Racial and Ethnic Disparities in the Diagnostic Performance of AI Systems for Medical Imaging in the United States: A Systematic Review.
Authors
Affiliations (6)
Affiliations (6)
- General Practice, Leeds Teaching Hospitals NHS Trust, Leeds, GBR.
- Radiology, Afe Babalola University Multi-system Hospital, Ado Ekiti, NGA.
- Cardiothoracic Surgery, Beacon Hospital, Dublin, IRL.
- Family Medicine, Alberta Health Services, Edmonton, CAN.
- Internal Medicine, Bridgeport Hospital, Bridgeport, USA.
- Radiology, University of Calabar, Calabar, NGA.
Abstract
AI-assisted diagnostic systems are increasingly being integrated into radiology and medical imaging practice in the United States. Concerns have emerged regarding differences in diagnostic performance across racial and ethnic populations, particularly among historically underserved groups. This study aimed to systematically review evidence on racial and ethnic disparities in the diagnostic performance of AI systems used in radiology and medical imaging in the United States. A systematic search of major databases identified 27 eligible studies published between 2015 and 2026. The included studies primarily consisted of retrospective observational and validation studies evaluating AI systems in chest radiography, breast ultrasound, and ophthalmologic imaging, as well as selected non-imaging predictive models. Several studies reported reduced diagnostic performance, lower sensitivity, higher rates of underdiagnosis, or subgroup disparities among racial and ethnic minority populations, particularly Black and Hispanic patients. Some studies also demonstrated that fairness-aware model development and subgroup validation may improve equity in AI performance. The findings were synthesized narratively and support the need for diverse training datasets, transparent demographic reporting, standardized fairness evaluations, and prospective multicenter validation before the widespread clinical implementation of AI systems in radiology and medical imaging.