1,451 FDA-Authorized AI/ML Devices — What the Tracker Shows

As of the latest FDA AI/ML-Enabled Medical Device tracker update, 1,451 devices have received authorization — most through the 510(k) pathway, with a minority via De Novo or PMA.

The composition reveals where AI has reached clinical viability:

• Radiology / imaging — ~76%. Dominant category, driven by mature imaging datasets, clear ground-truth labels, and well-defined integration points (PACS, RIS).
• Cardiology — ~10%. Echocardiography, ECG analysis, coronary calcium scoring.
• Neurology — ~5%. Stroke triage, seizure detection, EEG analysis.
• Hematology, ophthalmology, pathology, others — remainder.

The concentration in imaging reflects data availability and evaluation tractability, not that AI is most useful there in absolute terms. Specialties with less-standardized data capture (e.g., primary-care clinical notes, complex surgical decision-making) have fewer tools because building and validating AI on those domains is harder — not because the opportunity is smaller.

For clinicians: the tracker entry for a specific product contains the indication for use, clinical-performance summary, and (for De Novo) the decision memo. Do not rely on vendor marketing; read the FDA entry.

For product teams: 510(k) via the AI/ML submission pathway remains the dominant route. Predicate selection, substantial-equivalence argumentation, and clinical-performance data requirements are the three gates. The “algorithm change protocol” introduced in recent FDA guidance matters for iterative model updates post-authorization.