FDA Regulation of AI Medical Devices Quiz

Answers and explanations appear as you take the quiz.

1. What must a manufacturer show to obtain 510(k) clearance?
   • That the device is substantially equivalent to a legally marketed predicate device
   • That the device has completed three phases of randomized clinical trials
   • That the device has no predicate and is low to moderate risk
   • That the device learns continuously from patient data
2. The FDA authorized IDx-DR, an autonomous AI for diabetic retinopathy, in 2018. Which pathway did it use, and why?
   • PMA, because every diagnostic AI is Class III
   • 510(k), because many retinal cameras already existed
   • De Novo, because it was novel and had no suitable predicate
   • It needed no authorization under the Cures Act exclusion
3. What does a predetermined change control plan (PCCP) allow?
   • Selling a device before the FDA has reviewed it
   • Making specified, pre-validated modifications to an AI device without a new submission for each one
   • Letting a model rewrite its own intended use in the field
   • Skipping postmarket monitoring
4. Which three parts make up a PCCP?
   • Predicate list, clinical trial, and labeling
   • Risk class, user fee, and establishment registration
   • Algorithm source code, training data, and patent filings
   • Description of modifications, modification protocol, and impact assessment
5. Which statement about most FDA-authorized AI devices is accurate?
   • They are generally locked models that change only through controlled updates
   • They retrain themselves automatically on every patient they see
   • They are all Class III devices approved through PMA
   • They are mostly used in dermatology
6. Under the 21st Century Cures Act, which kind of software can fall outside the definition of a device?
   • Software that analyzes medical images to detect tumors
   • Any software sold by a hospital
   • Clinical decision support that shows clinicians the basis for its recommendations so they can review them independently
   • Any software that uses machine learning
7. Under the EU Medical Device Regulation, which rule classifies software, and where does most diagnostic software end up?
   • Rule 1, which puts it in Class I
   • Rule 11, which puts most of it in Class IIa or higher
   • Annex XVI, which exempts it
   • Rule 22, which places all software in Class III
8. Why do reviewers ask for AI device performance broken down by subgroup and acquisition equipment?
   • To calculate user fees
   • To choose a predicate device
   • To decide whether the device is Class I
   • Because models can fail when patient populations or equipment such as scanner vendors differ from the training data