ACR wants the FDA to enhance artificial intelligence device transparency: 4 suggestions
The American College of Radiology is calling on federal health officials to enhance reporting pathways for artificial intelligence tools and ensure providers have access to the latest data required for optimizing patient care.
ACR submitted written comments to the Food and Drug Administration last week detailing its requests. The college first presented the FDA with some changes on Oct. 14 during a virtual workshop dedicated to AI-enabled medical device transparency.
As it stands, the administration has cleared “hundreds” of AI tools for radiological use, ACR noted, but imaging providers often don’t have the detailed metrics they need to make informed decisions.
“Having access to meaningful information could help radiology providers discover appropriate innovations for their clinical needs and estimate software performance when paired with their respective patient populations and subpopulations, image acquisition/input devices and imaging protocols,” ACR Board of Chancellors Chair Howard B. Fleishon, MD, wrote in the Nov. 15 letter.
Below are four key points from the document.
1). AI data must become transparent. Software performance often changes between institutions and radiologists need this information to make purchasing decisions. Product labels and performance data should also be made available to providers, patients and the general public.
2). The FDA should create a new mechanism for reporting performance concerns that is accessible to doctors and consumers. The administration’s current Medical Device Reporting pathway is not well-suited to capture issues with AI-enabled radiology tools, the ACR noted.
3). Similarly, the FDA must have manufacturers release data detailing model performance across populations, equipment and protocols. Most facilities don’t have the resources to test tools for each possible use case, making public documentation crucial. Manufacturers should also work with end-users to monitor performance across the software’s lifecycle to ensure device safety.
4). Testing characteristics must be publicly accessible, potentially formatted as a “food label” style, high-level summary. Required metrics should include population demographics, acquisition devices (manufacturer, model, etc.), performance measurements, user qualifications, and more.
“As always, the ACR welcomes the opportunity for further communication with FDA regarding transparency enhancements and AI/machine learning-enabled medical device oversight in general,” the college concluded.
You can read the entire letter here.
Matt joined Chicago’s TriMed team in 2018 covering all areas of health imaging after two years reporting on the hospital field. He holds a bachelor’s in English from UIC, and enjoys a good cup of coffee and an interesting documentary.