Assessing ChatGPT as an Educational Tool for Image Generation in Dermatology

Comprehensive Summary

This research letter investigates whether ChatGPT can generate accurate dermatologic images of different skin tones for educational purposes. Using a standardized sentence structure, ChatGPT-5 was prompted to generate images of atopic dermatitis (AD) and psoriasis on Fitzpatrick skin types 1 through 5. The prompts were randomized by condition and skin type and each request was initiated in a separate chat session. One reviewer qualitatively evaluated the images for accuracy of the pigmentation and lesions. ChatGPT produced images that generally corresponded with Fitzpatrick pigmentation categories, except in type 3, which appeared to be darker than type 4. AD was typically shown as patches or thin plaques and psoriasis was shown as demarcated plaques with scale. The generated images of AD and psoriasis were not accurately representative of the real skin conditions, limiting their applicability in the real world. The authors conclude that although ChatGPT shows potential as an educational image generator, it lacks the precision for accurate teachings.

Outcomes and Implications

With limited exposure to diverse skin tones in dermatology, it can contribute to diagnostic disparities. AI-generated images could bridge this gap and expand the representation without needing to worry about patient privacy. ChatGPT could be a useful supplemental learning too, however, its current limitations prevent it from being utilized in a clinical or educational context. It needs to be further refined before broader use.