AI in Radiology: Workflow & Legal Liability – Michael Bernstein, MD

The integration of artificial intelligence into radiology workflows is reshaping perceptions of medical liability, according to recent research. A study published in Nature Health on March 10, 2026, suggests that how radiologists interact with AI-assisted diagnoses influences how mock jurors assign blame in cases of missed diagnoses. This has implications for the adoption of AI in healthcare and the potential for reducing, rather than increasing, legal risks.

How Review Frequency Impacts Perceived Responsibility

Researchers from Penn State College of Medicine, Brown University and Seton Hall University School of Law presented mock jurors with a hypothetical scenario: a patient suffered irreversible brain damage because a radiologist failed to detect a brain bleed on a CT scan, despite the AI correctly identifying the scan as abnormal. The study revealed a significant difference in juror responses based on the radiologist’s review process. Jurors were nearly 50% more likely to find the radiologist liable when the radiologist only reviewed the scan once after receiving the AI’s feedback, compared to when the radiologist reviewed the scan twice – once before and once after the AI analysis. Penn State News details these findings.

This suggests that a more thorough review process, even with AI assistance, can mitigate perceptions of negligence. Michael Bruno, a professor of radiology and medicine at Penn State College of Medicine, emphasized that AI’s promise to improve healthcare quality and safety could be hampered by concerns about legal liability. This research offers a nuanced perspective on those concerns.

Understanding the Nuances of AI-Assisted Diagnosis

The study doesn’t imply that AI eliminates liability, but rather that the way AI is used affects how responsibility is assigned. The researchers highlight that the number of times a radiologist independently reviews a case is a key factor. This is particularly relevant as AI becomes more integrated into clinical practice. The findings align with broader discussions about AI’s impact on radiology workflow and liability, as discussed by Michael Bernstein, MD, who notes the potential for automation bias and its effect on patient safety.

Automation bias refers to the tendency to favor suggestions from automated systems, even when those suggestions are incorrect. The study’s results suggest that relying solely on AI’s initial assessment, without independent verification, may increase perceived liability. This underscores the importance of maintaining a critical and engaged approach to AI-assisted diagnosis.

The Study’s Design and Limitations

The research employed a randomized study design, presenting mock jurors with the same hypothetical case but varying the radiologist’s review process. This approach allows researchers to isolate the impact of review frequency on juror perceptions. However, it’s crucial to acknowledge the limitations inherent in using mock jurors. Their responses may not perfectly reflect the decisions of real jurors in actual legal cases. The hypothetical nature of the scenario also means that the emotional weight and complexity of a real-life malpractice case are absent.

the study focused specifically on brain bleeds detected via CT scans. The findings may not be generalizable to other types of radiological examinations or medical specialties. Additional research is needed to explore how AI integration affects liability perceptions across a broader range of clinical scenarios.

Implications for Healthcare Providers and Legal Frameworks

The study’s findings have important implications for healthcare providers adopting AI technologies. It suggests that clear protocols for AI integration, emphasizing independent verification and thorough review, could help mitigate legal risks. This includes documenting the radiologist’s review process, including both the initial assessment and the subsequent review after receiving AI feedback.

The research also raises questions for legal frameworks governing medical malpractice. Current legal standards often focus on whether a physician met the standard of care. As AI becomes more prevalent, it may be necessary to refine these standards to account for the role of AI in the diagnostic process. This could involve clarifying the responsibilities of both the physician and the AI system developer.

Ongoing Research and the Evolving Landscape

Michael H. Bernstein, one of the authors of the study, has been actively researching the legal implications of AI in radiology. A randomized study published in May 2025, also authored by Bernstein, further explores the impact of AI on perceived legal liability for radiologists. This ongoing work is crucial for understanding the evolving legal landscape surrounding AI in healthcare.

The integration of AI into radiology is not simply a technological shift; it’s a complex process with legal, ethical, and practical implications. The findings from this study and ongoing research provide valuable insights for healthcare providers, legal professionals, and policymakers as they navigate this evolving landscape. The key takeaway is that thoughtful implementation, emphasizing human oversight and thorough review, is essential for maximizing the benefits of AI while minimizing potential risks.

What comes next: Further research is planned to investigate the impact of different AI interface designs and levels of AI accuracy on liability perceptions. Legal scholars are also beginning to examine how existing malpractice laws apply to AI-assisted diagnoses, and whether new legislation is needed to address the unique challenges posed by this technology. Healthcare organizations are encouraged to develop and implement clear policies regarding AI integration, ensuring that radiologists receive adequate training and support.