International Journal of Medical Advances and Discoveries

International Journal of Medical Advances and Discoveries | Vol. 17, No. 8, August 2026 | pp. 80–86

DOI: 10.46882/2026/IJMAD/170880

Original Research Article

Title: Diagnostic Performance of Multimodal LLMs in Analyzing Complex Dermatological Lesions

Names of Authors: Yukihiro Tanaka¹, Kenji M. Sato²

Authors’ Affiliations: ¹Department of Dermatology, University of Tokyo Hospital, Tokyo, Japan; ²Artificial Intelligence Research Center, Tokyo Institute of Technology, Tokyo, Japan

Abstract: The clinical integration of Large Language Models (LLMs) trained on both textual and visual biomedical datasets has altered diagnostic paradigms. This multi-center diagnostic accuracy study evaluated the performance of a cutting-edge multimodal biomedical LLM (Med-VLM v4.0) in classifying complex or atypical dermatological lesions. A validation dataset comprising 1,500 clinical and dermoscopic images—spanning melanoma, basal cell carcinoma, atypical nevi, and severe inflammatory dermatoses—was compiled. The diagnostic accuracy, sensitivity, and specificity of the multimodal LLM were compared directly against the consensus diagnoses of a panel of board-certified dermatologists. For the primary task of distinguishing malignant melanoma from benign atypical nevi, the model achieved an Area Under the Receiver Operating Characteristic curve (AUROC) of 0.94 (95% CI: 0.91–0.96). The AI system demonstrated a diagnostic sensitivity of 91.4% and a specificity of 89.1%, matching the average performance of senior dermatologists (sensitivity: 90.2%, specificity: 88.5%; p = 0.42). Furthermore, when presented with rare, complex dermatological presentations, the LLM correctly identified the underlying pathology within its top-three differential diagnoses in 94.6% of test cases. Crucially, the model provided cohesive textual rationales that cited relevant morphologic features such as pigment networks and border irregularity. These results show that advanced multimodal LLMs can provide diagnostic guidance equivalent to expert clinical assessments, highlighting their value as reliable clinical decision-support systems in dermatology.

Keywords: Multimodal LLM, Artificial intelligence, Dermatological oncology, Melanoma detection, Dermoscopy, Diagnostic accuracy, Medical computer vision

Manuscript Timeline: Received: May 02, 2026; Revised: June 11, 2026; Accepted: July 01, 2026; Published: August 08, 2026

Citation: Tanaka, Y., & Sato, K. M. (2026). Diagnostic Performance of Multimodal LLMs in Analyzing Complex Dermatological Lesions. International Journal of Medical Advances and Discoveries, 17(8), 80–86. doi:10.46882/2026/IJMAD/170880

Subscribe to read the full article: https://www.internationalscholarsjournals.org/subscribe-to-read