Comparative Analysis of Diagnostic Accuracy and Case Difficulty Assessment of Three Large Language Models

Overview

This prospective, blinded diagnostic accuracy study aims to compare the performance of three large language models-ChatGPT (GPT-5.5 Pro), Gemini 3.1 Pro, and Claude Opus 4.7-in endodontic diagnosis and case difficulty assessment. The models will be evaluated against expert consensus as the reference standard using standardized clinical data and periapical radiographs. Diagnostic accuracy, sensitivity, specificity, and agreement with expert consensus will be assessed to determine the potential of LLMs as clinical decision-support tools in endodontics.

This prospective, blinded diagnostic accuracy study aims to compare the diagnostic accuracy and endodontic case difficulty assessment of three LLMs-ChatGPT (GPT-5.5 Pro), Gemini 3.1 Pro, and Claude Opus 4.7-with expert consensus as the reference standard. Adult patients requiring primary endodontic treatment or retreatment will undergo routine clinical examination, pulp sensibility testing, and periapical radiographic assessment. Standardized clinical information and radiographs will be provided to each LLM and to four experienced endodontists independently. Expert consensus, defined as agreement among at least three of four experts, will serve as the reference standard. The primary outcomes are diagnostic accuracy, sensitivity, specificity, and agreement with expert consensus for pulpal and periapical diagnosis and endodontic case difficulty assessment according to the American Association of Endodontists (AAE) criteria. The findings will provide evidence regarding the reliability of LLMs as clinical decision-support tools in endodontics and their potential role in improving diagnostic consistency and treatment planning.

Conditions

Interventions