ChatGPT-4o在不同菲茨帕特里克皮肤类型的皮肤病诊断中的性能评估

Performance Evaluation of ChatGPT-4o in Dermatological Diagnoses Across Fitzpatrick Skin Types.

作者信息

Patel Housley Priya, Samia Arthur M, Motaparthi Kiran

机构信息

Department of Dermatology, University of Florida, Gainesville, USA.

出版信息

Cureus. 2025 Jul 6;17(7):e87368. doi: 10.7759/cureus.87368. eCollection 2025 Jul.

DOI:10.7759/cureus.87368

PMID:40765600

原文链接:https://pmc.ncbi.nlm.nih.gov/articles/PMC12323556/

Abstract

ChatGPT-4o, a multimodal AI model released in May 2024, offers free image analysis capabilities that patients may access prior to seeking medical care. We evaluated ChatGPT-4o's diagnostic performance across Fitzpatrick skin types (FSTs) using 324 biopsy-confirmed dermatologic images. The model generated three differential diagnoses and malignancy assessments per image, which were compared to histopathologic ground truths. Results showed significantly lower sensitivity, specificity, and accuracy for melanoma in darker skin tones (FSTs III-VI) compared to lighter tones (FSTs I-II). These findings highlight the potential risk of biased AI performance in underrepresented populations and underscore the need for more inclusive dermatologic datasets.

摘要

ChatGPT-4o是2024年5月发布的多模态人工智能模型，具有免费的图像分析功能，患者在寻求医疗护理之前即可使用。我们使用324张经活检证实的皮肤病图像评估了ChatGPT-4o在不同菲茨帕特里克皮肤类型（FST）中的诊断性能。该模型为每张图像生成了三种鉴别诊断和恶性肿瘤评估结果，并与组织病理学的真实情况进行了比较。结果显示，与较浅肤色（FST I-II）相比，黑色素瘤在较深肤色（FST III-VI）中的敏感性、特异性和准确性显著降低。这些发现凸显了人工智能在代表性不足人群中表现出偏差的潜在风险，并强调了需要更具包容性的皮肤病数据集。