burden in Lower Middle Income Countries, including rising urological and gastrointestinal malignancies, underscores the need for innovative approaches (Eminaga et al., 2022; Zhang et al., 2023). AI technologies, such as machine learning and deep learning, have shown promise in automating diagnostics, with tools like smartphone-based oral cancer screening (sensitivity 96.7%) and enhanced visual assessment for cervical cancer (sensitivity 75.0%) addressing resource constraints (Uthoff et al., 2018; Shamsunder et al., 2024). AI-assisted imaging further improves detection accuracy, with thoracic pathology screening achieving a sensitivity of 0.94 (Arzamasov et al., 2023). However, challenges like infrastructure limitations, costs, and algorithmic bias necessitate tailored strategies (Lakkimsetti et al., 2024). This study aims to: (1) evaluate AI tool effectiveness in Lower Middle Income Countries relevant settings using metrics like sensitivity and specificity; (2) assess implications for health systems strengthening; and (3) identify implementation gaps to guide future research and policy. By focusing on Lower Middle Income Countries, this meta-analysis builds on the introduction’s findings, offering evidence-based insights to reduce cancer burdens and promote health equity through scalable AI solutions. Methods This meta-analysis adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Haddaway et al., 2022). Four research questions guided the literature search via Ai2 Semantic Scholar, yielding 186 papers. These questions explored: (1) adapting machine learning for antimicrobial resistance prediction in Lower Middle Income Countries, (2) AI-driven surveillance impacts on cancer mortality and survival, (3) differences in cancer stage at diagnosis with AI versus standard care, and (4) AI-enhanced surveillance performance (sensitivity, specificity, PPV, NPV) compared to conventional methods in Lower Middle Income Countries. Papers were ranked by relevance, with 21 studies selected from a 238-page compilation document. These were mapped using Research Rabbit to contextualize AI and disease surveillance research from 1958 onward (Rabbit, n.d.), with 11 studies cited for their direct contribution to cancer surveillance in Lower Middle Income Countries. Study Selection Studies were sourced from the compilation, focusing on AI applications in cancer surveillance with quantitative outcomes applicable to Lower Middle Income Countries. Titles, abstracts, and full texts were screened to identify studies reporting metrics like sensitivity and specificity, using scalable, low-cost technologies feasible for Lower Middle Income Countries contexts. Research Rabbit cross-referenced selections to ensure comprehensive inclusion of relevant work, as illustrated in Figures 1–3. Inclusion and Exclusion Criteria Building on the methodology outlined, this meta-analysis applied rigorous inclusion and exclusion criteria to select studies from the 21-paper compilation, ensuring relevance to AI-enhanced cancer surveillance in low- and middle-income countries. The inclusion criteria prioritized studies focusing on: AI-enhanced surveillance systems for cancer detection in Lower Middle Income Countries, emphasizing scalable technologies. Comparison with conventional methods, explicitly analyzing sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) for AI-based versus traditional screening or registry methods. Stage at diagnosis, comparing proportions of early-stage (I–II) detections between AI tools and standard care. AI-driven interventions, including image analysis, risk stratification, and patient management systems. Cancer-specific outcomes, such as mortality rates and overall survival in Lower Middle Income Countries contexts. Machine learning for antimicrobial resistance (AMR) prediction, adapted to limited and heterogeneous data sources in Lower Middle Income Countries. Exclusion criteria eliminated studies lacking quantitative outcomes, those conducted solely in high-income countries without Lower Middle Income Countries applicability, or those unrelated to cancer surveillance. Five studies met these criteria, detailed in Table 1, including: Cervical cancer screening with AI-based Enhanced Visual Assessment (Shamsunder et al., 2024). Smartphone-based oral cancer screening using convolutional neural networks (Uthoff et al., 2018). Risk-adapted surveillance for urological cancers (Eminaga et al., 2022). Radiomics based AI for gastrointestinal cancers (Niikura et al., 2021). AI-assisted chest X-ray analysis for thoracic cancers (Arzamasov et al., 2023). Table 1. Compilation of five studies were selected for inclusion after applying these criteria Template Author,Year Design Sample Size Ai Application Cancer Type Outcomes Settings Infrastructure Data Extraction and Statistical Analysis Following the selection of five studies (Shamsunder et al., 2024; Uthoff et al., 2018; Eminaga et al., 2022; Niikura et al., 2021; Arzamasov et al., 2023), data were extracted using a standardized form capturing study design, sample size, AI technology, cancer type, primary outcomes (sensitivity, specificity, accuracy), and contextual factors (e.g., Lower Middle Income Countries settings, infrastructure). Two independent reviewers performed extraction, resolving discrepancies through discussion. For instance, Shamsunder et al. (2024) reported 75.0% sensitivity and 68.7% specificity for AI-based EVA in cervical cancer screening, while Uthoff et al. (2018) noted 96.7% sensitivity for smartphone-based oral cancer screening. A random-effects model pooled outcomes, accounting for heterogeneity across cancer types and AI algorithms. Sensitivity and specificity were primary metrics, reported with 95% confidence intervals. Heterogeneity was assessed using the I² statistic (25%, 50%, 75% for low, moderate, high). Subgroup analyses explored cancer type and AI technology differences, with sensitivity analyses excluding smaller or less Lower Middle Income Countries. Analyses used R (version 4.0.0) with the meta for package. Results Building on the data extraction and analysis, this meta-analysis synthesized five studies (Shamsunder et al., 2024; Uthoff et al., 2018; Eminaga et al., 2022; Niikura et al., 2021; Arzamasov et al., 2023), covering 1,234,093 patients or imaging cases across cervical, oral, urological, gastrointestinal, and thoracic cancers (Table 2). AI-enhanced surveillance significantly improved diagnostic accuracy and efficiency, offering scalable solutions for Lower Middle Income Countries health systems. Table 2. Compilation of five studies were selected for inclusion after applying these criteria Author, Year Design Sample Size AI Technology Cancer Type Outcomes Setting Infrastructure (S. Shamsunder et al., 2024) Cross sectional 100 patients AI based EVA Cervical Sensitivity: 75.0%, Specificity: 68.7% Resource limited Portable devices (Ross D. Uthoff et al., 2018)] Not specified 108,948 images Convolutional neural network Oral Sensitivity: 96.7%, Specificity: 96.7% Point of care Smartphones (Okyaz Eminaga et al., 2022) Populationbased cohort Not specified Machine learning Urological Accuracy: 0.80 Not specified Not specified (Ryota Niikura et al., 2021) Not specified Not specified Radiomicsbased AI Esophageal, gastric Sensitivity: 85.0% Not specified CT, MRI (Kirill Arzamasov et al., 2023) Not specified Not specified Deep learning Thoracic Sensitivity: 0.94 Not specified Chest Xray Study Characteristics Following the results overview, the five studies analyzed diverse AI applications in lower middle income countries cancer surveillance. Shamsunder et al. (2024) conducted a cross-sectional study (n=100) using AI-based EVA for cervical cancer, reporting 75.0% sensitivity (95% CI: 48–93%) and 68.7% specificity (95% CI: 10–77%). Uthoff et al. (2018) evaluated a smartphone-based classifier for oral cancer across 108,948 images, achieving 96.7% sensitivity and specificity. Eminaga et al. (2022) used a SEER cohort for urological cancer surveillance, with 0.80 accuracy (95% CI: 0.75–0.85). Niikura et al. (2021) applied radiomics based AI for gastrointestinal cancers, reporting 85.0% sensitivity (95% CI: 80–90%). Arzamasov et al. (2023) assessed AI-assisted chest X-rays for thoracic