Dana-Farber experts recommend actionable steps to enhance the rigor and transparency of FDA validation standards for radiology AI software.
Key Details
- 1The current FDA regulatory framework for SaMD validation is flexible but lacks explicit, consistent standards for radiology AI.
- 2Common radiology SaMD types include CADq, CADt, CADe, CADx, CADe/x, and CADa/o.
- 3Retrospective validation is typical for assistive radiology AI; prospective studies are standard for autonomous devices.
- 4Concerns exist over inadequate evaluation metrics for generative models and the use of synthetic data for some devices.
- 5Recommendations include removing nonclinical validation for certain devices, requiring more reader and prospective studies, adopting mandatory reporting checklists, and creating a public validation database.
Why It Matters
Clear and consistent FDA standards are crucial for ensuring the clinical reliability, performance, and reporting transparency of AI tools used in radiology, directly impacting adoption and trust among professionals and patients.
Source
AuntMinnie
Related News
•Radiology Business
Radiology Societies Urge Patient Safety in FDA Generative AI Regulation
RSNA and ACR call for patient safety and ongoing monitoring as the FDA considers regulations for generative AI in radiology.
•AuntMinnie
LLMs Rarely Recommend Unnecessary Imaging, Except for Thyroid Ultrasound
Large language models seldom over-recommend imaging, but thyroid ultrasound guidance shows substantial discrepancies.
•AuntMinnie
HPI Rebuts AI's Supposed Impact on Radiology Workforce Shortage
Harvey L. Neiman HPI challenges recent claims that AI has lessened the radiology workforce burden, citing evidence of rising workloads and imaging volumes.