AI is moving into clinical workflows fast, but if your software diagnoses, treats, or informs clinical decisions, it may be a medical device, and FDA regulates it. The hard part of AI in healthcare isn’t the model; it’s earning clinical and regulatory trust.
When AI is a medical device
Software that performs a medical purpose, on its own, is Software as a Medical Device (SaMD). An AI tool that flags findings on an image, triages patients, or recommends treatment is very likely SaMD. Its risk classification (and therefore its pathway) depends on the significance of the information it provides and the state of the patient’s condition.
The pathway
Like other devices, AI/ML SaMD typically clears via 510(k), De Novo, or PMA depending on risk and whether a predicate exists. The submission must show the model is clinically valid and reliable: a representative dataset, a sound validation methodology, defined performance metrics, and an honest account of limitations and failure modes.
Good Machine Learning Practice
FDA, with international partners, has published Good Machine Learning Practice (GMLP) principles, covering data quality and representativeness, avoiding bias, rigorous evaluation, transparency to users, and lifecycle monitoring. These are the expectations a credible AI submission is built around.
The PCCP: updating models without re-submitting
A unique challenge for AI is that models improve over time, but you can’t refile every time you retrain. FDA’s Predetermined Change Control Plan (PCCP) lets you specify, up front, the changes you intend to make to the model, how you’ll implement them, and how you’ll validate them, so you can update within an authorized envelope without a new submission. A well-written PCCP is one of the most powerful tools in regulated AI.
Build for trust from day one
The organizations that win in medical AI engineer clinical validity, explainability, security, and a regulatory path from the start, rather than retrofitting them after the model works. That combination, real ML plus a defensible regulatory strategy, is exactly where AI and medical devices meet.
Sequence Group and Protostar AI work at the intersection of applied AI and FDA-regulated products, from model validation strategy to SaMD submissions and PCCPs. Get in touch.