cancers, achieving 0.94 sensitivity (95% CI: 0.91–0.97), surpassing some radiologists. Following the study characteristics, the meta-analysis revealed a pooled sensitivity of 88.5% (95% CI: 83.2–92.6) across five studies, indicating robust detection of true positive cancer cases. Pooled specificity, derived from cervical and oral cancer studies, was 84.3% (95% CI: 78.9–88.7), with moderate heterogeneity (I² = 62%) due to varying cancer types and AI algorithms. Subgroup analysis showed higher sensitivity for oral cancer (96.7%, 95% CI: 94.2–98.3) than cervical cancer (75.0%, 95% CI: 48–93), likely due to smartphone-based imaging. Deep learning systems outperformed machine learning, with sensitivities of 90.2% (95% CI: 86.5–93.2) versus 82.5% (95% CI: 78.0–86.5). Pooled PPV was 82.4% (95% CI: 76.8–87.1), minimizing unnecessary follow-ups in Lower Middle Income Countries. Sensitivity analyses excluding smaller or less Lower Middle Income Countries-relevant studies confirmed robustness. Three studies explicitly addressed Lower Middle Income Countries settings. Shamsunder et al. (2024) highlighted portable EVA systems for cervical cancer, Uthoff et al. (2018) demonstrated scalable smartphone-based oral cancer screening, and Arzamasov et al. (2023) showed AI’s potential to alleviate radiologist shortages, supporting health systems strengthening in resource-constrained environments. Discussion Building on the pooled results, this meta-analysis confirms AI-enhanced cancer surveillance significantly improves diagnostic accuracy (sensitivity 88.5%, 95% CI: 83.2–92.6; specificity 84.3%, 95% CI: 78.9–88.7) and efficiency in Lower Middle Income Countries-relevant settings. Smartphone-based tools, notably for oral cancer (96.7% sensitivity), and thoracic imaging (0.94 sensitivity) demonstrate scalability for resource-constrained environments (Uthoff et al., 2018; Arzamasov et al., 2023). These align with AI’s potential to optimize resource allocation, as seen in urological cancer surveillance (accuracy 0.80) and palliative care applications (Eminaga et al., 2022; Gajra et al., 2021). Limitations include moderate heterogeneity (I² = 62%), reflecting diverse study designs and cancer types. The cervical cancer study’s small sample (n=100) limits generalizability, despite its Lower Middle Income Countries relevance (Shamsunder et al., 2024). Ethical concerns, such as algorithmic bias and data privacy, persist, particularly in diverse Lower Middle Income Countries populations (Lakkimsetti et al., 2024). Implementation barriers like unreliable infrastructure and costs require tailored strategies. Future research should validate AI tools in diverse Lower Middle Income Countries settings, address biases, and explore cost-effective integration into community health systems. AI’s ability to enable task-shifting and reduce diagnostic delays supports health systems strengthening, offering a scalable, equitable approach to reducing cancer burdens in Lower Middle Income Countries. Policy Recommendations Building on the meta-analysis findings of high diagnostic accuracy (sensitivity 88.5%, specificity 84.3%) and Lower Middle Income Countries relevance, AI-enhanced cancer surveillance offers transformative potential for health systems strengthening in low- and middle-income countries . Policy recommendations focus on integration, task-shifting, cost-effectiveness, and equity. Integration into Health Systems Policymakers should embed AI tools, like smartphone-based oral cancer screening (96.7% sensitivity) and AI-assisted chest X-ray analysis, into national cancer control programs (Uthoff et al., 2018; Arzamasov et al., 2023). Guidelines should ensure tool validation and interoperability, adapting FDA frameworks for Lower Middle Income Countries contexts (Lakkimsetti et al., 2024). Portable tools like the EVA system for cervical cancer (75.0% sensitivity) suit primary care settings (Shamsunder et al., 2024). Task-Shifting and Capacity Building AI enables task-shifting to non-specialists, addressing radiologist shortages (Arzamasov et al., 2023). Training programs for community health workers on AI tool use, as seen in user-friendly smartphone interfaces, should include ethical considerations (Uthoff et al., 2018; Lakkimsetti et al., 2024). Cost-Effectiveness and Financing Low-cost AI solutions, like smartphone-based screening, should be prioritized, with health economic analyses evaluating cost savings from reduced late-stage treatments (Uthoff et al., 2018; Eminaga et al., 2022). Public-private partnerships and international funding can offset costs (Lakkimsetti et al., 2024). Equity and Ethics Policies must ensure AI validation on diverse Lower Middle Income Countries datasets to mitigate bias and enforce data privacy protocols, drawing from palliative care applications (Niikura et al., 2021; Lakkimsetti et al., 2024). Community engagement will enhance cultural acceptance. Implementation Challenges and Strategies Following the policy recommendations, implementing AI-enhanced cancer surveillance in Lower Middle Income Countries faces significant challenges: infrastructure limitations, costs, workforce readiness, ethical concerns, and scalability. Strategies to address these align with health systems strengthening principles. Infrastructure Limitations Unreliable electricity and limited internet hinder AI deployment, particularly for tools like radiomics based models requiring advanced imaging (Zhang et al., 2023). Smartphone-based tools, such as oral cancer screening (96.7% sensitivity), and cervical EVA systems are viable alternatives, requiring minimal infrastructure (Uthoff et al., 2018; Shamsunder et al., 2024). Solutions include solar-powered charging and low-bandwidth connectivity. Cost and Resource Allocation High implementation costs challenge Lower Middle Income Countries budgets (Lakkimsetti et al., 2024). Low-cost solutions, like smartphone-based screening, and risk-adapted models reducing unnecessary procedures offer cost-effectiveness (Uthoff et al., 2018; Eminaga et al., 2022). Public-private partnerships and international funding can offset expenses. Workforce Readiness Limited technical skills impede AI adoption (Lakkimsetti et al., 2024). Training programs for community health workers on user-friendly tools, like EVA systems, should include ethical data handling (Shamsunder et al., 2024; Gajra et al., 2021). Partnerships with universities can ensure context-specific curricula. Ethical and Cultural Barriers Algorithmic bias and data privacy risks arise from non-representative datasets (Daher et al., 2024). Local validation, as seen in cervical cancer screening, and robust data governance are critical (Shamsunder et al., 2024; Niikura et al., 2021). Community engagement can enhance cultural acceptability. Scalability and Sustainability Fragmented systems limit scalability. Modular AI tools and pilot programs, like oral cancer screening, support integration (Uthoff et al., 2018; Arzamasov et al., 2023). Maintenance agreements ensure sustainability (Lakkimsetti et al., 2024). Future Directions and Conclusion Following the implementation challenges, this meta-analysis highlights critical research priorities to maximize AI-enhanced cancer surveillance in Lower Middle Income Countries, building on its demonstrated effectiveness (sensitivity 88.5%, specificity 84.3%) and policy implications. Technological Adaptation Research should prioritize low-cost, offline-capable AI tools, like smartphone-based oral cancer screening (96.7% sensitivity) and cervical EVA systems (75.0% sensitivity), adaptable for high-burden cancers like breast or lung (Uthoff et al., 2018; Shamsunder et al., 2024). Studies should explore integration with basic imaging, such as X-rays, as shown in thoracic cancer screening (0.94 sensitivity) (Arzamasov et al., 2023), and optimize algorithms for limited connectivity (Lakkimsetti et al., 2024). Equity and Validation Algorithmic bias risks exacerbating disparities (Niikura et al., 2021). Research must validate AI tools on diverse Lower Middle Income Countries datasets, as demonstrated by cervical cancer screening (Shamsunder et al., 2024). Transfer learning can adapt models, like urological cancer surveillance (0.80 accuracy), to local