What is the most common FDA AI device category?

Radiology AI, including detection, quantification, triage, and diagnostic assistance across multiple imaging modalities.

How is imaging AI classified in the EU?

Under MDR Rule 11, typically Class IIa or IIb requiring Notified Body assessment, plus high-risk classification under the AI Act Annex III.

What validation is required for radiology AI?

Validation using representative datasets with diversity across demographics, equipment, protocols, and disease prevalence, with metrics including confidence intervals and subgroup analyses.

Can radiology AI work autonomously?

Most are designed to assist radiologists. Fully autonomous systems exist but face higher scrutiny and typically require stronger clinical evidence.

How should imaging AI handle equipment variability?

Validate across multiple manufacturers and protocols, and implement post-market monitoring to detect performance degradation outside validation parameters.

Quick answer

AI diagnostic imaging systems are regulated as medical devices, with radiology AI the largest FDA-authorized category and classified as high-risk under the EU AI Act.

Updated June 2026 · MmowW AI Compliance

AI Diagnostic Imaging Regulations: Radiology and Pathology Compliance (2026)

AI in Diagnostic Imaging

AI in diagnostic imaging represents the most mature medical AI segment, with hundreds of FDA-authorized products. Systems analyzing medical images face rigorous oversight because they directly influence clinical decisions. Radiology AI accounts for the majority of FDA-authorized AI/ML devices, ranging from triage tools to automated measurement and diagnostic aids.

FDA Regulation

Function	Example	Typical Class	Pathway
CADe (Detection)	Flagging lung nodules	Class II	510(k)
CADx (Diagnosis)	Characterizing lesions	Class II-III	510(k) or PMA
Triage	Prioritizing stroke cases	Class II	De Novo / 510(k)
Quantification	Cardiac measurements	Class II	510(k)
Autonomous diagnosis	Diabetic retinopathy	Class II-III	De Novo / PMA

Clinical Validation

Clinical validation requires demonstrating performance across diverse patient populations, imaging equipment, and clinical settings. Validation must address variability from different scanner manufacturers, acquisition protocols, and image quality. Performance metrics depend on intended use: sensitivity/specificity/AUC-ROC for detection, accuracy/precision for quantification.

EU Requirements

Under MDR Rule 11, most imaging AI falls into Class IIa or IIb requiring Notified Body assessment. The AI Act adds high-risk requirements. European regulations emphasize transparency so healthcare professionals understand system capabilities and limitations.

Digital Pathology

AI in digital pathology presents unique challenges. Whole slide imaging systems are themselves medical devices, and AI algorithms analyzing digital slides add another regulatory layer. Pathology AI must address variability in tissue preparation, staining, and scanning.

Post-Market Surveillance

Post-market monitoring should track performance metrics over time, watching for degradation from changes in equipment, protocols, or patient populations. Organizations should establish drift detection with predetermined investigation thresholds.

Compliance Checklist

Classify based on function (detection, diagnosis, triage, quantification, autonomous)
Identify appropriate premarket pathway
Design validation with representative datasets spanning equipment and demographics
Document metrics with confidence intervals and subgroup analyses
Address imaging variability in validation strategy
Implement post-market monitoring with drift detection
Ensure labeling communicates intended use, performance, and limitations

Check your AI compliance readiness — free.

Take the Readiness Check 3 minutes · 10 questions · no signup required

This article is for informational purposes only and does not constitute legal advice. Regulatory requirements change frequently — verify current rules with official sources. Built by Sawai Gyoseishoshi Office, Hiroshima, Japan.