The FDA’s database of AI/ML-enabled medical devices crossed 521 authorized products in 2023, a figure that represents a 521-fold increase since 2008, when the first AI-based medical device — a computer-aided detection system for mammography — was cleared through the 510(k) pathway. The growth curve tells a story not just about the maturation of medical AI technology, but about a regulatory framework that has been learning to accommodate a fundamentally different class of software in real time.
The Specialty Breakdown: Radiology Dominates
Of the 521 AI-enabled devices authorized through 2023, radiology and imaging accounted for approximately 75% — roughly 391 devices. This concentration reflects both the maturity of computer vision techniques on medical imaging and the relatively well-defined performance benchmarks available for image analysis tasks. Within radiology, the sub-specialty breakdown favors:
- Cardiovascular imaging (CT and echo): approximately 91 devices
- Neurology/neuroradiology (brain imaging, stroke detection): approximately 67 devices
- Chest imaging (pulmonary nodule detection, pneumonia screening): approximately 59 devices
- Mammography and breast imaging: approximately 44 devices
- Musculoskeletal imaging: approximately 35 devices
Outside radiology, pathology, ophthalmology (particularly diabetic retinopathy screening), and cardiology (arrhythmia detection from ECG and wearable data) have seen growing numbers of clearances. Primary care and behavioral health AI tools remain a small fraction of the total.
510(k), PMA, and De Novo: Understanding the Pathways
The FDA’s regulatory pathways for medical devices are not interchangeable. The pathway chosen significantly affects the rigor of the pre-market review and the post-market obligations of the manufacturer.
The 510(k) pathway — which accounts for the overwhelming majority of AI medical device clearances — does not require clinical trials demonstrating safety and efficacy. It requires substantial equivalence to a legally marketed predicate device. For the first AI imaging tools, this was largely straightforward; as AI devices have become more sophisticated, the identification of appropriate predicates has become increasingly strained.
The De Novo pathway is used for novel, low-to-moderate risk devices with no predicate. It results in a risk classification and special controls that can then serve as a predicate for future 510(k) applications. Several landmark AI device authorizations — including DermaSensor’s EDS in 2024 and several stroke detection tools — came through De Novo, establishing new regulatory pathways for device categories that did not previously exist.
The PMA (Premarket Approval) pathway is reserved for Class III (high-risk) devices and requires clinical evidence of safety and effectiveness from controlled studies. As of 2025, very few AI medical devices had been approved through PMA, reflecting both the regulatory complexity and the tendency to design AI tools for lower-risk intended uses that avoid Class III classification.
The Adaptive AI Problem
The most unresolved regulatory challenge for clinical AI is the adaptive algorithm problem: AI/ML-based Software as Medical Device (SaMD) may be designed to continuously learn from new data after deployment, potentially changing its performance characteristics without new pre-market review. The FDA’s 2021 Action Plan for AI/ML-Based SaMD described a proposed framework for predetermined change control plans — a mechanism by which manufacturers would pre-specify the types of modifications the algorithm might undergo and the associated validation requirements, avoiding the need for a new 510(k) for every update. As of 2025, this framework was still being developed through public comment processes and pilot programs.
What the 2026 Pipeline Signals
The composition of the AI device pipeline as it entered 2026 shows several emerging trends. First, multimodal systems — AI that integrates imaging with structured EHR data, genomics, or wearable signals — are moving through development at a faster rate than in prior years. Second, ambient clinical intelligence tools (AI that listens to physician-patient encounters and generates structured clinical notes) are entering regulatory review, with a different risk profile than imaging AI. Third, international regulatory harmonization efforts, including cooperation with the European Medicines Agency on the AI Act’s medical device provisions, are creating pressure toward more rigorous pre-market clinical validation.
Key Takeaway
521 FDA-authorized AI medical devices represents significant technological maturation, but the regulatory framework has not fully caught up with the capabilities of modern adaptive AI systems. The 510(k) pathway’s reliance on predicate equivalence is increasingly strained for novel AI architectures. The adaptive algorithm problem — how to regulate continuously learning AI after deployment — remains the central unresolved challenge for the field.
Sources
1. FDA. Artificial Intelligence and Machine Learning (AI/ML)-Enabled Medical Devices. fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-and-machine-learning-aiml-enabled-medical-devices. Updated 2023.
2. FDA. Artificial Intelligence and Machine Learning (AI/ML)-Based Software as a Medical Device (SaMD) Action Plan. January 2021.
3. Hwang TJ, Kesselheim AS, Vokinger KN. Lifecycle Regulation of Artificial Intelligence- and Machine Learning-Based Software Devices in Medicine. JAMA. 2019;322(23):2285–2286.
Medical Disclaimer: This article is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional for medical decisions.