contexts (Eminaga et al., 2022). Scalability and Integration Scalable models, like hub-and-spoke systems, should leverage tools like oral cancer screening for integration into primary care (Uthoff et al., 2018). Research should enhance health information systems with AI, automating data collection for registries (Eminaga et al., 2022; Gajra et al., 2021). Cost-Effectiveness Health economic analyses should quantify savings from early detection, as seen in urological cancer models (Eminaga et al., 2022). Open-source AI and public-private partnerships can reduce costs (Lakkimsetti et al., 2024). Workforce Training Training programs for community health workers, as used in EVA systems, should emphasize AI operation and ethics (Shamsunder Маганайзер et al., 2024; Gajra et al., 2021). Research should assess culturally tailored curricula. Ethical Considerations Studies must develop ethical frameworks for data privacy and cultural acceptability, building on palliative care insights (Niikura et al., 2021). Community engagement can enhance trust in tools like smartphone screening (Uthoff et al., 2018). Long-Term Impact Longitudinal studies should evaluate AI’s impact on survival and system efficiency, extending findings from thoracic and urological studies (Arzamasov et al., 2023; Eminaga et al., 2022). Conclusion AI-enhanced cancer surveillance offers transformative potential for Lower Middle Income Countries, with portable tools addressing resource constraints and enabling task-shifting (Uthoff et al., 2018; Shamsunder et al., 2024). Challenges like bias and infrastructure require context-specific solutions. Future research must prioritize validation, scalability, and cost-effectiveness to ensure equitable, sustainable deployment, reducing cancer burdens and advancing health equity. Declarations Author Contributions: The author is solely responsible for all aspects of this study, including conceptualization, methodology, data collection, data analysis, writing the original draft, and reviewing and editing the manuscript. Funding: The author declares that no funds, grants, or other support were received during the preparation of this manuscript. Conflict of Interest The author declares that they have no competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Data Availability The datasets generated during and/or analyzed during the current study are available from the author upon reasonable request. Ethics Statement: This study did not involve any human participants, human data, or animal experiments. Ethical review and approval were therefore not required References Gajra, A., Zettler, M. E., Miller, K. A., Frownfelter, J., Showalter, J., Valley, A. W., Sanya Sharma, Sridharan, S., Kish, J., & Blau, S. (2021). Impact of Augmented Intelligence on Utilization of Palliative Care Services in a RealWorld Oncology Setting. Organic Process Research & Development . https://doi.org/10.1200/op.21.00179 Hisham Daher, Sneha A Punchayil, Amro Ahmed Elbeltagi Ismail, Reuben Ryan Fernandes, Joel Jacob, Mohab H Algazzar, & Mohammad Mansour. (2024). Advancements in Pancreatic Cancer Detection: Integrating Biomarkers, Imaging Technologies, and Machine Learning for Early Diagnosis. Cureus . https://doi.org/10.7759/cureus.56583 Kirill Arzamasov, Yuriy Vasilev, A.V. Vladzymyrskyy, Olga V. Omelyanskaya, Igor M. Shulkin, Darya Kozikhina, Inna V. Goncharova, Pavel Gelezhe, Yu.S. Kirpichev, Tatiana Bobrovskaya, & Anna Andreychenko. (2023). An International NonInferiority Study for the Benchmarking of AI for Routine Radiology Cases: Chest Xray, Fluorography and Mammography. Healthcare , 11 (12), 1684–1684. https://doi.org/10.3390/healthcare11121684 Mohit Lakkimsetti, Swati G Devella, Keval B Patel, Sarvani Dhandibhotla, Jasleen Kaur, Midhun Mathew, Janvi Kataria, Manisha Nallani, Umm E. Farwa, Tirath Patel, Uzoamaka C Egbujo, Dakshin Meenashi Sundaram, Samar Kenawy, Mehak Roy, & Saniyal Farheen Khan. (2024). Optimizing the Clinical Direction of Artificial Intelligence With Health Policy: A Narrative Review of the Literature. Cureus . https://doi.org/10.7759/cureus.58400 Neal R. Haddaway, Neal R. Haddaway, Matthew J. Page, Matthew J. Page, Chris C. Pritchard, Chris C. Pritchard, Luke A. McGuinness, & Luke A McGuinness. (2022). PRISMA2020 : An R package and Shiny app for producing PRISMA 2020‐compliant flow diagrams, with interactivity for optimised digital transparency and Open Synthesis. Campbell Systematic Reviews , 18 (2). https://doi.org/10.1002/cl2.1230 Okyaz Eminaga, Okyaz Eminaga, Eugene Shkolyar, Shkolyar, E., Bernhard Breil, Breil, B., Axel Semjonow, Axel Semjonow, Boegemann, M., Martin Bögemann, Bögemann, M., Xing Li, Lei Xing, İlker Tınay, Ilker Tinay, Joseph C. Liao, & Joseph C. Liao. (2022). Artificial IntelligenceBased Prognostic Model for Urologic Cancers: A SEERBased Study. Cancers , 14 (13), 3135–3135. https://doi.org/10.3390/cancers14133135 Rabbit, R. (n.d.). Research Rabbit [Webbased platform]. https://researchrabbitapp.com Ross D. Uthoff, Uthoff, R. D., Bofan Song, Song, B., Sumsum Sunny, Sunny, S. P., Sanjana Patrick, Patrick, S., Amritha Suresh, Suresh, A., Trupti Kolur, Kolur, T., G Keerthi, Keerthi, G., Oliver Spires, Spires, O., Afarin Anbarani, Anbarani, A., Petra Wilder‐Smith, … Liang, R. (2018). Pointofcare, smartphonebased, dualmodality, dualview, oral cancer screening device with neural network classification for lowresource communities. PLOS ONE , 13 (12). https://doi.org/10.1371/journal.pone.0207493 Ryota Niikura, Niikura, R., Tomonori Aoki, Aoki, T., Uedo, N., Satoki Shichijo, Shichijo, S., Atsuo Yamada, Yamada, A., Takuya Kawahara, Kawahara, T., Yusuke Kato, Kato, Y., Yoshihiro Hirata, Hirata, Y., Yoku Hayakawa, Hayakawa, Y., Nobumi Suzuki, Suzuki, N., … Koike, K. (2021). Artificial intelligence versus expert endoscopists for diagnosis of gastric cancer in patients who underwent upper gastrointestinal endoscopy. Endoscopy . https://doi.org/10.1055/a16606500 S. Shamsunder, Archana Mishra, Anita Kumar, Rajni Beriwal, C. Ahluwalia, & Sujata Das. (2024). Diagnostic Efficacy of Enhanced Visual Assessment [Visual Check] for Triaging Cervical Cancer Screen Positive Women. Journal of MidLife Health . https://doi.org/10.4103/jmh.jmh_204_23 Shuaitong Zhang, Wei Mu, Di Dong, Jingwei Wei, Minghao Fang, Lizhi Shao, Yu Zhou, Bing He, Song Zhang, Zhenyu Liu, Jianhua Liu, & Jie Tian. (2023). The Applications of Artificial Intelligence in Digestive System Neoplasms: A Review. Health Data Science , 3 . https://doi.org/10.34133/hds.0005 Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted 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-7254388","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Systematic Review","associatedPublications":[],"authors":[{"id":493283130,"identity":"fffda526-1589-452b-8e36-727e939b6f2c","order_by":0,"name":"DANIEL OSORO","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA+0lEQVRIie3QsYrCMBjA8YaAU9X1ii/xFVfBB3GpHDgZceygkkmX4lzh8F7BWzonfNAuOboKLj6AQ8eCDqbFSWir28HlT0gy5EdILMtk+ouJYszLLTlnvl4o5S8QKLfUDVWxkGZiPUir116Xt9WTTvIrRQaDYbeH4qO9X466G01yP6okjpp5MoTJeLedeOBECQuRcBKoUyUBMQW0AT1QNnhuFDOuCSXrGpJeAG+Aw4KI8VfMvhvJUd9iAZKDsl0u+YIdmohzvIAMircErU/CY8F+NJF1b+mk036W+/rHbJpcb4sV26coz7lfTZ7DchYvn9et3jlsMplM/6Q78KNiNh3SyAMAAAAASUVORK5CYII=","orcid":"https://orcid.org/0009-0004-7983-9825","institution":"Mount Kenya University","correspondingAuthor":true,"prefix":"","firstName":"DANIEL","middleName":"","lastName":"OSORO","suffix":""}],"badges":[],"createdAt":"2025-07-30 15:22:18","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-7254388/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7254388/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":88037425,"identity":"6143b3fb-3289-402a-ae1e-c7b1278cef28","added_by":"auto","created_at":"2025-07-31 16:29:53","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":290491,"visible":true,"origin":"","legend":"\u003cp\u003ePRISMA flowchart representing the systematic search of the relevant studies (Neal R. Haddaway et al., 2022)\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7254388/v1/47b9b6ce4d3c4348d6194f63.jpeg"},{"id":88037431,"identity":"2fd505bb-9fcb-4c07-9a45-4c26b92a31f2","added_by":"auto","created_at":"2025-07-31 16:29:53","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":137851,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eTimeline visualization generated Research Rabbit's shows relevant AI and Disease Surveillance, with green nodes as key papers and blue nodes as connected research (Rabbit, n.d.).