AI medical device safety is no longer just about model accuracy — this evidence map shows why real risk emerges across the full lifecycle: retraining, workflow integration, postmarket surveillance, cybersecurity, and human oversight. Built from 146 references and 151 original studies, the full map gives a practical view of where AI medical devices are already improving safety, where reporting and regulation still fail, and which safeguards matter before deployment.
Human-verified editorial review
Verified by World ID proof-of-human. This editorial layer was submitted from a SAIMSARA account verified as a unique human.
Abstract: To synthesize the available structured evidence on AI medical device safety, with emphasis on lifecycle risk management, postmarket surveillance, clinical deployment, cybersecurity, usability, workflow integration, and regulatory oversight. The review uses 146 references and builds its evidence map from 151 original studies with 1823692 total participants/sample observations (topic-deduplicated ΣN). Across the mapped evidence, AI medical device safety emerges predominantly as a lifecycle governance challenge rather than a one-time performance question, with risks shifting after deployment through model updates, workflow integration, and connected infrastructure. Recurrent signals indicate that postmarket reporting is often insufficient to attribute harms to AI components, with 34.5% of FDA medical device reports lacking adequate AI/ML information and less than 2% of approved AI devices updated through retraining despite documented site-specific performance shifts. Promising mitigations span deployment safety cases, usability engineering, threshold customization, and IoMT cybersecurity monitoring, but their clinical impact remains supported mainly by experimental and early-phase evidence. This suggests a practical role for adaptive, auditable oversight frameworks that link premarket clearance to continuous real-world monitoring. Future research should prioritize prospective deployment studies with standardized AI-specific safety reporting to close the gap between technical promise and demonstrated patient benefit.
Keywords: AI medical devices; Medical device safety; Post-market surveillance; Adverse event reporting; Regulatory compliance; Software as medical device; Model retraining; Distribution shift; Predictive maintenance; Clinical workflow safety
Review Stats
Final search date and database lock: 2026-05-14 22:21:05 CEST
Plan: Pro (expanded craft tokens; source: Semantic Scholar)
Source: Semantic Scholar
Total Abstracts/Papers: 707
Downloaded Abstracts/Papers: 707
Included original and non-original Abstracts/Papers (all): 295
Included original Abstracts/Papers (Vote counting by direction of effect): 151
Reference Index (links used in paper): 146
Total participants/sample observations (topic deduplicated ΣN): 1823692
Get access to the full paper
Unlock the full evidence map
The full evidence review, including the Introduction, Methods, Results, Discussion, Conclusion, figures, and complete reference index, opens after purchase or sign-in.
The Evidence Object JSON is a separate machine-readable evidence product: a concentrated synthesis of results, topic-level evidence, and discussion across original and non-original studies. It can be directly input into your LLM, agent, or RAG workflow.
[7] AI as a Medical Device Adverse Event Reporting in Regulatory Databases: Protocol for a Systematic Review — https://doi.org/10.2196/48156
[9] Real-Time Edge AI with Explainable Deep Learning for Predictive Maintenance of Medical Devices: *Instantaneous Detection of Medical Device Failures using Edge AI — https://doi.org/10.1109/isiber68248.2026.11470555
[12] Global Harmonization of Artificial Intelligence-Enabled Software as a Medical Device Regulation: Addressing Challenges and Unifying Standards — https://doi.org/10.1016/j.mcpdig.2024.100191
[15] Post-Marketing Surveillance and Regulation of the Pharmaceutical and Medical Device Market: Exploring the Possibilities of Healthcare Artificial Intelligence Applications — https://doi.org/10.61093/pgrl.1(1).51-64.2025
[28] Methodology for Conducting Post-Marketing Surveillance of Software as a Medical Device Based on Artificial Intelligence Technologies — https://doi.org/10.17691/stm2022.14.5.02
[31] Risk Assessment and Classification of Medical Device Software for the Internet of Medical Things: Challenges arising from connected, intelligent medical devices — https://doi.org/10.1145/3567445.3571104
[33] AI-Powered QA in Healthcare Software: Leveraging Predictive Analytics and Digital Twins for Safe, Cost-Effective, and Agile Medical Systems — https://doi.org/10.32996/jcsts.2025.4.1.70
[43] AI-based assessment of Clinical Activity Score and detection of active thyroid eye disease using facial images: validation of Glandy CAS — https://doi.org/10.1136/bmjophth-2025-002264
[74] Stakeholder Perspectives on Early Feasibility Studies for Digital Health Technologies in the European Union: Qualitative Interview Study — https://doi.org/10.2196/77982
[80] Edge-Based Computation of Super-Resolution Superlet Spectrograms for Real-Time Estimation of Heart Rate Using an IoMT-Based Reference-Signal-Less PPG Sensor — https://doi.org/10.1109/jiot.2023.3322947
[92] OSAIRIS: Lessons Learned From the Hospital-Based Implementation and Evaluation of an Open-Source Deep-Learning Model for Radiotherapy Image Segmentation. — https://doi.org/10.1016/j.clon.2024.10.032
[97] Evaluating the Safety and Usability of an Over-the-Counter Medical Device for Adults With Mild to Moderate Hearing Loss: Formative and Summative Usability Testing — https://doi.org/10.2196/65142
