What question did this study set out to answer?

This research aims to compare the accuracy of various AI models in assisting clinical decisions for endodontic treatments in patients with systemic conditions.

February 12, 2026Odontology3 citationsOpen Access

Clinical decision accuracy in endodontic treatment of patients with systemic diseases: a comparative analysis using different artificial intelligence models

AEAyşegül EroğluErciyes University İAİpek Eraslan Akyüz EYEmre YılmazErciyes University

Key Points

This research aims to compare the accuracy of various AI models in assisting clinical decisions for endodontic treatments in patients with systemic conditions.
Conducted a scenario-based, cross-sectional educational study with 40 standardized clinical scenarios.
Developed scenarios based on endodontic and medical guidelines, reviewed by experts.
Queried four AI models using identical prompts in isolated environments.
Evaluated AI-generated responses by two calibrated endodontists using a 10-point scoring system.
No statistically significant differences in overall clinical decision accuracy among AI models (Friedman test, p > 0.05).
Categorical analysis showed a difference in high-accuracy responses (Cochran’s Q, p = 0.007).
No pairwise differences were significant in post hoc comparisons.
All models were comparably close to expert-defined optimal decisions, with higher variability for Gemini 2.5 Flash.

Abstract

Abstract This study aims to compare the clinical decision-making accuracy of different artificial intelligence (AI) models in endodontic treatment planning for patients with systemic diseases. A scenario-based, cross-sectional educational study was conducted using 40 standardized clinical scenarios representing ten commonly encountered systemic conditions affecting endodontic care. Scenarios were developed based on international endodontic and medical guidelines and reviewed by medical specialists and experienced endodontists. Four AI models, ChatGPT-5.1, Gemini 2.5 Pro, Gemini 2.5 Flash, and ChatGPT-3.5, were queried using identical, standardized prompts within fully isolated interaction environments to prevent contextual memory effects. AI-generated responses were independently evaluated by two calibrated endodontists using a predefined 10-point scoring system across four clinical domains. Clinical accuracy was categorized as high, partial, or incorrect. Nonparametric statistical analyses were performed. No statistically significant differences were observed among AI models in overall clinical decision accuracy or domain-specific scores (Friedman test, p > 0.05). Although categorical analysis revealed an overall difference in the proportion of high-accuracy responses (Cochran’s Q, p = 0.007), post hoc comparisons did not demonstrate significant pairwise differences. Deviation analysis revealed comparable proximity of all models to the expert-defined optimal decisions, with greater variability observed for the Gemini 2.5 Flash. Current AI models demonstrate comparable clinical decision-making performance in endodontic scenarios involving medically compromised patients. While descriptive trends were observed, no single model consistently outperformed others. AI systems may serve as supportive decision-making tools when used under professional supervision, but should not replace clinical judgment.

Ask AI

Helpful

Bookmark

View Full Paper