Library / learning-module
Understanding medical AI devices
Audience: patients-public, students-learners ยท Review state: published
When AI becomes a medical device
Software that diagnoses, screens, or guides treatment decisions is regulated as a medical device in many jurisdictions, and AI versions of that software now number in the hundreds of authorized products. Regulators maintain public lists of AI-enabled devices, most of them in radiology, cardiology and ophthalmology.
The device framing matters: it means an AI product must show evidence for its stated use, be manufactured under quality systems, and be monitored after release, just like a physical device.
What authorized AI actually does
The largest group reads medical images: flagging strokes on CT scans, measuring ejection fraction on echocardiograms, screening diabetic eye disease from retinal photos. Others triage worklists so urgent cases are read first, or monitor vital signs for early deterioration.
Nearly all authorized systems are locked models that support a clinician rather than autonomous decision-makers, and their labels state exactly what they may be used for.
How regulators evaluate them
Review asks the evidence questions: was performance demonstrated on data representative of the intended patients, does the model behave on data from hospitals it never saw, and how will the manufacturer detect performance drift after deployment. Lifecycle expectations now include monitoring plans, since a model can silently degrade as equipment, protocols and populations shift.
Generative and continuously learning systems are stretching these frameworks, and regulators are actively adapting rules for models that change over time.
A patient's practical view
If AI was involved in your scan or result, the final interpretation and responsibility still sit with the clinical team; authorized AI is an instrument they use, not a second opinion you must negotiate with.
Be more cautious with AI outside the clinic: consumer health chatbots and wellness apps mostly operate outside medical device regulation, and their answers carry no comparable evidence requirement.
Limitations
Educational overview of AI as medical devices, not an evaluation of any specific product and not medical advice.