Case-report–based Retrieval-Augmented Generation of Large Language Models for Identification of Malignant Melanocytic Lesions by Anatomic Site

Description

Supplementary Figure 1. Schematic Representation of Study Design and Three-Phase Workflow for Evaluating Multimodal LLM Performance in Dermoscopic Melanocytic Lesion Classification Supplementary Table 1. Case reports (n=648) with DOI Supplementary Table 2. Exact System and User Prompts for Phases 1 and 3 Supplementary Table 3. Raw study data generated using get-5-mini (n = 9,999), including ISIC image ID, histopathologically confirmed diagnosis, anatomical region, Phase 1 Diagnosis, and Phase 3 Diagnosis Supplementary Table 4. Raw study data generated using gemini-3.0-flash (n = 9,999), including ISIC image ID, histopathologically confirmed diagnosis, anatomical region, Phase 1 Diagnosis, and Phase 3 Diagnosis

Steps to reproduce

Full Schematic Representation of Study Design found in Supplementary Figure 1 1. Download images from the ISIC archive (see Supplementary Tables 3 and 4 for ISIC archive reference numbers) 2. Upload the dermoscopic image to both gpt-5-mini and gemini-3.0-flash attached as a multimodal image input with each API call with prompts (provided in Supplementary Table 2) 4. Provide structured diagnostic reference guide provided in Supplementary Table 1 along with dermoscopic image to both gpt-5-mini and gemini-3.0-flash attached as a multimodal image input with each API call

Institutions

• Sam Houston State University

  Texas, Huntsville
• University of Houston

  Texas, Houston

Categories

Licence