Comparative diagnostic accuracy of ChatGPT in second mesiobuccal canal detection

BACKGROUND: The presence of a second mesiobuccal canal (MB2) in maxillary first molars is a critical factor in endodontic success. Accurate identification and treatment of this canal significantly reduce the risk of persistent infection and treatment failure. This study evaluated the diagnostic accuracy of ChatGPT in identifying MB2 canals using structured data derived from CBCT images. METHODS: This retrospective study included 400 CBCT scans obtained from patients aged 18–60. Structured data derived from axial, coronal, and sagittal CBCT slices were evaluated by both an experienced endodontist and ChatGPT. Images were preprocessed using ImageJ software for enhanced clarity. The model was asked to classify each case as MB2 present or absent. Model predictions were compared with expert evaluations. Diagnostic performance was assessed by calculating sensitivity, specificity, accuracy, positive and negative predictive values and Cohen’s Kappa coefficient. RESULTS: ChatGPT predicted MB2 presence in 228 cases, 227 of which were confirmed by expert evaluation. Eight false negatives and one false positive were observed. The model achieved 96.6% sensitivity, 99.4% specificity, and 97.8% overall accuracy, with a Cohen’s Kappa coefficient of 0.954 (p < 0.001). CONCLUSIONS: ChatGPT demonstrated high agreement with expert analysis using structured CBCT data. Despite not processing raw image data, the model showed strong pattern recognition capabilities. These findings support the potential of language models in clinical decision support, although further refinement and multimodal integration are recommended.