# [Medical ML] EU Experience: CE-Marked Diagnostic AI EU Experience: CE-Marked Diagnostic AI Article #8 in Medical ML for Ukrainian Doctors Series EU CE-marked diagnostic AI experience and standards By Oleh Ivchenko | Researcher, ONPU | Stabilarity Hub | February 8, 2026 📋 Key Questions Addressed How does the European regulatory framework for medical AI differ from the US FDA approach? What is the impact of the EU AI Act on medical device software, and how does dual regulation work? What lessons can Ukrainian healthcare draw from the EU's evidence-focused approach? Context: Why This Matters for Ukrainian Healthcare Ukraine's regulatory trajectory aligns with the EU Medical Device Regulation (MDR) through ongoing European integration reforms. Understanding the European CE marking process—with its emphasis on clinical evidence and post-market surveillance—directly informs how Ukrainian hospitals should evaluate AI diagnostic tools. A Different Regulatory Philosophy: EU vs US Aspect 🇺🇸 US FDA 🇪🇺 EU MDR Primary pathway510(k) (substantial equivalence)Conformity assessment (risk-based) Clinical evidenceOften optional (510(k))Required for Classes IIa-III Third-party reviewLimitedNotified Body mandatory for IIa+ Post-market surveillanceMAUDE reportingActive PMS, PMCF studies AI-specific regulationFDA guidance (non-binding)EU AI Act (legally binding) CE-Marked AI Devices: The Numbers ~500-700 Total CE-marked AI devices ~53% Radiology focus ~60% Class IIa devices ~35% Class IIb/III (high-risk) The EU AI Act: A New Layer of Regulation The World's First Comprehensive AI Law The EU AI Act (Regulation EU 2024/1689), adopted March 2024, represents the world's first comprehensive legal framework for artificial intelligence. For medical devices, this creates a dual regulatory burden. ```mermaid graph TD A[Medical AI Device] --> B{MDR Classification?} B -->|Class I| C[Not High-Risk AI] B -->|Class IIa+| D[High-Risk AI] D --> E[MDR Compliance + AI Act Compliance] E --> F[Risk Management Art. 9] E --> G[Data Governance Art. 10] E --> H[Technical Documentation Art. 11] E --> I[Transparency Art. 13] E --> J[Human Oversight Art. 14] style D fill:#dc3545,color:#fff style E fill:#ffc107,color:#000 ``` AI Act Risk Classification Category Description Examples Prohibited AIBanned practicesSocial scoring, manipulative AI High-risk AIStrict compliance requiredMedical AI devices (via MDR Annex II) Limited-risk AITransparency obligationsChatbots, emotion recognition Minimal-risk AINo specific requirementsSpam filters 🔑 Automatic High-Risk: Any AI-enabled medical device software (MDSW) classified above Class I (MDR) automatically qualifies as high-risk AI under the AI Act. Timeline for AI Act Compliance ```mermaid timeline title EU AI Act Compliance Milestones August 2024 : AI Act enters into force May 2025 : General-purpose AI obligations August 2026 : Most high-risk AI obligations August 2027 : Medical device AI obligations (36-month transition) ``` The Dual Regulatory Framework: MDR + AI Act Integration, Not Duplication Article 11(2) of the AI Act allows manufacturers to maintain a single technical documentation file that combines MDR and AI Act requirements. Requirements for High-Risk Medical AI Requirement Area AI Act Additions (Beyond MDR) Risk ManagementAI-specific risks to fundamental rights Data GovernanceTraining data quality, bias assessment, demographic considerations Technical DocumentationDetailed AI architecture, training methodology, computational resources TransparencyDisclosure of accuracy levels, foreseeable limitations Human OversightDesign for human supervision capability Record-keepingAutomatic logging capabilities over system lifetime European Market Leaders and Notable Devices Top Companies in European Medical AI Company HQ Key Products Siemens Healthineers🇩🇪 GermanyAI-Rad Companion, ALFA Philips Healthcare🇳🇱 NetherlandsIntelliSpace, HealthSuite Imaging Qure.ai🇮🇳 IndiaqXR (TB/chest), qER Contextflow🇦🇹 AustriaAI radiology search ScreenPoint🇳🇱 NetherlandsTranspara (mammography) Notable CE-Marked AI Applications 🎀 Breast Cancer Transpara (ScreenPoint) Mammography triage, Class IIb 🫁 Chest X-ray qXR (Qure.ai) TB/pneumonia detection, Class IIa 🧠 Stroke