AI-Powered Diagnostics: Where the EU AI Act Meets Medical Device Regulation
AI diagnostic tools must satisfy both the EU AI Act and the Medical Device Regulation. Understanding the crossover is critical for medtech companies and hospitals.
Dual Regulation: The Diagnostic AI Challenge
AI-powered diagnostic tools — radiology AI, sepsis prediction algorithms, dermatology AI — must simultaneously satisfy the EU Medical Device Regulation (MDR) or In Vitro Diagnostic Regulation (IVDR) and the EU AI Act's requirements for high-risk AI systems. Both are demanding. Together, they create the most complex compliance environment in any sector.
MDR vs. AI Act: Where Requirements Align and Conflict
Where they align: MDR's technical documentation requirements (Annex II) significantly overlap with AI Act Annex IV — both require development process documentation, software lifecycle records, clinical/validation evidence, risk management records, and post-market surveillance data. Organizations building MDR documentation can structure it to simultaneously satisfy AI Act requirements. Where they diverge: MDR requires mandatory Notified Body involvement for Class IIa+ devices. The AI Act allows self-assessment conformity for most high-risk AI (Notified Body mandatory only for biometric/critical infrastructure AI). Post-market surveillance cadence also differs between MDR's PSURs and AI Act's post-market monitoring requirements.
The Critical Question: Medical Device vs. AI System Classification
Not every clinical AI tool is a medical device. MDCG guidance distinguishes between tools that "drive clinical decision-making" (SaMD) and general-purpose tools that process clinical data without clinical claims. This distinction matters enormously — SaMD faces MDR's demanding clinical evidence requirements. Companies should get early clarity from a Notified Body before committing to a compliance pathway.
Hospital Deployer Obligations
Hospitals must: conduct Fundamental Rights Impact Assessments before deploying diagnostic AI, ensure clinician oversight (every AI-influenced diagnosis requires qualified clinician review), monitor AI performance in their local patient population (demographic differences from training populations significantly affect diagnostic accuracy), and report serious incidents to providers within 15 days.
Need medtech AI compliance expertise? Browse our directory of compliance consultants specializing in healthcare AI and MDR/AI Act intersection.