ChatGPT-4o Proves Effective in Accurately Diagnosing Dermoscopic Images of Benign and Malignant Skin Conditions

Abstract

Introduction: The integration of artificial intelligence (AI) and AI chatbots (AIC) into medicine is progressing rapidly, particularly within dermatology, where clinical, dermoscopic, and histological pattern recognition is essential. Methods: To assess the diagnostic capabilities of ChatGPT-4o, ten skin lesions were selected: basal cell carcinoma (BCC), melanoma, invasive squamous cell carcinoma (SCC), actinic keratosis (AK), cherry angioma (CA), solar lentigo (SL), seborrheic keratosis (SK), blue nevus (BN), dermatofibroma (DF), and lichen planus (LP). For each condition, fifteen dermoscopic images were sourced from DermNet NZ and Dermoscopedia, yielding a total of 150 images. Rank order top 5 differential diagnosis were generated for each image and sensitivity, specificity, and accuracy were computed for each. Results: ChatGPT-4o correctly diagnosed 117 out of 150 (78%) images at the first attempt, and 144 out of 150 (96%) images within the top five differential diagnoses. Melanoma was correctly identified on initial diagnosis (15/15, 100% sensitivity, 91.1% specificity, 91.99% accuracy). Discussion: Our research demonstrates the promising potential diagnostic capabilities of ChatGPT-4o in accurately identifying various benign and malignant skin lesions using dermoscopic images, novel findings in comparison to prior studies.