Medical Devices; Radiology Devices; Classification of the Radiological Machine Learning-Based Quantitative Imaging Software With Predetermined Change Control Plan
The Food and Drug Administration (FDA) is classifying the radiological machine learning-based quantitative imaging software with predetermined change control plan into class II (special controls). The special controls that apply to the device type are identified in this order and will be part of the codified language for classification of the radiological machine learning-based quantitative imaging software with predetermined change control plan. We are taking this action because we have determined that classifying the device into class II will provide a reasonable assurance of safety and effectiveness of the device. We believe this action will also enhance patients' access to beneficial innovative devices, in part by reducing regulatory burdens.
What this rule actually says
The FDA just classified a specific type of AI software—machine learning models that analyze medical images (mainly X-rays, CT scans, MRIs) to measure or quantify things—as a Class II medical device. This means it's not a free-for-all, but it's not the most restrictive category either. The key twist: the FDA will accept these products if they come with a documented plan for how updates and changes will be controlled going forward.
Who it applies to you
- If you're building AI that analyzes radiological images (chest X-rays, lung CT scans, bone density scans, etc.) to produce measurements, volumes, densities, or other numerical outputs—this applies to you.
- If you're building AI that analyzes non-radiological medical images (pathology slides, ultrasounds, endoscopy footage) to quantify features—check with FDA first; this rule is narrowly written for radiology.
- If you're building a chatbot, hiring assistant, or general medical scribe that doesn't analyze medical images—this does not apply.
- Jurisdiction: United States only. If you're operating in the EU, Canada, or elsewhere, different rules apply (and often stricter ones).
- Data scope: Radiological images and the measurements your model outputs. You're not exempt from handling patient data safely, but that's separate from this classification.
What founders need to do
- Determine if your product is in scope (1 day). Ask: Does my AI actually analyze radiological images and output quantitative measurements? If no, stop. If yes, continue.
- Draft a predetermined change control plan (3-5 days). Document how you'll handle updates to your model, retraining decisions, and validation processes. This doesn't have to be elaborate—FDA is looking for evidence you've thought through what happens when things change.
- Prepare for FDA submission (2-4 weeks). You'll likely need a 510(k) submission (not full PMA). Gather performance data, clinical validation, and risk analysis. Consider hiring an FDA consultant (~$5k–$15k); worth it to avoid rejection.
- Plan for post-market surveillance (ongoing). Class II requires you to monitor real-world performance and report serious problems to FDA. Budget for data logging and incident tracking.
- Do not launch in the US without clearance. Selling an unclassified radiological AI tool is technically illegal, even if nobody's caught you yet.
Bottom line
If you're building radiological AI quantification software, act now—you need FDA clearance before launch, but the Class II pathway is faster and cheaper than older alternatives.