Comparative Assessment of Diagnostic Concordance Among Clinical Examination, ChatGPT-4, Google Lens, and Histopathology in Dermatological Lesions

Abstract

Introduction: Artificial intelligence-supported systems are increasingly utilized in dermatology as adjunctive diagnostic tools. Objectives: To evaluate the diagnostic performance of ChatGPT and Google Lens in skin lesions requiring biopsy, comparing the concordance of dermatologists' clinical diagnoses and AI-generated diagnoses with histopathological findings. Methods: In this prospective observational study, 152 adult patients with biopsy indications were included. Clinical diagnoses were established by dermatologists. Standardized lesion images were uploaded to ChatGPT (GPT-4 model, accompanied by standardized anamnesis and localization) and Google Lens (single clinical image). Diagnoses were compared with histopathological results to calculate accuracy rates using SPSS 26.0. Results: Mean patient age was 52.37 ± 16.63 years (55.3% male). Dermatologists achieved significantly higher diagnostic accuracy than ChatGPT (86.2% vs. 56.6%; absolute difference: 29.6 percentage points; 95% CI: 20.7–37.9) and Google Lens (86.2% vs. 31.6%; absolute difference: 54.6 percentage points; 95% CI: 45.4–62.2). By disease subtype, dermatologists outperformed both AI systems in inflammatory and drug-induced dermatoses, whereas for autoimmune bullous diseases, dermatologists and ChatGPT showed significantly higher accuracy than Google Lens. Conclusions: Under the defined protocol conditions, clinical evaluation demonstrated higher diagnostic concordance than ChatGPT and Google Lens.