Systematic Review Finds AI Medical Imaging Systems May Perform Differently Across Racial and Ethnic Groups in the US
A growing body of research is examining whether artificial intelligence systems used in medical imaging diagnostics perform consistently across different patient populations. A new systematic review synthesizing studies conducted within the United States has found evidence that AI diagnostic tools may exhibit performance disparities when evaluated across racial and ethnic groups.
The review assessed multiple AI systems deployed or evaluated for tasks such as detecting diseases from X-rays, CT scans, and other imaging modalities. While the specific findings vary by system and clinical context, the overall evidence suggests that algorithmic models can produce unequal outcomes depending on the demographic characteristics of the patient population being analyzed.
Researchers note that several factors may contribute to these disparities, including imbalanced training datasets that underrepresent certain groups, differences in imaging acquisition parameters across clinical settings, and underlying variations in disease presentation. The findings highlight the importance of rigorous validation processes that test AI systems across diverse patient demographics before clinical adoption.
The review underscores the need for standardized reporting of demographic performance metrics in AI research and for regulatory frameworks that require bias auditing as part of medical AI approval processes. Addressing these disparities is seen as critical to ensuring equitable healthcare delivery as AI becomes more integrated into clinical workflows.