Artificial intelligence-generated images in skin cancer education: an experimental study
Jevan Cevik, Ishith Seth, Warren M. Rozen
Abstract
Introduction: Artificial intelligence (AI) has emerged as a powerful tool with considerable potential for enriching the quality of medical education. Generative adversarial networks, a branch of AI renowned for image-generation capabilities, particularly hold promise. Methods: In this study, we assess the ability of three prominent AI models—DALL-E, Midjourney and BlueWillow—to generate authentic-looking images of skin cancers to be used as educational resources for plastic surgery. These models were prompted to generate images of squamous cell carcinoma, basal cell carcinoma and melanoma. The AI-generated images were scrutinised by two plastic surgery residents and a specialist plastic surgeon, using a Likert scale for evaluation. Additionally, the images underwent an accuracy check using AI Dermatologist’s machine learning algorithm. Results: Only DALL-E had the ability to generate the distinctive features associated with each type of skin cancer, however its images cannot currently replace traditional patient photographs in terms of authenticity and practical learning value. The findings of AI Dermatology showed some discrepancies with those of the authors, indicating further refinement of its machine-learning algorithm is needed. Images created by Midjourney and BlueWillow had exaggerated fictionalised features that the authors deemed not appropriate for clinical education and could not be analysed by AI Dermatologist. Conclusion: The use of AI-generated images shows significant promise for augmenting traditional teaching practices in medicine.