Utilizing GPT-4 and generative artificial intelligence platforms for surgical education: an experimental study on skin ulcers

Background The advancement of artificial intelligence (AI), specifically Generative Adversarial Networks (GANs), offers exciting possibilities for the enhancement of medical education, with its image-generation capabilities becoming a topic of interest. This novel study evaluates the aptitude of combining a large language model, ChatGPT, with GANs DALL-E 2, Midjourney, and Blue Willow in producing authentic images of ulcers, with a goal to enrich educational resources for surgery. Methods First, ChatGPT-4 was prompted with definitions of different skin ulcers, and its response was inputted into the GAN models. Generated AI images were evaluated by four board-certified plastic surgeons and three plastic surgeon residents with extensive experience using a Likert scale. Results Among the three GANs, only DALL-E showed an acceptable level of accuracy, portraying the unique characteristics of each ulcer type. However, it cannot replace conventional patient photographs in terms of authenticity and educational value. Despite presenting aesthetically pleasing images, Midjourney and Blue Willow produced highly stylized, exaggerated features unsuitable for clinical education. Conclusions Despite these shortcomings, the future of AI-generated images remains promising, given the continuous progress of technology, in augmenting traditional medical education methodologies. Level of evidence: Not gradable.