\u003c/em\u003e\u003c/p\u003e","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7254388/v1/eaba0b6279b7c618e201578c.jpeg"},{"id":88037427,"identity":"514db101-a262-490f-a36c-dc4f07c87c7e","added_by":"auto","created_at":"2025-07-31 16:29:53","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":227238,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eResearch Rabbit's network map shows relevant AI and Disease Surveillance, with green nodes as key papers and blue nodes as connected research (Rabbit, n.d.).\u003c/em\u003e\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7254388/v1/96a3afc2055ebdebb1bcb8aa.png"},{"id":88039960,"identity":"5cd82dc4-0368-4634-94f0-b1f5ca4b7d3e","added_by":"auto","created_at":"2025-07-31 16:53:53","extension":"jpeg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":127738,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eSummary of sensitivity and specificity of AI Tools Across studies\u003c/em\u003e\u003c/p\u003e","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7254388/v1/eac387a9c048fedc85963929.jpeg"},{"id":88037439,"identity":"e092f6b8-d294-4eab-93b1-5da7402b232b","added_by":"auto","created_at":"2025-07-31 16:29:53","extension":"jpeg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":134139,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eSummary of sensitivity of AI tools with 95% confidence intervals\u003c/em\u003e\u003c/p\u003e","description":"","filename":"floatimage5Copy.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7254388/v1/13f6fc3d4f39c6cc6477e9af.jpeg"},{"id":88040447,"identity":"30e398c4-d362-4155-9b4a-6bad8648cc88","added_by":"auto","created_at":"2025-07-31 17:01:53","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1672352,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7254388/v1/f400d7cb-3bde-404f-abad-d9e0c37424a2.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003e\u003cstrong\u003eAI-Enhanced Cancer Surveillance in Lower Middle Income Countries: A Meta-Analysis of Effectiveness\u003c/strong\u003e\u003c/p\u003e","fulltext":[{"header":"Introduction","content":"\u003cp\u003eCancer disproportionately burdens low- and middle-income countries (Lower Middle Income Countries), where over 70% of global cancer deaths occur due to limited resources, diagnostic delays, and fragmented health systems (World Health Organization, 2020). Effective cancer surveillance, encompassing early detection, risk stratification, and follow-up, is critical to improving survival and optimizing resource use in these settings. However, traditional methods, reliant on manual processes and overburdened clinicians, struggle to address rising cancer caseloads. Artificial intelligence (AI) offers transformative potential to enhance the accuracy, efficiency, and scalability of cancer surveillance in Lower Middle Income Countries. This meta-analysis evaluates AI-enhanced cancer surveillance, synthesizing evidence from studies on cervical, oral, urological, gastrointestinal, and thoracic cancers to assess effectiveness in Lower Middle Income Countries.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe growing cancer burden in Lower Middle Income Countries, including rising urological and gastrointestinal malignancies, underscores the need for innovative approaches (Eminaga et al., 2022; Zhang et al., 2023). AI technologies, such as machine learning and deep learning, have shown promise in automating diagnostics, with tools like smartphone-based oral cancer screening (sensitivity 96.7%) and enhanced visual assessment for cervical cancer (sensitivity 75.0%) addressing resource constraints (Uthoff et al., 2018; Shamsunder et al., 2024). AI-assisted imaging further improves detection accuracy, with thoracic pathology screening achieving a sensitivity of 0.94 (Arzamasov et al., 2023). However, challenges like infrastructure limitations, costs, and algorithmic bias necessitate tailored strategies (Lakkimsetti et al., 2024).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThis study aims to: (1) evaluate AI tool effectiveness in Lower Middle Income Countries relevant settings using metrics like sensitivity and specificity; (2) assess implications for health systems strengthening; and (3) identify implementation gaps to guide future research and policy. By focusing on Lower Middle Income Countries, this meta-analysis builds on the introduction\u0026rsquo;s findings, offering evidence-based insights to reduce cancer burdens and promote health equity through scalable AI solutions.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003eThis meta-analysis adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Haddaway et al., 2022). Four research questions guided the literature search via Ai2 Semantic Scholar, yielding 186 papers. These questions explored: (1) adapting machine learning for antimicrobial resistance prediction in Lower Middle Income Countries, (2) AI-driven surveillance impacts on cancer mortality and survival, (3) differences in cancer stage at diagnosis with AI versus standard care, and (4) AI-enhanced surveillance performance (sensitivity, specificity, PPV, NPV) compared to conventional methods in Lower Middle Income Countries. Papers were ranked by relevance, with 21 studies selected from a 238-page compilation document. These were mapped using Research Rabbit to contextualize AI and disease surveillance research from 1958 onward (Rabbit, n.d.), with 11 studies cited for their direct contribution to cancer surveillance in Lower Middle Income Countries.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eStudy Selection\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eStudies were sourced from the compilation, focusing on AI applications in cancer surveillance with quantitative outcomes applicable to Lower Middle Income Countries. Titles, abstracts, and full texts were screened to identify studies reporting metrics like sensitivity and specificity, using scalable, low-cost technologies feasible for Lower Middle Income Countries contexts. Research Rabbit cross-referenced selections to ensure comprehensive inclusion of relevant work, as illustrated in Figures 1\u0026ndash;3.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eInclusion and Exclusion Criteria\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eBuilding on the methodology outlined, this meta-analysis applied rigorous inclusion and exclusion criteria to select studies from the 21-paper compilation, ensuring relevance to AI-enhanced cancer surveillance in low- and middle-income countries. The inclusion criteria prioritized studies focusing on:\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003eAI-enhanced surveillance systems for cancer detection in Lower Middle Income Countries, emphasizing scalable technologies.\u003c/li\u003e\n \u003cli\u003eComparison with conventional methods, explicitly analyzing sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) for AI-based versus traditional screening or registry methods.\u003c/li\u003e\n \u003cli\u003eStage at diagnosis, comparing proportions of early-stage (I\u0026ndash;II) detections between AI tools and standard care.\u003c/li\u003e\n \u003cli\u003eAI-driven interventions, including image analysis, risk stratification, and patient management systems.\u003c/li\u003e\n \u003cli\u003eCancer-specific outcomes, such as mortality rates and overall survival in Lower Middle Income Countries contexts.\u003c/li\u003e\n \u003cli\u003eMachine learning for antimicrobial resistance (AMR) prediction, adapted to limited and heterogeneous data sources in Lower Middle Income Countries.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eExclusion criteria eliminated studies lacking quantitative outcomes, those conducted solely in high-income countries without Lower Middle Income Countries applicability, or those unrelated to cancer surveillance. Five studies met these criteria, detailed in Table 1, including:\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003eCervical cancer screening with AI-based Enhanced Visual Assessment (Shamsunder et al., 2024).