Evaluation of Chat-GPT 5.1 for the Detection of Apical Lesions in Panoramic Radiography

Objective: The aim of this study was to evaluate the diagnostic performance of ChatGPT-5.1 in determining the presence or absence of apical lesions on panoramic radiographs based on visual input and to analyze the obtained results on a jaw-specific basis. Materials and Methods: A total of 207 anonymized panoramic radiographs were retrospectively analyzed. In each radiograph, the region in which an apical lesion was present was recorded as “lesion-present,” whereas the contralateral jaw region without an apical lesion on the same radiograph was considered “lesion-absent.” In this context, each lesion-present and lesion-absent region was treated as an independent unit of analysis. All evaluations were independently performed by ChatGPT-5.1 using standardized and anatomically restricted prompts that clearly defined the jaw (maxilla/mandible), side (right/left), and anatomical region. Model outputs were classified as true positive, true negative, false positive, or false negative. Sensitivity, specificity, accuracy, and F1 score were calculated for overall performance and on a jaw-specific basis.Results: Overall sensitivity, specificity, accuracy, and F1 score of ChatGPT-5.1 were 67.15%, 60.87%, 64.01%, and 65.11%, respectively. Tooth-level detection sensitivity was 67.6%. Mandibular performance was higher than maxillary performance (accuracy: 67.52% vs. 57.14%; tooth-level sensitivity: 69.89% vs. 63.04%). Concusion: ChatGPT-5.1 demonstrated a moderate level of diagnostic performance in detecting apical lesions on panoramic radiographs. The findings indicate that the model is not suitable for use as a standalone reliable diagnostic tool.