AI-generated physician images show sex and race bias vs real-world demographics

Cross-sectional study: DALL-E 3 generated 78% female critical care physicians vs real-world 27% (P<0.001); Midjourney 9% female. Non-White 90% vs 46% real. AI images biased, may skew education & perception.

• Design: 300 AI-generated images (DALL-E 3, Midjourney 6.0, Stable Diffusion 3.0) compared with AAMC physician workforce data for adult and pediatric critical care.
• Other platforms: Stable Diffusion showed 60% female for pediatric CCM vs 49.6% real (P<0.001); Midjourney underrepresented across categories.
• Limitation: Single study limited to critical care; findings may not generalize to other medical imaging contexts or specialties.

RadPigeon summaries are original and for information only. They are not clinical advice.