Evaluating the Ability of ChatGPT in Generating Differential Diagnosis in Clinicopathological Conferences of Neurodegenerative Disorders

Abstract

Abstract This study explores the utility of the artificial intelligence model ChatGPT in predicting neuropathologic diagnoses from clinical summaries presented at clinicopathological conferences. Twenty-five cases of various neurodegenerative disorders were analyzed using ChatGPT versions 3.5 and 4. The primary neuropathological diagnoses predicted by the models aligned with the official neuropathologic diagnoses in 32% (ChatGPT-3.5) and 52% (ChatGPT-4) of cases. When considering all proposed diagnoses, accuracy increased to 76% and 84% for ChatGPT-3.5 and ChatGPT-4, respectively. These findings highlight the potential of artificial intelligence tools like ChatGPT in neuropathology, suggesting they may facilitate more comprehensive discussions in clinicopathological conferences.