Development of AI-based tools for assessing temporomandibular joint disorders using MRI images

Temporomandibular joint disorders (TMDs) are diagnostically challenging due to the complexity of MRI interpretation and high inter- observer variability among clinicians. Therefore, it is of interest to develop and evaluate artificial intelligence-based tools for automated assessment of TMDs using magnetic resonance imaging. Hence, a total of 2,847 TMJ MRI examinations were used to train and test deep learning models for disc displacement classification, osteoarthritic change detection and joint effusion identification. The convolutional neural network achieved diagnostic accuracies of 94.2%, 91.8% and 93.5%, respectively, with area under the ROC curve values exceeding 0.92 and strong agreement with expert radiologists (κ = 0.87-0.91). The AI system reduced interpretation time by 68%, demonstrating its potential to improve diagnostic accuracy, consistency and efficiency in clinical TMJ evaluation.