Diagnostic Performance of Artificial Intelligence in Detecting Root Canal Filling in Panoramic and Periapical Radiographs: A Retrospective Study

Aim: The current retrospective study aimed to evaluate the diagnostic performance of two artificial intelligence (AI) models, ThakaaMed Detect and Intelligent Reporting Software, in detecting root canal filling in panoramic and periapical radiographs. Materials and methods: The study included 1,000 radiographic images comprising 500 periapical and 500 panoramic radiographs, each showing at least one tooth with root canal filling. This study included four groups—group I: AI model—ThakaaMed Detect; group II: AI model— Intelligent Reporting Software; group III: Endodontist; and group IV: Radiologist. All the groups independently assessed the same set of 500 periapical and 500 panoramic radiographs with manual evaluation serving as the gold standard. Classification indicators—true positive, true negative, false positive, and false negative—were used to calculate performance metrics. To investigate the diagnostic performance of the AI models, sensitivity, specificity, predictive values, diagnostic accuracy, area under the receiver operating characteristic (ROC) curve (AUC), and 95% confidence interval (95% CI) of the AUC for AI models were computed. Results: The results showed that the diagnostic accuracy of the ThakaaMed Detect model was 98.3% and 92.7%, sensitivity 90.8%, 91.8%, specificity 98.8%, 93.3%, positive predictive value (+PV) 84.6%, 90.3%, and negative predictive value (−PV) 99.4%, 94.3% in panoramic and periapical radiographs, respectively. For Intelligent Reporting Software, the results showed that diagnostic accuracy was 93.9% and 96.7%, sensitivity 93.9%, 96.7%, specificity 100%, 92.4%, +PV 100%, 89.7%, and −PV 99.6%, 97.6% in panoramic and periapical radiographs, respectively. The diagnostic accuracy for the endodontist and radiologist was 100%. Conclusion: The findings of this study showed that both AI models, ThakaaMed Detect and Intelligent Reporting Software, exhibited high diagnostic performance in detecting root canal fillings in panoramic and periapical radiographs. ThakaaMed Detect showed superior accuracy in panoramic radiographs while Intelligent Reporting Software achieved high accuracy in both modalities. Clinical significance: AI models have shown accuracy in detection of root canal filling, suggesting their potential as valuable adjuncts in clinical practice, improving efficiency in radiographic assessment, particularly in busy clinical settings or for practitioners with limited experience in radiographic imaging.