In a March 5 letter, the American College of Radiology(ACR), the Radiological Society of North America (RSNA) and the Society for Imaging Informatics in Medicine (SIIM) urged U.S. Department of Health and Human Services (HHS) officials to reject a “midnight” proposal by the immediate-past HHS Secretary to permanently exempt certain medical devices from the FDA’s 510(k) premarket notification requirements.

The proposal would include several types of devices used by radiologists in patient care, including artificial intelligence/machine learning (AI/ML)-enabled software for computer-assisted/aided triage, detection or diagnosis. It was noteworthy for the lack of FDA involvement and for contradicting the agency’s stated plans for enhancing oversight of AI/ML-enabled software.

“The proposal is extraordinarily concerning from a patient safety perspective. Although we do not anticipate implementation by the current administration, informatics experts must inform regulators of the potentially harmful impacts resulting from this idea in case the proposal resurfaces,” says Howard B. Fleishon, MD, MMM, FACR, chair of the ACR Board of Chancellors.

While the FDA has cleared or authorized many AI/ML-enabled devices for the radiology market, the agency continues to face oversight challenges due to the impact on algorithm performance of changing practice environments, diverse patient populations and input devices. The ACR will continue to work with the FDA on these issues, organization officials say.

“We encourage the FDA to bolster their capabilities to ensure algorithms perform as intended in the real world and over time,” says Christoph Wald, MD, PhD, MBA, FACR, chair of the ACR Commission on Informatics. “This should include multisite validation, post-market monitoring of longitudinal performance, and other measures.”

Diagnostic radiologist Bibb Allen Jr., MD, FACR, from the ACR Data Science Institute adds, “The priority of the FDA in this space must be to ensure safety and effectiveness—effectively doing so will ultimately help establish trust and promote clinical adoption of AI/ML-enabled innovations.”