[104] User Interface Improvement for Usability Test of Artificial Intelligence-based Software as a Medical Device for Diagnosis of Neovascular Age-related Macular Degeneration — https://doi.org/10.17480/psk.2024.68.2.121
[108] Clinical Trial Design and Regulatory Requirements for Artificial Intelligence as a Medical Device: A PRISMA-ScR-Guided Scoping Review of Global Guidance and Evidence (2017-2025). — https://doi.org/10.3390/jcm15051937
[116] The AI cycle of health inequity and digital ageism: mitigating biases through the EU regulatory framework on medical devices — https://doi.org/10.1093/jlb/lsad031
[120] Health Disparities and Reporting Gaps in Artificial Intelligence (AI) Enabled Medical Devices: A Scoping Review of 692 U.S. Food and Drug Administration (FDA) 510k Approvals — https://doi.org/10.1101/2024.05.20.24307582
[125] Assuring the safety of AI-based clinical decision support systems: a case study of the AI Clinician for sepsis treatment — https://doi.org/10.1136/bmjhci-2022-100549
[130] Safety and User Experience of a Generative Artificial Intelligence Digital Mental Health Intervention: Exploratory Randomized Controlled Trial — https://doi.org/10.2196/67365
[134] Generative AI’s healthcare professional role creep: a cross-sectional evaluation of publicly accessible, customised health-related GPTs — https://doi.org/10.3389/fpubh.2025.1584348
[136] Key Information Influencing Patient Decision-Making About AI in Health Care: Survey Experiment Study — https://doi.org/10.2196/75615
[138] Real-Time Monitoring of Personal Protective Equipment Adherence Using On-Device Artificial Intelligence Models — https://doi.org/10.3390/s25072003
[149] Tiny Deep Ensemble: Uncertainty Estimation in Edge AI Accelerators via Ensembling Normalization Layers with Shared Weights — https://doi.org/10.1145/3676536.3676804
[151] Artificial Intelligence–Enabled Intensive Care Unit (AI‑ICU): Integrated Framework for Real‑Time Decision Support, Automated Clinical Handoffs, and Intelligent Wound Monitoring — https://doi.org/10.36948/ijfmr.2025.v07i06.63700
[164] Knowledge-based in silico models and dataset for the comparative evaluation of mammography AI for a range of breast characteristics, lesion conspicuities and doses — https://doi.org/10.48550/arxiv.2310.18494
[167] How threshold customisation affects the performance of a multiclass X-ray AI model for primary care triage: a retrospective study — https://doi.org/10.1136/bmjopen-2025-111127
[175] Safeguarding Connected Health: Leveraging Trustworthy AI Techniques to Harden Intrusion Detection Systems Against Data Poisoning Threats in IoMT Environments — https://doi.org/10.58496/bjiot/2023/005
[177] Clinical Effectiveness of an Artificial Intelligence-Based Prediction Model for Cardiac Arrest in General Ward-Admitted Patients: A Non-Randomized Controlled Trial — https://doi.org/10.3390/diagnostics16020335
[198] Leveraging the smarts in your phone: An artificial intelligence-driven iOS application for neurosurgical navigation of external ventricular drains — https://doi.org/10.36922/aih.8195
[223] CBRNEmedicine Project: CBRNE Point of Care Testing (CBRNE POCT)–An Innovative Solution for Rapid Detection and Identification of Hazardous Materials for Emergency Medical System, Medical Hubs, and Hospitals — https://doi.org/10.1017/s1049023x26104798
[225] Design and performance evaluation of a green LED OFDM LiFi system for an electromagnetic interference sensitive hospital network — https://doi.org/10.1007/s10791-026-09906-0
[226] Wearable Biosensors and AI Analytics for Continuous Health Monitoring and Early Disease Detection — https://doi.org/10.59675/e225
[248] Can Artificial Intelligence Aid Diagnosis by Teleguided Point-of-Care Ultrasound? A Pilot Study for Evaluating a Novel Computer Algorithm for COVID-19 Diagnosis Using Lung Ultrasound — https://doi.org/10.3390/ai4040044
[249] Improving internet of health things security through anomaly detection framework using artificial intelligence driven ensemble approaches — https://doi.org/10.1038/s41598-025-10016-y
[252] MedGlasses: A Wearable Smart-Glasses-Based Drug Pill Recognition System Using Deep Learning for Visually Impaired Chronic Patients — https://doi.org/10.1109/access.2020.2967400
[254] The feasibility of double automated reading of chest radiographic screening results (based on the Moscow experiment on computer vision in radiology) — https://doi.org/10.22328/2079-5343-2026-17-1-77-87
[267] Unveiling Healthcare Professionals’ Perspectives through a Knowledge, Attitude, and Practice Study on Artificial Intelligence in Materiovigilance – An Interventional Study — https://doi.org/10.4103/ajprhc.ajprhc_130_25
[268] 112-LB: Proof-of-Concept Testing of an Artificial Intelligence–Based Fully Closed-Loop System in Hospitalized Patients with Diabetes — https://doi.org/10.2337/db23-112-lb
[272] Single-Use Autoinjector Functionality And Reliability For At-Home Administration Of Benralizumab For Patients With Severe Asthma: GRECO Trial Results — https://doi.org/10.2147/JAA.S224266
[273] Evaluation of Four Artificial Intelligence–Assisted Self-Diagnosis Apps on Three Diagnoses: Two-Year Follow-Up Study — https://doi.org/10.2196/18097
[283] Multi-Layered Unsupervised Learning Driven by Signal-to-Noise Ratio-Based Relaying for Vehicular Ad Hoc Network-Supported Intelligent Transport System in eHealth Monitoring — https://doi.org/10.3390/s24206548
[285] Using artificial intelligence for personal protective equipment guidance for healthcare workers in the COVID-19 pandemic and beyond. — https://doi.org/10.33321/cdi.2022.46.51