Detection Viz LVO (Viz.ai) Large vessel occlusion alert Clinical Evidence: The European Approach MDR's Clinical Evidence Requirements Class Clinical Evidence Requirements Class ILiterature review may suffice Class IIaClinical evaluation, may require clinical investigation Class IIbClinical investigation usually required Class IIIMandatory clinical investigation for most Post-Market Clinical Follow-up (PMCF) A distinctive EU requirement is ongoing PMCF studies after market entry: PMCF Plan: Proactive collection of clinical data post-CE mark PMCF Studies: Prospective/retrospective studies in real-world use PMCF Report: Regular updates, part of technical documentation Periodic Safety Update Report: Mandatory for higher-risk devices European AI in Action: Case Studies Case Study 1: AI in European Breast Cancer Screening Country Program AI Role Results 🇸🇪 SwedenScreenTrustIndependent reader44% reduced workload 🇳🇱 NetherlandsDutch screening pilotsSecond readerImproved cancer detection 🇬🇧 UKNHS pilotsTriage/prioritizationOngoing evaluation Case Study 2: The Colonoscopy Deskilling Warning ⚠️ Critical Finding A pivotal 2025 Lancet study on AI-assisted colonoscopy revealed: Initial AI assistance: Improved adenoma detection After 6 months: Physician detection rates FELL when AI was withdrawn Implication: Over-reliance creates skill degradation "Researchers found that over six months of using AI, clinicians' ability to spot concerning features in colonoscopies declined when AI was not used during specific procedures." — Time, January 2025 Challenges Facing European Medical AI ```mermaid mindmap root((EU Medical AI Challenges)) Regulatory Complexity Dual MDR + AI Act Two sets of documentation Definition ambiguity Notified Body Bottleneck Only ~40 MDR bodies active AI Act accreditation emerging Limited dual-accredited Market Fragmentation National reimbursement Language requirements HTA variation SME Impact High documentation costs Disproportionate burden Innovation barrier ``` Practical Implications for Ukrainian Healthcare What the EU Experience Teaches EU Lesson Ukrainian Application Evidence requirements matterBuild clinical validation into AI procurement criteria Dual regulation is comingPrepare for MDR-aligned + AI-specific requirements Post-market surveillance worksImplement PMCF-style monitoring from deployment AI augments, doesn't replaceDesign workflows for human-AI collaboration Recommendations for Ukrainian Hospitals Prioritize CE-marked devices: EU certification provides stronger clinical evidence than 510(k)-only clearance Demand PMCF data: Ask vendors for post-market clinical follow-up results Plan for dual compliance: Expect both MDR-style and AI Act-style requirements Build monitoring infrastructure: Implement local performance tracking Train for collaboration: Prevent deskilling through proper training Conclusions: Original Insights 🌍 Global Future The EU's dual regulatory framework represents the global future—other jurisdictions will likely adopt similar models ✅ Quality Filter Stricter clinical evidence requirements may result in fewer but better-validated tools ⚠️ Deskilling Risk European studies show AI assistance can degrade physician skills when over-relied upon 📊 PMCF Advantage Devices with robust post-market data will increasingly differentiate themselves Questions Answered ✅ How does the EU regulatory framework differ from the US FDA approach? The EU requires stronger clinical evidence through MDR conformity assessment, mandatory Notified Body review, and ongoing Post-Market Clinical Follow-up. The AI Act adds a second regulatory layer. ✅ What is the impact of the EU AI Act on medical device software? Medical AI devices (Class IIa and above) are automatically classified as "high-risk AI," requiring dual compliance by August 2027. ✅ What lessons can Ukraine draw from the EU approach? The EU model emphasizes evidence-based approval, ongoing monitoring, and human-AI collaboration—principles Ukraine should adopt as it aligns with EU standards. Next in Series: Article #9 - UK NHS AI Lab: Lessons Learned Series: Medical ML for Ukrainian Doctors | Stabilarity Hub Research Initiative Author: Oleh Ivchenko | ONPU Researcher | Stabilarity Hub