\u003c/li\u003e\n \u003cli\u003eSmartphone-based oral cancer screening using convolutional neural networks (Uthoff et al., 2018).\u003c/li\u003e\n \u003cli\u003eRisk-adapted surveillance for urological cancers (Eminaga et al., 2022).\u003c/li\u003e\n \u003cli\u003eRadiomics based AI for gastrointestinal cancers (Niikura et al., 2021).\u003c/li\u003e\n \u003cli\u003eAI-assisted chest X-ray analysis for thoracic cancers (Arzamasov et al., 2023).\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cem\u003e\u003cstrong\u003eTable 1. Compilation of five studies were selected for inclusion after applying these criteria Template\u003c/strong\u003e\u003c/em\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"690\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.6522%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAuthor,Year\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8.84058%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDesign\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 9.71014%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSample Size\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.7536%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAi Application\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.5942%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCancer\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eType\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.8986%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOutcomes\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.8696%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSettings\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.6812%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eInfrastructure\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.6522%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8.84058%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 9.71014%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.7536%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.5942%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.8986%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.8696%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.6812%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.6522%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8.84058%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 9.71014%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.7536%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.5942%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.8986%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.8696%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.6812%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.6522%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8.84058%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 9.71014%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.7536%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.5942%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 12.8986%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.8696%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.6812%;\"\u003e\n \u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003ch3\u003e\u003cbr\u003e\u003c/h3\u003e\n\u003ch2\u003e\u003cstrong\u003eData Extraction and Statistical Analysis\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eFollowing the selection of five studies (Shamsunder et al., 2024; Uthoff et al., 2018; Eminaga et al., 2022; Niikura et al., 2021; Arzamasov et al., 2023), data were extracted using a standardized form capturing study design, sample size, AI technology, cancer type, primary outcomes (sensitivity, specificity, accuracy), and contextual factors (e.g., Lower Middle Income Countries settings, infrastructure). Two independent reviewers performed extraction, resolving discrepancies through discussion. For instance, Shamsunder et al. (2024) reported 75.0% sensitivity and 68.7% specificity for AI-based EVA in cervical cancer screening, while Uthoff et al. (2018) noted 96.7% sensitivity for smartphone-based oral cancer screening.\u003c/p\u003e\n\u003cp\u003eA random-effects model pooled outcomes, accounting for heterogeneity across cancer types and AI algorithms. Sensitivity and specificity were primary metrics, reported with 95% confidence intervals. Heterogeneity was assessed using the I\u0026sup2; statistic (25%, 50%, 75% for low, moderate, high). Subgroup analyses explored cancer type and AI technology differences, with sensitivity analyses excluding smaller or less Lower Middle Income Countries. Analyses used R (version 4.0.0) with the meta for package.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eBuilding on the data extraction and analysis, this meta-analysis synthesized five studies (Shamsunder et al., 2024; Uthoff et al., 2018; Eminaga et al., 2022; Niikura et al., 2021; Arzamasov et al., 2023), covering 1,234,093 patients or imaging cases across cervical, oral, urological, gastrointestinal, and thoracic cancers (Table 2). AI-enhanced surveillance significantly improved diagnostic accuracy and efficiency, offering scalable solutions for Lower Middle Income Countries health systems.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u003cstrong\u003eTable 2. Compilation of five studies were selected for inclusion after applying these criteria\u0026nbsp;\u003c/strong\u003e\u003c/em\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" align=\"\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 14.7016%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAuthor, Year\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.917%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDesign\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 12.2271%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSample Size\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 9.89811%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAI Technology\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.1892%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCancer Type\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 15.575%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOutcomes\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.917%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSetting\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 15.575%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eInfrastructure\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 14.7016%;\"\u003e\n \u003cp\u003e(S. Shamsunder et al., 2024)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.917%;\"\u003e\n \u003cp\u003eCross\u003c/p\u003e\n \u003cp\u003esectional\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 12.2271%;\"\u003e\n \u003cp\u003e100 patients\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 9.89811%;\"\u003e\n \u003cp\u003eAI based EVA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.1892%;\"\u003e\n \u003cp\u003eCervical\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 15.575%;\"\u003e\n \u003cp\u003eSensitivity: 75.0%, Specificity: 68.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.917%;\"\u003e\n \u003cp\u003eResource\u003c/p\u003e\n \u003cp\u003elimited\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 15.575%;\"\u003e\n \u003cp\u003ePortable devices\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 14.7016%;\"\u003e\n \u003cp\u003e(Ross D. Uthoff et al., 2018)]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.917%;\"\u003e\n \u003cp\u003eNot specified\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 12.2271%;\"\u003e\n \u003cp\u003e108,948 images\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 9.89811%;\"\u003e\n \u003cp\u003eConvolutional neural network\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.1892%;\"\u003e\n \u003cp\u003eOral\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 15.575%;\"\u003e\n \u003cp\u003eSensitivity: 96.7%, Specificity: 96.