Diagnostic Accuracy of Educated Large Language Models in Endodontic Diagnosis and Case Difficulty Assessment

Accuracy of Educated Large Language Models Compared With Endodontic Experts in the Diagnosis and Difficulty Assessment of Endodontic Cases: A Diagnostic Test Accuracy Study

This diagnostic test accuracy (DTA) study aims to evaluate the diagnostic performance of educated large language models (Educated ChatGPT (GPT-5.5 Pro), Educated Gemini 3.1 Pro, and Educated Claude Opus 4.7) in endodontic practice. Their ability to establish pulpal and periapical diagnoses and assess endodontic case difficulty will be compared with the reference standard established by a panel of endodontic experts. Clinical and radiographic information from patients presenting for primary endodontic treatment or nonsurgical endodontic retreatment will be provided to both the AI models and the expert panel. The primary outcomes are the sensitivity, specificity, and the overall accuracy of the educated LLMs, with the objective of determining their potential role as reliable decision-support tools in endodontic diagnosis and treatment planning.

Inclusion Criteria:

1. Age above 16 years old.
2. Requiring primary endodontic treatment or retreatment.
3. Availability of complete clinical examination records.
4. Availability of diagnostic radiographs.
5. Restorable teeth.
6. Patient's acceptance to participate in the study.

Exclusion Criteria:

1. Incomplete records
2. Pregnant women.
3. No restorability: Hopeless tooth.
4. Traumatic dental injuries
5. Periapical radiographic images of sub-optimal quality or artifacts/high scatter interfering with proper assessment.