Artificial Intelligence-generated Clinical Images for Plastic Surgery Education: A Prospective Randomized...

High-quality clinical images are essential for plastic surgery education and board examinations, but privacy concerns and limited case diversity constrain their use. The educational validity of generative artificial intelligence (AI) images remains uncertain.

In a prospective randomized double-blind study, 47 plastic surgeons (40 residents, 7 senior surgeons) each evaluated 60 randomized AI-generated and real clinical images depicting board-style scenarios. Participants rated realism and educational usefulness (0-10), suitability for replacing real photographs (yes/no), and artificial elements (yes/no). Participant-level means and proportions were compared using paired t tests and Wilcoxon tests. Test-retest reliability at 1 month was assessed using intraclass correlation coefficients.

Realism ratings did not differ significantly between AI-generated and real images (8.46 versus 8.36; P = 0.5044). AI-generated images received higher educational usefulness ratings (9.13 versus 8.75; P = 0.0035). Replacement endorsement did not differ significantly (0.95 versus 0.93; P = 0.4140). Artificial elements were reported less frequently in AI-generated images (0.02 versus 0.15; P < 0.0001). Findings were similar across experience levels. All aggregated measures showed good test-retest agreement (intraclass correlation coefficients > 0.8).

AI-generated images received high realism ratings and slightly higher educational usefulness ratings than real photographs in this cohort. These findings support curated AI imagery as a potentially privacy-preserving educational resource. Diagnostic performance, learning outcomes, and demographic bias require evaluation before widespread curricular or examination use.