Artificial Intelligence-Based Approach for Determining the Risk of Temporomandibular Disorders

This study aims to predict TMD using ML approaches based on clinical and sociodemographic variables to aid in the early detection and risk assessment.

The study was on patients with and without TMD who met the inclusion criteria in a retrospective-prospective manner. Clinical and demographic variables of 340 patients were collected during the study. Data were split into 80% training and 20% test sets. Five different machine learning models (DANets, KAN, KNN, MLP, SVM) were used. Model performance was evaluated using accuracy, weighted precision, recall, and F1-score. A model interpretability analysis was performed using SHapley Additive ExPlanations (SHAP).

For result classification, the DANets model achieved the highest performance (accuracy = 0.8529). The DANets model achieved the strongest performance in diagnostic classification by far (accuracy = 0.8382, F1 = 0.8212). SHAP-based analyses revealed that demographic factors (especially age and gender) and certain clinical symptoms (e.g., joint pain, clicking, and limited mouth opening) had the highest contribution to the models' decision-making processes.

AI-based models-particularly deep learning algorithms-may provide substantial support to clinicians in the diagnosis and risk assessment of TMD. These tools may reduce diagnostic uncertainty, especially for inexperienced practitioners, facilitate early diagnosis and referral processes, and decrease the risk of TMD progression.