Only 0.3% of FDA-cleared radiology AI devices have registered trials on patient outcomes, raising concerns about evidence standards.
Key Details
- 11,059 FDA-cleared AI devices are used in radiology; only three had registered trials on patient outcomes.
- 2Out of 1,357 total FDA-cleared AI medical devices, just three evaluated mortality, morbidity, or readmissions.
- 3Only 2.5% were linked to prospective trials, 0.9% posted results or had peer-reviewed publications.
- 462% of studies used small, homogeneous cohorts; vulnerable populations often excluded (e.g., pregnant, pediatric, non-English speakers).
- 5Authors recommend a three-phase evidence ladder with mandatory demographic reporting and larger, prospective trials.
- 6FDA's 510(k) pathway only requires substantial equivalency, not proof of clinical benefit to patients.
Why It Matters
This report highlights a significant gap between regulatory clearance and actual clinical outcome testing for radiology AI, calling attention to the need for stronger pre- and post-market evidence requirements and greater transparency. Radiology professionals and purchasers should scrutinize the clinical validation of AI tools before adoption.
Source
AuntMinnie
Related News
•Radiology Business
Radiology Partners Urges FDA for Clarity on Imaging AI Regulations
Radiology Partners and Mosaic Clinical Technologies petition the FDA for clearer regulatory guidelines on imaging AI vision-language models.
•AuntMinnie
AI Framework Boosts MRI Quality Control for Prostate Cancer
A deep learning AI framework improved radiologist and model accuracy for prostate cancer detection by assessing MRI image quality.
•Cardiovascular Business
AI Finds Hidden Aortic Stenosis in Routine Chest CTs
AI algorithms can identify undiagnosed aortic stenosis on routine chest CT scans, potentially improving early detection and patient care.