AI/ML-Enabled Medical Device
Medical device that uses artificial intelligence or machine learning to perform its intended use.
Definition
An AI/ML-enabled device incorporates models that learn patterns from data to provide diagnostic, predictive, monitoring, or therapeutic functions. FDA maintains a public list of AI/ML-enabled devices that have been authorized. Most AI-enabled devices on FDA's list are in radiology, followed by cardiology and neurology, and most were cleared through 510(k). They are regulated as software functions, so they follow the same pathways, but FDA expects extra content on data management, model development, performance across subgroups and monitoring, as described in its draft guidance on AI-enabled device software functions issued in January 2025. In the EU, they must also meet the AI Act.What this means in practice
Most authorized AI devices today are 'locked' - the model weights do not change post-deployment. Adaptive systems require a Predetermined Change Control Plan (PCCP) to enable continued learning while maintaining oversight.Examples
- An AI/ML-enabled diagnostic device uses a locked algorithm trained on a large dataset of medical images to identify anomalies, requiring no post-market model updates.
- An adaptive AI/ML therapeutic device for personalized drug dosing continuously learns from a patient's physiological data to adjust treatment, operating under an FDA-approved Predetermined Change Control Plan (PCCP).
- A predictive AI/ML monitoring device analyzes continuous patient data streams to forecast potential adverse events, necessitating regular performance monitoring and validation to ensure accuracy.
- •Failing to establish a robust data management plan for training and validation data can lead to biased or ineffective AI/ML models.
- •Assuming that an AI/ML model’s performance in a controlled development environment will perfectly translate to real-world clinical settings is a common misconception.
- •Neglecting to implement a post-market surveillance strategy for AI/ML performance can result in undetected degradation or errors over time.
- •Underestimating the cybersecurity risks associated with AI/ML models, including data poisoning or adversarial attacks, can compromise device safety and effectiveness.
- •Not adequately documenting the AI/ML model’s design, development, and validation processes can lead to difficulties in demonstrating regulatory compliance.
Frequently asked questions
Cross-references
Related terms
Grouped by themeEditor's picks
· Hand-selected related conceptsFDA mechanism to pre-authorize specific modifications to AI/ML-enabled devices.
Guiding principles for the development of AI/ML-enabled medical devices.
Software intended for medical purposes that performs without being part of a hardware device.
FDA framework integrating premarket and post-market oversight across a device's life.
AI/ML Devices Deep Dive
· From this learning pathMethods (SHAP, saliency maps, prototype explanations) that help users understand why a model made a given prediction.
Large pretrained model adaptable to many downstream clinical tasks via fine-tuning or prompting.
Devices that use LLMs or diffusion models to generate text, images, or recommendations within a clinical workflow.
Structured documentation of a machine learning model's intended use, performance, and limitations.
Where this term appears across MedTech Terms.
- AI/ML Devices Deep DiveLesson 1 of 9
Latest in MedTech
Sources
3 sourcesEvery citation below opens the original document. Each is graded against our source-tier hierarchy so you can see what rests on binding law versus commentary.
- 1FDA AI/ML-Enabled Device ListTier 2 VerifiedFDAfda.gov
- 2MDCG Software GuidanceTier 2 VerifiedMDCGhealth.ec.europa.eu
- 3FDA - Software as a Medical Device (SaMD)Tier 2 VerifiedFDAfda.gov
Inline markers like [1] jump to the matching reference above.