Radiology societies weigh in on FDA’s push to regulate generative AI

Two major American groups for radiology doctors are sharing their views on the U.S. Food and Drug Administration's (FDA) plans to make rules for generative artificial intelligence (AI).

Back in August, the FDA asked the public for ideas about how to control this new technology. Generative AI can create things like pictures and sounds. The FDA recently stopped accepting comments and received more than 300 responses, according to a report from radiologist Rick Abramson, MD, who leads the FDA's Digital Health Center of Excellence.

The American College of Radiology (ACR) and the Radiological Society of North America (RSNA) recently published their comments online. Both groups say that keeping patients safe should be the most important goal. They also want the FDA to work closely with radiology professionals when creating rules and guidance for AI.

In comments sent on October 19, RSNA said patient safety should always come first. It said that even if the rules allow some flexibility, there must be clear safety requirements that can be enforced. These rules should make it easy to understand how AI works, make sure people and companies take responsibility, check that AI works safely in medical care, and keep watching how it performs when used with real patients.

RSNA explained that generative AI systems may keep changing after people start using them. These changes could create new dangers that cannot be caught by checking a product only once before it goes on the market. Instead, RSNA wants regulators to create a plan to keep checking new generative AI products over time. These checks should look at how well the AI works, how people use it, how it affects daily medical work, and whether it is being used in the right way.

RSNA believes the companies that make AI tools, not radiologists, should remain responsible for making sure these medical products are safe and work properly.

RSNA explained that hospitals, healthcare workers, professional groups, and organizations that create standards all have important jobs. Each can help based on what it can see and control. But their work should not replace the safety responsibilities of the companies that make AI products.

RSNA also believes the rules should require AI companies to quickly share clear, detailed information whenever they change their generative AI models or the way their systems are set up. The group wants users to have clear ways to report when AI does not work as expected and to ask for changes to be reviewed based on what they see while using the tools in real medical settings. Companies should also explain their systems clearly enough to help hospitals safely adjust AI tools for their own patients and the way their medical teams work.

Meanwhile, ACR also stressed that AI tools need to be checked regularly after they enter the market. This would help make sure doctors and hospitals use AI responsibly. ACR said it wants to keep working with the FDA and other radiology professionals to help the agency handle this challenge.

ACR shared several important goals for rules about generative AI. It wants the FDA to look at everything an AI tool does when used to care for patients, along with all the possible risks. ACR also wants the FDA to let AI technology improve and change over time. It believes the agency should balance flexible rules before products enter the market with careful checks of how they work in real medical settings. Most importantly, ACR wants regulators to measure AI mistakes that could actually affect patient care.

ACR explained that AI should be judged mainly on mistakes that matter for medical care. It said regulators should not focus only on general numbers showing how often AI makes up false information, gives different answers to the same question, or produces text that is similar to other text.

ACR said it can help the FDA decide which kinds of AI mistakes are important. These include making false statements, leaving out important information, comparing things incorrectly, and giving recommendations without enough evidence. The group wants the FDA to work closely with medical imaging experts when making these decisions. Before publishing draft rules, ACR suggests that the FDA hold a public workshop about radiology, arrange a meeting of its digital health advisory committee, or ask imaging specialists for their opinions.

ACR ended its letter by saying it would welcome more discussions with the FDA about these issues and other topics that affect radiology.

You can read the American College of Radiology's full comments here and the Radiological Society of North America's full comments here.