FDA denies a petition to exempt certain radiology AI software from 510(k) review, stressing ongoing regulatory oversight.
Key Details
- 1FDA rejected Harrison.ai's petition seeking 510(k) exemption for four types of radiology AI (CADx, CADe, CADt, CADe/x) under specific classifications.
- 2At least 255 radiology AI devices received 510(k) clearance in 2025 alone, per FDA records.
- 3FDA clarified prior clearance does not guarantee manufacturer competence for future device development.
- 4Manufacturer-proposed self-evaluation and risk plans were dismissed due to insufficient FDA oversight mechanisms.
- 5Most public comments opposed the exemption, and FDA detailed significant differences between CADe, CADx, and CADe/x devices impacting regulatory needs.
- 6The agency emphasized vulnerabilities in relying on end-users (radiologists) to catch all performance issues.
Why It Matters
This decision underscores the FDA’s commitment to robust regulatory oversight for radiology AI, potentially affecting product development timelines and market strategies. The nuanced treatment of different device categories and the rejection of broad user detection arguments highlight stringent evidentiary requirements that developers must meet.
Source
AuntMinnie
Related News
•AuntMinnie
ONC Official at SIIM Calls for Smarter AI and Better Image Sharing
ONC's Dr. Jason Funderburk urges improvements in medical image sharing and AI readiness at SIIM 2024.
•AuntMinnie
SIIM Study: LLMs Excel in Numerical Radiology Tasks
At SIIM 2026, LLMs demonstrated high accuracy on radiology-based numerical tasks, particularly in extraction and judgment tests.
•AuntMinnie
AI Systems Predict Breast Cancer Risk Years Before Diagnosis
AI-enabled mammography systems can flag individuals at increased risk for breast cancer up to six years ahead of diagnosis, outperforming traditional risk measures.