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

Area of Science:

• Endodontics
• Artificial Intelligence in Dentistry
• Diagnostic Accuracy

Background:

• Accurate endodontic diagnosis is critical but challenging in clinical practice.
• Artificial intelligence (AI), specifically large language models (LLMs), is emerging as a potential clinical support tool.

Purpose of the Study:

• To compare the diagnostic performance of three leading LLMs (ChatGPT-5, Gemini, Perplexity) against an experienced endodontist.
• To evaluate the accuracy, sensitivity, and specificity of LLM-based diagnostic support in endodontics.

Main Methods:

• An experienced endodontist evaluated 40 anonymized clinical cases.
• The diagnostic accuracy of ChatGPT-5, Gemini, and Perplexity was compared to the expert's judgment.
• Statistical analysis included accuracy and Cohen's kappa coefficient (κ).

Main Results:

• ChatGPT-5 achieved perfect agreement (100% accuracy, κ=1.00) with the expert.
• Gemini demonstrated 97.5% accuracy (κ=0.95) with one false positive.
• Perplexity showed 92.5% accuracy (κ=0.85), missing three positive cases, but maintained full specificity.

Conclusions:

• State-of-the-art LLMs can achieve expert-level diagnostic performance in endodontics.
• ChatGPT-5 and Gemini exhibited superior sensitivity, while ChatGPT-5 and Perplexity demonstrated full specificity.
• Further validation is required before widespread clinical adoption of LLMs for endodontic diagnosis.