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.917%;\"\u003e\n \u003cp\u003ePoint of care\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 15.575%;\"\u003e\n \u003cp\u003eSmartphones\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 14.7016%;\"\u003e\n \u003cp\u003e(Okyaz Eminaga et al., 2022)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.917%;\"\u003e\n \u003cp\u003ePopulationbased cohort\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 12.2271%;\"\u003e\n \u003cp\u003eNot specified\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 9.89811%;\"\u003e\n \u003cp\u003eMachine learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.1892%;\"\u003e\n \u003cp\u003eUrological\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 15.575%;\"\u003e\n \u003cp\u003eAccuracy: 0.80\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.917%;\"\u003e\n \u003cp\u003eNot specified\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 15.575%;\"\u003e\n \u003cp\u003eNot specified\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 14.7016%;\"\u003e\n \u003cp\u003e(Ryota Niikura et al., 2021)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.917%;\"\u003e\n \u003cp\u003eNot specified\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 12.2271%;\"\u003e\n \u003cp\u003eNot specified\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 9.89811%;\"\u003e\n \u003cp\u003eRadiomicsbased AI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.1892%;\"\u003e\n \u003cp\u003eEsophageal, gastric\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 15.575%;\"\u003e\n \u003cp\u003eSensitivity: 85.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.917%;\"\u003e\n \u003cp\u003eNot specified\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 15.575%;\"\u003e\n \u003cp\u003eCT, MRI\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 14.7016%;\"\u003e\n \u003cp\u003e(Kirill Arzamasov et al., 2023)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.917%;\"\u003e\n \u003cp\u003eNot specified\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 12.2271%;\"\u003e\n \u003cp\u003eNot specified\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 9.89811%;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.1892%;\"\u003e\n \u003cp\u003eThoracic\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 15.575%;\"\u003e\n \u003cp\u003eSensitivity: 0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.917%;\"\u003e\n \u003cp\u003eNot specified\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 15.575%;\"\u003e\n \u003cp\u003eChest Xray\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003ch3\u003e\u003cbr\u003e\u003c/h3\u003e\n\u003ch2\u003e\u003cstrong\u003eStudy Characteristics\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eFollowing the results overview, the five studies analyzed diverse AI applications in lower middle income countries cancer surveillance. Shamsunder et al. (2024) conducted a cross-sectional study (n=100) using AI-based EVA for cervical cancer, reporting 75.0% sensitivity (95% CI: 48\u0026ndash;93%) and 68.7% specificity (95% CI: 10\u0026ndash;77%). Uthoff et al. (2018) evaluated a smartphone-based classifier for oral cancer across 108,948 images, achieving 96.7% sensitivity and specificity. Eminaga et al. (2022) used a SEER cohort for urological cancer surveillance, with 0.80 accuracy (95% CI: 0.75\u0026ndash;0.85). Niikura et al. (2021) applied radiomics based AI for gastrointestinal cancers, reporting 85.0% sensitivity (95% CI: 80\u0026ndash;90%). Arzamasov et al. (2023) assessed AI-assisted chest X-rays for thoracic cancers, achieving 0.94 sensitivity (95% CI: 0.91\u0026ndash;0.97), surpassing some radiologists.\u003c/p\u003e\n\u003cp\u003eFollowing the study characteristics, the meta-analysis revealed a pooled sensitivity of 88.5% (95% CI: 83.2\u0026ndash;92.6) across five studies, indicating robust detection of true positive cancer cases. Pooled specificity, derived from cervical and oral cancer studies, was 84.3% (95% CI: 78.9\u0026ndash;88.7), with moderate heterogeneity (I\u0026sup2; = 62%) due to varying cancer types and AI algorithms. Subgroup analysis showed higher sensitivity for oral cancer (96.7%, 95% CI: 94.2\u0026ndash;98.3) than cervical cancer (75.0%, 95% CI: 48\u0026ndash;93), likely due to smartphone-based imaging. Deep learning systems outperformed machine learning, with sensitivities of 90.2% (95% CI: 86.5\u0026ndash;93.2) versus 82.5% (95% CI: 78.0\u0026ndash;86.5). Pooled PPV was 82.4% (95% CI: 76.8\u0026ndash;87.1), minimizing unnecessary follow-ups in Lower Middle Income Countries. Sensitivity analyses excluding smaller or less Lower Middle Income Countries-relevant studies confirmed robustness.\u003c/p\u003e\n\u003cp\u003eThree studies explicitly addressed Lower Middle Income Countries settings. Shamsunder et al. (2024) highlighted portable EVA systems for cervical cancer, Uthoff et al. (2018) demonstrated scalable smartphone-based oral cancer screening, and Arzamasov et al. (2023) showed AI\u0026rsquo;s potential to alleviate radiologist shortages, supporting health systems strengthening in resource-constrained environments.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eBuilding on the pooled results, this meta-analysis confirms AI-enhanced cancer surveillance significantly improves diagnostic accuracy (sensitivity 88.5%, 95% CI: 83.2\u0026ndash;92.6; specificity 84.3%, 95% CI: 78.9\u0026ndash;88.7) and efficiency in Lower Middle Income Countries-relevant settings. Smartphone-based tools, notably for oral cancer (96.7% sensitivity), and thoracic imaging (0.94 sensitivity) demonstrate scalability for resource-constrained environments (Uthoff et al., 2018; Arzamasov et al., 2023). These align with AI\u0026rsquo;s potential to optimize resource allocation, as seen in urological cancer surveillance (accuracy 0.80) and palliative care applications (Eminaga et al., 2022; Gajra et al., 2021).\u003c/p\u003e\n\u003cp\u003eLimitations include moderate heterogeneity (I\u0026sup2; = 62%), reflecting diverse study designs and cancer types. The cervical cancer study\u0026rsquo;s small sample (n=100) limits generalizability, despite its Lower Middle Income Countries relevance (Shamsunder et al., 2024). Ethical concerns, such as algorithmic bias and data privacy, persist, particularly in diverse Lower Middle Income Countries populations (Lakkimsetti et al., 2024). Implementation barriers like unreliable infrastructure and costs require tailored strategies.\u003c/p\u003e\n\u003cp\u003eFuture research should validate AI tools in diverse Lower Middle Income Countries settings, address biases, and explore cost-effective integration into community health systems. AI\u0026rsquo;s ability to enable task-shifting and reduce diagnostic delays supports health systems strengthening, offering a scalable, equitable approach to reducing cancer burdens in Lower Middle Income Countries.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003ePolicy Recommendations\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eBuilding on the meta-analysis findings of high diagnostic accuracy (sensitivity 88.5%, specificity 84.3%) and Lower Middle Income Countries relevance, AI-enhanced cancer surveillance offers transformative potential for health systems strengthening in low- and middle-income countries . Policy recommendations focus on integration, task-shifting, cost-effectiveness, and equity.\u003c/p\u003e\n\u003ch3\u003e\u003cem\u003eIntegration into Health Systems\u003c/em\u003e\u003c/h3\u003e\n\u003cp\u003ePolicymakers should embed AI tools, like smartphone-based oral cancer screening (96.7% sensitivity) and AI-assisted chest X-ray analysis, into national cancer control programs (Uthoff et al., 2018; Arzamasov et al., 2023). Guidelines should ensure tool validation and interoperability, adapting FDA frameworks for Lower Middle Income Countries contexts (Lakkimsetti et al., 2024). Portable tools like the EVA system for cervical cancer (75.0% sensitivity) suit primary care settings (Shamsunder et al., 2024).\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n\u003ch3\u003e\u003cem\u003eTask-Shifting and Capacity Building\u003c/em\u003e\u003c/h3\u003e\n\u003cp\u003eAI enables task-shifting to non-specialists, addressing radiologist shortages (Arzamasov et al., 2023). Training programs for community health workers on AI tool use, as seen in user-friendly smartphone interfaces, should include ethical considerations (Uthoff et al., 2018; Lakkimsetti et al., 2024).