知镜 Cognispect · 医学文献与研究工作台

Image-based diagnostic accuracy of artificial intelligence models in acute otitis media and otitis media with effusion: a comparison of ChatGPT and Gemini.

Alpay M, Aslier NGY.
European archives of oto-rhino-laryngology : official journal of the European Federation of Oto-Rhino-Laryngological Societies (EUFOS) : affiliated with the German Society for Oto-Rhino-Laryngology - Head and Neck Surgery · 2026-11-08 · PMID 43000558

文献摘要

Objective The aim of this study is to comparatively evaluate two multimodal large language models (LLMs); ChatGPT (GPT-5.2) and Gemini 3.0 Flash, in terms of diagnostic accuracy, reliability, and clinical consistency for the diagnosis of pathologic tympanic membrane (PTM), acute otitis media (AOM) and otitis media with effusion (OME) prompting oto-endoscopic images. Methods This prospective diagnostic accuracy study included 515 high-resolution tympanic membrane (TM) images from patients aged between 6 months and 80 years. Nine questions, including final reference diagnoses (normal, AOM or OME) were set and the reference standard was based on concordant independent assessments by two senior otolaryngologists who were blinded to AI outputs. Both models were evaluated using the same standardized zero-shot prompt and the same nine-item structured image-assessment framework. Expert-model agreement was assessed using Cohen's kappa. Diagnostic performance was summarized using sensitivity, specificity, positive and negative predictive values, accuracy, and area under the receiver operating characteristic curve (AUC), with 95% confidence intervals; AUCs were compared using the DeLong method. Results ChatGPT (GPT-5.2) and Gemini 3.0 Flash showed good agreement (κ = 0.662, κ = 0.697 respectively; p Conclusion LLMs offer significant potential in image-based diagnosis of OME and AOM. Gemini 3.0 Flash demonstrated higher diagnostic accuracy for identifying pathologic TMs compared with GPT-5.2. However, both models demonstrate similar accuracy with different sensitivities and specificities for OME and AOM, suggesting that AI-based systems may be better positioned as clinical decision support tools to complement clinical evaluation rather than as standalone diagnostic solutions.

在知镜精读原文来源DOI 原文

知镜提供医学文献检索、阅读笔记、人体图谱、统计绘图与研究复现工具。