ChatGPT-4o as a diagnostic tool for skin cancer: Diagnostic accuracy in melanoma and non-melanoma detection

Comprehensive Summary

The study evaluated whether ChatGPT-4o could accurately diagnose melanoma and non-melanoma skin cancers using clinical images of skin lesions. Researchers analyzed 90 histopathologically confirmed cases, providing the AI with macroscopic images and combined macroscopic-dermoscopic images and asking it to generate one primary diagnosis and three differential diagnoses. The model achieved moderate accuracy (about 66-73%), with slightly higher accuracy when multiple possible diagnoses were considered, though dermoscopic images did not consistently improve performance. The authors concluded that while ChatGPT-4o shows potential as a clinical decision-support tool, its inconsistent accuracy means it should not replace dermatologists for diagnosing skin cancer.

Outcomes and Implications

The findings of this study suggest that artificial intelligence tools such as ChatGPT-4o could play a supportive role in dermatology by assisting clinicians with the initial evaluation of skin lesions. With moderate diagnostic accuracy, AI systems may help flag potentially suspicious lesions and contribute to earlier detection of skin cancer, particularly in settings with limited access to dermatologists. However, the model’s inconsistent performance and tendency to recommend biopsies for all lesions highlight the risk of overdiagnosis and unnecessary procedures if used independently. Therefore, AI technologies should be integrated cautiously as decision-support tools alongside trained medical professionals, with further research needed to improve accuracy and reliability before widespread clinical use.