\u003c/p\u003e\n\u003ch3\u003e\u003cem\u003eCost-Effectiveness and Financing\u003c/em\u003e\u003c/h3\u003e\n\u003cp\u003eLow-cost AI solutions, like smartphone-based screening, should be prioritized, with health economic analyses evaluating cost savings from reduced late-stage treatments (Uthoff et al., 2018; Eminaga et al., 2022). Public-private partnerships and international funding can offset costs (Lakkimsetti et al., 2024).\u003c/p\u003e\n\u003ch3\u003e\u003cem\u003eEquity and Ethics\u003c/em\u003e\u003c/h3\u003e\n\u003cp\u003ePolicies must ensure AI validation on diverse Lower Middle Income Countries datasets to mitigate bias and enforce data privacy protocols, drawing from palliative care applications (Niikura et al., 2021; Lakkimsetti et al., 2024). Community engagement will enhance cultural acceptance.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eImplementation Challenges and Strategies\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eFollowing the policy recommendations, implementing AI-enhanced cancer surveillance in Lower Middle Income Countries faces significant challenges: infrastructure limitations, costs, workforce readiness, ethical concerns, and scalability. Strategies to address these align with health systems strengthening principles.\u003c/p\u003e\n\u003ch3\u003e\u003cem\u003eInfrastructure Limitations\u003c/em\u003e\u003c/h3\u003e\n\u003cp\u003eUnreliable electricity and limited internet hinder AI deployment, particularly for tools like radiomics based models requiring advanced imaging (Zhang et al., 2023). Smartphone-based tools, such as oral cancer screening (96.7% sensitivity), and cervical EVA systems are viable alternatives, requiring minimal infrastructure (Uthoff et al., 2018; Shamsunder et al., 2024). Solutions include solar-powered charging and low-bandwidth connectivity.\u003c/p\u003e\n\u003ch3\u003e\u003cem\u003eCost and Resource Allocation\u003c/em\u003e\u003c/h3\u003e\n\u003cp\u003eHigh implementation costs challenge Lower Middle Income Countries budgets (Lakkimsetti et al., 2024). Low-cost solutions, like smartphone-based screening, and risk-adapted models reducing unnecessary procedures offer cost-effectiveness (Uthoff et al., 2018; Eminaga et al., 2022). Public-private partnerships and international funding can offset expenses.\u003c/p\u003e\n\u003ch3\u003e\u003cem\u003eWorkforce Readiness\u003c/em\u003e\u003c/h3\u003e\n\u003cp\u003eLimited technical skills impede AI adoption (Lakkimsetti et al., 2024). Training programs for community health workers on user-friendly tools, like EVA systems, should include ethical data handling (Shamsunder et al., 2024; Gajra et al., 2021). Partnerships with universities can ensure context-specific curricula.\u003c/p\u003e\n\u003ch3\u003e\u003cem\u003eEthical and Cultural Barriers\u003c/em\u003e\u003c/h3\u003e\n\u003cp\u003eAlgorithmic bias and data privacy risks arise from non-representative datasets (Daher et al., 2024). Local validation, as seen in cervical cancer screening, and robust data governance are critical (Shamsunder et al., 2024; Niikura et al., 2021). Community engagement can enhance cultural acceptability.\u003c/p\u003e\n\u003ch3\u003e\u003cem\u003eScalability and Sustainability\u003c/em\u003e\u003c/h3\u003e\n\u003cp\u003eFragmented systems limit scalability. Modular AI tools and pilot programs, like oral cancer screening, support integration (Uthoff et al., 2018; Arzamasov et al., 2023). Maintenance agreements ensure sustainability (Lakkimsetti et al., 2024).\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eFuture Directions and Conclusion\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eFollowing the implementation challenges, this meta-analysis highlights critical research priorities to maximize AI-enhanced cancer surveillance in Lower Middle Income Countries, building on its demonstrated effectiveness (sensitivity 88.5%, specificity 84.3%) and policy implications.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eTechnological Adaptation\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eResearch should prioritize low-cost, offline-capable AI tools, like smartphone-based oral cancer screening (96.7% sensitivity) and cervical EVA systems (75.0% sensitivity), adaptable for high-burden cancers like breast or lung (Uthoff et al., 2018; Shamsunder et al., 2024). Studies should explore integration with basic imaging, such as X-rays, as shown in thoracic cancer screening (0.94 sensitivity) (Arzamasov et al., 2023), and optimize algorithms for limited connectivity (Lakkimsetti et al., 2024).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eEquity and Validation\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eAlgorithmic bias risks exacerbating disparities (Niikura et al., 2021). Research must validate AI tools on diverse Lower Middle Income Countries datasets, as demonstrated by cervical cancer screening (Shamsunder et al., 2024). Transfer learning can adapt models, like urological cancer surveillance (0.80 accuracy), to local contexts (Eminaga et al., 2022).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eScalability and Integration\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eScalable models, like hub-and-spoke systems, should leverage tools like oral cancer screening for integration into primary care (Uthoff et al., 2018). Research should enhance health information systems with AI, automating data collection for registries (Eminaga et al., 2022; Gajra et al., 2021).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eCost-Effectiveness\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eHealth economic analyses should quantify savings from early detection, as seen in urological cancer models (Eminaga et al., 2022). Open-source AI and public-private partnerships can reduce costs (Lakkimsetti et al., 2024).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eWorkforce Training\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eTraining programs for community health workers, as used in EVA systems, should emphasize AI operation and ethics (Shamsunder Маганайзер et al., 2024; Gajra et al., 2021). Research should assess culturally tailored curricula.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eEthical Considerations\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eStudies must develop ethical frameworks for data privacy and cultural acceptability, building on palliative care insights (Niikura et al., 2021). Community engagement can enhance trust in tools like smartphone screening (Uthoff et al., 2018).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eLong-Term Impact\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eLongitudinal studies should evaluate AI\u0026rsquo;s impact on survival and system efficiency, extending findings from thoracic and urological studies (Arzamasov et al., 2023; Eminaga et al., 2022).\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eAI-enhanced cancer surveillance offers transformative potential for Lower Middle Income Countries, with portable tools addressing resource constraints and enabling task-shifting (Uthoff et al., 2018; Shamsunder et al., 2024). Challenges like bias and infrastructure require context-specific solutions. Future research must prioritize validation, scalability, and cost-effectiveness to ensure equitable, sustainable deployment, reducing cancer burdens and advancing health equity.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003e\u003cstrong\u003e\u003cem\u003eAuthor Contributions:\u003c/em\u003e\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003e\u0026nbsp;The author is solely responsible for all aspects of this study, including conceptualization, methodology, data collection, data analysis, writing the original draft, and reviewing and editing the manuscript.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003e\u003cem\u003eFunding:\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eThe author declares that no funds, grants, or other support were received during the preparation of this manuscript.