Use of AI complicates legal liabilities for radiologists, study finds

Medical malpractice suits may face more complex challenges with the rise of artificial intelligence, according to a study led by Brown researchers that suggested that radiologists are viewed as more culpable if they fail to find an abnormality detected by AI.

The findings, published in June in the NEJM AI, have implications for patients, attorneys, and radiologists, says lead author Michael Bernstein, PhD, associate professor of radiology.

“There is a real potential for legal repercussions if radiologists fail to find an abnormality that AI correctly finds,” Bernstein says. “It could be worse for them than if they fail to find something with no AI in the first place.”

The study presented more than 1,300 participants with two hypothetical cases in which a radiologist failed to identify an abnormality and was being sued: In one, they didn’t detect bleeding in the brain of a stroke patient, and in the other they missed an abnormality in a patient's chest, ultimately leading to a delay in treatment and death. Participants essentially acted as jurors, determining whether or not the radiologist was liable for the oversight.

“We picked these as reasonable types of cases that would be presented to a real jury,” Bernstein says. “We did not recruit a sample of radiologists to participate, only everyday people.”

The study randomized participants to one of five conditions representing various levels of information pertaining to AI. The control left out AI entirely, while the other four conditions included the use of AI assisting the radiologist and whether or not the AI detected an abnormality. Additionally, Bernstein says, some participants were given information about error rates for AI.

The authors concluded that radiologists would be judged “more harshly” by a jury when they disagree with AI that discovers abnormalities. For example, in the brain bleed scenario, when no AI was described to participants, they sided with the plaintiff over 56 percent of the time. When both the AI and the radiologist did not find a brain bleed, participants sided with the plaintiff half of the time. However, when AI disagreed with the radiologist by instead finding the brain bleed, participants sided with the plaintiff nearly three-quarters of the time, which is a notable increase relative to the other two conditions. Bernstein says this shows that AI ultimately hurt the radiologist’s defense when it found a pathology the radiologist did not find.

“We were surprised by the magnitude of the effect of AI,” Bernstein says.

The percentages change when participants are made aware of AI’s error rates, however. Like before, when both the AI and the radiologist did not find a brain bleed, participants sided with the plaintiff half of the time. When participants were told the false omission rate—the proportion of AI results that AI incorrectly flagged as negative—that percentage fell to one-third.

“Those error rates clearly reduced the perceived liability when jurors were told them,” Bernstein says. “The percentages made them more radiologist friendly.”

Bernstein explains that radiologists may face an “AI penalty” if an AI system detects an abnormality that they missed, but jurors can be swayed by presenting data on AI’s error rates. As radiologists grapple with the emergence of AI, he says more research is needed in determining how to mitigate misconceptions about AI in diagnostic imaging.

“We think these findings might be due to the fact that people often fall into a trap of thinking that AI is magic and gets everything perfect. AI, like real radiologists, make errors. Our research shows that when you make people aware of the fact that AI isn’t perfect, people have greater sympathy for radiologists who contradict AI,” he says.

The study was coauthored by Grayson Baird, PhD, vice chair of research in the Department of Radiology, and researchers from Seton Hall University School of Law, the Pennsylvania State University College of Medicine, and Rhode Island Hospital. All authors are members of the Brown Radiology, Psychology and Law Lab.