Related Experiment Video
Updated: Jan 11, 2026
06:37
Temporomandibular Joint Pain Measurement by Bite Force and Von Frey Filament Assays in Mice
Published on: September 13, 2024
1.8K
An Interpretable Machine Learning Model Based on MRI Features for Predicting Pain Severity in Temporomandibular
Chuanfang Xu1, Xianyan Wu1, Shibin Li2
1Department of Radiology, The Affiliated People's Hospital of Fujian University of Traditional Chinese Medicine, Fuzhou, China.
Journal of Oral Rehabilitation
|November 18, 2025
Summary
Machine learning models using temporomandibular joint (TMJ) magnetic resonance imaging (MRI) features can predict temporomandibular disorders (TMD) pain severity. The best model identified age, disc position, and condylar movement as key predictors for objective pain assessment.
Area of Science:
- Medical imaging and machine learning applications in healthcare.
- Radiology and diagnostic imaging interpretation.
- Biomedical data science and predictive modeling.
Background:
- Chronic pain in the temporomandibular joint (TMJ) and masticatory muscles is a hallmark of temporomandibular disorders (TMD).
- The predictive value of specific magnetic resonance imaging (MRI) features for TMD-related pain intensity is not well-established.
- This research addresses the need for objective pain assessment in TMD patients.
Purpose of the Study:
- To develop and interpret machine learning (ML) models utilizing MRI characteristics for predicting pain severity in patients diagnosed with TMD.
- To enhance the clinical utility of TMJ MRI by correlating imaging findings with patient-reported pain levels.
- To identify key imaging biomarkers associated with TMD pain intensity.
Main Methods:
- A retrospective analysis of 755 TMJ MRI datasets from 584 TMD patients (2022-2024) was performed.
- Pain severity was quantified using the Visual Analogue Scale (VAS).
- Eleven ML models were trained and evaluated using demographic and detailed MRI features (e.g., disc position, effusion, bony changes), with performance assessed via AUC, accuracy, and SHAP interpretability. Feature selection was also explored.
Main Results:
- The LightGBM model achieved the highest predictive performance with an Area Under the Curve (AUC) of 0.899.
- Shapley Additive Explanations (SHAP) analysis highlighted age, disc position, and condylar movement as the most influential features in pain prediction.
- Optimized models using the top nine SHAP-ranked features demonstrated strong diagnostic performance (AUC = 0.829).
Conclusions:
- An interpretable, high-performing MRI-based ML model was successfully developed for objective pain assessment in TMD.
- The integration of imaging features and clinical data via ML and SHAP analysis offers a promising approach for identifying at-risk TMD patients.
- This methodology has the potential to guide personalized treatment strategies for individuals suffering from TMD-related pain.
