FDA medical device regulation
FDA puts radiology AI software into class II
The new order covers software that uses machine learning on radiological images and includes a predetermined change control plan. FDA says the category can keep evolving under special controls, with 510(k) review still required.
FDA has classified radiological machine-learning-based quantitative imaging software with a predetermined change control plan as a class II device. The order took effect June 17, 2026, and FDA says the framework should preserve safety and effectiveness while giving developers a clearer path for updates.
- FDA put a radiology AI software category into class II.
- Special controls will apply under 21 CFR Part 892.
- The order took effect June 17, 2026.
- FDA says the change should reduce regulatory burden and improve access.
- Radiology software that uses machine learning to turn scans into quantitative outputs now has a clearer federal lane
Radiology software that uses machine learning to turn scans into quantitative outputs now has a clearer federal lane. The Food and Drug Administration, part of the federal Department of Health and Human Services, said June 17 that it is classifying radiological machine-learning-based quantitative imaging software with a predetermined change control plan, or PCCP, into class II.
That matters because the order took effect June 17, 2026, and FDA said the classification was applicable as of Feb. 24, 2023. In plain terms, the agency is saying this category belongs under special controls, not the stricter automatic class III track.
What the class II label changes
Class II devices still have to clear federal safety and effectiveness requirements, but they are governed by special controls tailored to the product type. Here, FDA says those controls are identified in the order itself and folded into 21 CFR Part 892, the part of the regulations covering radiology devices.
The software covered by the classification is not a single product. It is software that uses machine learning on radiological images to produce quantitative outputs such as view selection, segmentation and landmarking, while also laying out in advance how certain changes will be handled through a predetermined change control plan. That is the mechanism that lets a developer build some future updates into the device’s design from the start.
Why the label matters to patients and developers
For developers, the point is a clearer route to market and less regulatory friction as the software changes over time. For hospitals, imaging centers and radiologists, the practical value is that tools in this category can keep evolving without forcing the whole product back through the most burdensome path every time a planned update arrives.
FDA said the framework should enhance patients’ access to beneficial innovative devices while preserving a reasonable assurance of safety and effectiveness. The agency is not approving a single brand-name product here. It is drawing the line for a whole class of radiology software that can change in controlled ways over time.