Advances in radiology AI raise thorny medicolegal concerns

2021 01 12 19 59 7375 Computer Legal Justice Scales 400

How will artificial intelligence (AI) technology affect the medicolegal liability of radiology when it eventually begins outperforming humans? A team of authors tackled the topic in an article published February 1 in the Journal of the American College of Radiology.

"Although the [advanced AI] scenarios we envision here will likely not arrive in clinics and hospitals anytime soon, the time to plan for them is now," wrote John Banja, PhD, of Emory University, Dr. Rolf Dieter Hollstein of Advanced Radiology Services, and Dr. Michael Bruno of Penn State Milton S. Hershey Medical Center.

The authors noted that liability among clinicians and vendors is best managed by using advanced AI models that can explain the rationales behind their decisions, rather than simply operating as a 'black box.' "

"In any event, radiologists will need to be involved in developing the content of explanatory models for physician-to-physician communications and for physician to patient discussions," they wrote.

When clinicians disagree in the future with an AI model's suggested diagnosis and its related rationale(s), several strategies should be available to diminish their liability, according to Banja, et al. These could include utilizing an adjudicator algorithm developed by professional groups to serve as a tiebreaker, or simply discussing the disagreement directly with the patient as part of an informed consent or shared decision-making approach.

Furthermore, healthcare organizations and their vendors could develop contractual agreements on how liability will be apportioned in the event of malpractice verdicts or settlements. Another strategy could be to have vendors assume total liability for a system's mistake when a clinician simply signs off on an autonomous model's decision, according to the authors.

They also noted that liability for clinicians or AI software providers could be relieved if regulatory, judicial, or legal bodies enact or rule on policies. And if AI ever serves autonomously -- like the infamous HAL computer in the movie 2001: A Space Odyssey -- clinicians might consider reducing their liability by arguing that the technology should have legal standing as a "person," according to the authors.

"Still, it would be reasonable to anticipate that [advanced AI systems] would significantly reduce the frequency and severity of harm-causing errors, thereby realizing an overall reduction in physicians' malpractice actions and payouts and, presumably, a significant

reduction in their malpractice premiums," they wrote.

The authors concluded that tight vendor-clinician business relationships will be inevitable in the development and deployment of advanced AI models. But these relationships will also commingle their legal exposure.

"Consequently, it is in everyone's best interests to develop liability-sharing arrangements that are informed by the best analytical and strategic thinking available," they wrote. "The legal and ethical principles and theories undergirding these arrangements should be publicly discussed, analyzed, constructed, and reconstructed to secure the most reasonable liability protection available."

Advanced AI will also likely witness a rocky road in its early development and implementation, the authors said.

"There will be plenty of bugs, glitches, technology breakdowns, communications failures, misunderstandings, and inconsistent or confusing judicial decisions handed down, along with regulatory efforts resulting in unintended consequences," they wrote.

Malpractice actions will result from these bumps in the road, reflecting the learning curve of implementing these advanced AI algorithms, according to the authors.

"But if the history of the imaging sciences has taught us anything thus far, it is that continued technological progress in image recognition will continue relentlessly, despite the difficulties," the authors wrote. "Consequently, conducting increasingly concentrated and informed analyses of the issues we raise here seems an imperative."

Page 1 of 364
Next Page