Key points from article :
FDA proposes a framework to regulate medical devices that rely on AI algorithms.
Outlines two types of algorithms - locked and adaptive.
Locked algorithms provide the same result each time they’re fed the same input.
Adaptive algorithms will change its behaviour using a defined learning process.
Need balance between change that improve software over time and the safety of a patient.
Algorithms retrained with new training data would have to be resubmitted for FDA approval.
New approval would also be needed for system to expand beyond its original scope.
