Correlación diagnóstica entre modelos de inteligencia artificial y criterios clínicos radiográficos en patologías pulpares.

Abstract

Introduction: Artificial intelligence has gained relevance in the field of health for its great capacity to process clinical data to optimize care. In endodontics, large language models (LLMs) could be a supporting tool for identifying pulpar pathologies, the performance of which is subject to factors such as clinical case complexity. Objective: To determine the diagnostic correlation between artificial intelligence models and clinical-radiographic criteria in the diagnosis of pulpar and periapical pathologies. Materials and methods: A study was conducted with 156 clinical cases. The analysis was performed using SPSS (IBM, version 29), applying Cohen’s Kappa coefficient, the Friedman test with Bonferroni adjustment, and the Fisher-Freeman-Halton exact test. Results: There was a moderate and statistically significant agreement. No significant differences were observed between models, although concordance decreased as clinical complexity increased, particularly in Class III cases. Conclusions: ChatGPT, Gemini, and Claude showed potential usefulness as complementary diagnostic support tools, although their performance depends on the complexity of the case and does not replace professional clinical evaluation.