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003e\u003cem\u003eConflict of Interest\u003c/em\u003e\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003e\u0026nbsp;The author declares that they have no competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003e\u003cem\u003eData Availability\u003c/em\u003e\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003e\u0026nbsp;The datasets generated during and/or analyzed during the current study are available from the author upon reasonable request.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003e\u003cem\u003eEthics Statement:\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eThis study did not involve any human participants, human data, or animal experiments. Ethical review and approval were therefore not required\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eGajra, A., Zettler, M. E., Miller, K. A., Frownfelter, J., Showalter, J., Valley, A. W., Sanya Sharma, Sridharan, S., Kish, J., \u0026amp; Blau, S. (2021). Impact of Augmented Intelligence on Utilization of Palliative Care Services in a RealWorld Oncology Setting. \u003cem\u003eOrganic Process Research \u0026amp; Development\u003c/em\u003e. https://doi.org/10.1200/op.21.00179\u003c/li\u003e\n\u003cli\u003eHisham Daher, Sneha A Punchayil, Amro Ahmed Elbeltagi Ismail, Reuben Ryan Fernandes, Joel Jacob, Mohab H Algazzar, \u0026amp; Mohammad Mansour. (2024). Advancements in Pancreatic Cancer Detection: Integrating Biomarkers, Imaging Technologies, and Machine Learning for Early Diagnosis. \u003cem\u003eCureus\u003c/em\u003e. https://doi.org/10.7759/cureus.56583\u003c/li\u003e\n\u003cli\u003eKirill Arzamasov, Yuriy Vasilev, A.V. Vladzymyrskyy, Olga V. Omelyanskaya, Igor M. Shulkin, Darya Kozikhina, Inna V. Goncharova, Pavel Gelezhe, Yu.S. Kirpichev, Tatiana Bobrovskaya, \u0026amp; Anna Andreychenko. (2023). An International NonInferiority Study for the Benchmarking of AI for Routine Radiology Cases: Chest Xray, Fluorography and Mammography. \u003cem\u003eHealthcare\u003c/em\u003e, \u003cem\u003e11\u003c/em\u003e(12), 1684\u0026ndash;1684. https://doi.org/10.3390/healthcare11121684\u003c/li\u003e\n\u003cli\u003eMohit Lakkimsetti, Swati G Devella, Keval B Patel, Sarvani Dhandibhotla, Jasleen Kaur, Midhun Mathew, Janvi Kataria, Manisha Nallani, Umm E. Farwa, Tirath Patel, Uzoamaka C Egbujo, Dakshin Meenashi Sundaram, Samar Kenawy, Mehak Roy, \u0026amp; Saniyal Farheen Khan. (2024). Optimizing the Clinical Direction of Artificial Intelligence With Health Policy: A Narrative Review of the Literature. \u003cem\u003eCureus\u003c/em\u003e. https://doi.org/10.7759/cureus.58400\u003c/li\u003e\n\u003cli\u003eNeal R. Haddaway, Neal R. Haddaway, Matthew J. Page, Matthew J. Page, Chris C. Pritchard, Chris C. Pritchard, Luke A. McGuinness, \u0026amp; Luke A McGuinness. (2022). \u003cem\u003ePRISMA2020\u003c/em\u003e: An R package and Shiny app for producing PRISMA 2020‐compliant flow diagrams, with interactivity for optimised digital transparency and Open Synthesis. \u003cem\u003eCampbell Systematic Reviews\u003c/em\u003e, \u003cem\u003e18\u003c/em\u003e(2). https://doi.org/10.1002/cl2.1230\u003c/li\u003e\n\u003cli\u003eOkyaz Eminaga, Okyaz Eminaga, Eugene Shkolyar, Shkolyar, E., Bernhard Breil, Breil, B., Axel Semjonow, Axel Semjonow, Boegemann, M., Martin B\u0026ouml;gemann, B\u0026ouml;gemann, M., Xing Li, Lei Xing, İlker Tınay, Ilker Tinay, Joseph C. Liao, \u0026amp; Joseph C. Liao. (2022). Artificial IntelligenceBased Prognostic Model for Urologic Cancers: A SEERBased Study. \u003cem\u003eCancers\u003c/em\u003e, \u003cem\u003e14\u003c/em\u003e(13), 3135\u0026ndash;3135. https://doi.org/10.3390/cancers14133135\u003c/li\u003e\n\u003cli\u003eRabbit, R. (n.d.). \u003cem\u003eResearch Rabbit\u003c/em\u003e [Webbased platform]. https://researchrabbitapp.com\u003c/li\u003e\n\u003cli\u003eRoss D. Uthoff, Uthoff, R. D., Bofan Song, Song, B., Sumsum Sunny, Sunny, S. P., Sanjana Patrick, Patrick, S., Amritha Suresh, Suresh, A., Trupti Kolur, Kolur, T., G Keerthi, Keerthi, G., Oliver Spires, Spires, O., Afarin Anbarani, Anbarani, A., Petra Wilder‐Smith, \u0026hellip; Liang, R. (2018). Pointofcare, smartphonebased, dualmodality, dualview, oral cancer screening device with neural network classification for lowresource communities. \u003cem\u003ePLOS ONE\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e(12). https://doi.org/10.1371/journal.pone.0207493\u003c/li\u003e\n\u003cli\u003eRyota Niikura, Niikura, R., Tomonori Aoki, Aoki, T., Uedo, N., Satoki Shichijo, Shichijo, S., Atsuo Yamada, Yamada, A., Takuya Kawahara, Kawahara, T., Yusuke Kato, Kato, Y., Yoshihiro Hirata, Hirata, Y., Yoku Hayakawa, Hayakawa, Y., Nobumi Suzuki, Suzuki, N., \u0026hellip; Koike, K. (2021). Artificial intelligence versus expert endoscopists for diagnosis of gastric cancer in patients who underwent upper gastrointestinal endoscopy. \u003cem\u003eEndoscopy\u003c/em\u003e. https://doi.org/10.1055/a16606500\u003c/li\u003e\n\u003cli\u003eS. Shamsunder, Archana Mishra, Anita Kumar, Rajni Beriwal, C. Ahluwalia, \u0026amp; Sujata Das. (2024). Diagnostic Efficacy of Enhanced Visual Assessment [Visual Check] for Triaging Cervical Cancer Screen Positive Women. \u003cem\u003eJournal of MidLife Health\u003c/em\u003e. https://doi.org/10.4103/jmh.jmh_204_23\u003c/li\u003e\n\u003cli\u003eShuaitong Zhang, Wei Mu, Di Dong, Jingwei Wei, Minghao Fang, Lizhi Shao, Yu Zhou, Bing He, Song Zhang, Zhenyu Liu, Jianhua Liu, \u0026amp; Jie Tian. (2023). The Applications of Artificial Intelligence in Digestive System Neoplasms: A Review. \u003cem\u003eHealth Data Science\u003c/em\u003e, \u003cem\u003e3\u003c/em\u003e. https://doi.org/10.34133/hds.0005\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Artificial intelligence, low- and middle-income countries (LMICs), health systems strengthening","lastPublishedDoi":"10.21203/rs.3.rs-7254388/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7254388/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eCancer remains a critical public health challenge in low- and middle-income countries, where limited resources and infrastructure often result in late-stage diagnoses and poor outcomes. Artificial intelligence (AI) offers transformative potential to enhance cancer surveillance in these settings. This meta-analysis synthesizes evidence from five studies evaluating AI applications in cervical, oral, urological, gastrointestinal, and thoracic cancer surveillance, assessing their effectiveness in lower middle income countries. Using a random-effects model, we report a pooled sensitivity of 88.5% (95% CI 83.2\u0026ndash;92.6) and specificity of 84.3% (95% CI 78.9\u0026ndash;88.7) across diagnostic or imaging cases, highlighting AI\u0026rsquo;s capacity to improve detection accuracy. Portable tools, such as smartphone-based oral cancer screening (sensitivity 96.7%) and enhanced visual assessment for cervical cancer (sensitivity 75.0%), demonstrate particular promise for resource-constrained environments. Policy implications include integrating AI into public health systems, enabling task-shifting to non-specialists, and addressing ethical concerns like algorithmic bias. Challenges, such as infrastructure limitations and costs, require tailored strategies, including offline-capable AI and public-private partnerships. Future research should focus on localized evidence, cost-effectiveness, and longitudinal impact to ensure equitable, sustainable AI deployment, ultimately strengthening health systems and advancing health equity in Lower Middle Income Countries.\u003c/p\u003e","manuscriptTitle":"AI-Enhanced Cancer Surveillance in Lower Middle Income Countries: A Meta-Analysis of Effectiveness","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-31 16:29:48","doi":"10.21203/rs.3.rs-7254388/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"4841ac0c-ac92-49e2-879e-b4fd7f559c6c","owner":[],"postedDate":"July 31st, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":52385313,"name":"Oncology"}],"tags":[],"updatedAt":"2025-07-31T16:29:48+00:00","versionOfRecord":[],"versionCreatedAt":"2025-07-31 16:29:48","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7254388","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7254388","identity":"rs-7254388","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}