Development and Validation of a Magnetic Resonance Imaging-Based Machine Learning Model for TMJ Pathologies

Area of Science:

• Medical Imaging
• Machine Learning
• Radiomics

Background:

• Temporomandibular joint (TMJ) disorders present diagnostic challenges.
• Accurate classification of TMJ pathologies like condylar changes and disc displacements is crucial for effective treatment.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) model for classifying TMJ pathologies using magnetic resonance (MR) images.
• To assess the performance of various ML algorithms in predicting TMJ condylar bone changes and disc displacements.

Main Methods:

• A retrospective cohort study analyzed 214 TMJs from 107 patients.
• Radiomics features were extracted from MR images, including first-order statistics, shape, texture, GLCM, GLRLM, and GLSZM.
• Six ML classifiers (LR, RF, DT, KNN, XGBoost, SVM) were trained and evaluated using sensitivity, specificity, and ROC curves.

Main Results:

• K-nearest neighbors (KNN) and random forest (RF) demonstrated optimal performance in predicting TMJ pathologies.
• The models achieved high AUC, sensitivity, and specificity values on both training and testing datasets for condylar changes and disc displacements.
• Specific radiomic features were identified as significant predictors for condylar bone changes and disc displacements.

Conclusions:

• A machine learning model utilizing KNN and RF on TMJ MR images can effectively classify condylar changes and TMJ disc displacements.
• This approach offers a promising tool for the objective diagnosis and management of TMJ disorders.