Large Language Models for Endodontic Diagnosis: A Comparative Study Against an Expert Reference Standard

Accurate diagnosis is the cornerstone of successful endodontic care, yet it often remains a challenge in daily practice. Artificial intelligence (AI), particularly large language models (LLMs), is being explored as a possible support tool for clinicians. In this study, we compared the diagnostic ability of three LLMs: ChatGPT-5, Gemini, and Perplexity with the judgment of an experienced endodontist across 40 anonymized clinical cases. ChatGPT-5 reached perfect agreement with the expert (100% accuracy, κ = 1.00). Gemini (97.5% accuracy, κ = 0.95) and Perplexity (92.5% accuracy, κ = 0.85) also performed well but showed different error patterns: Gemini made one false positive, while Perplexity missed three positive cases. Overall, ChatGPT-5 and Gemini showed the best sensitivity, and ChatGPT-5 and Perplexity maintained full specificity. These results show that the latest LLMs can approach expert-level diagnostic performance, but further testing is needed before they can be relied upon in everyday clinical practice.