Decoding adolescent TMJ osteoarthritis with multimodal machine learning.

Area of Science:

• Orthopedics
• Radiology
• Machine Learning

Background:

• Adolescent temporomandibular joint osteoarthritis (TMJ-OA) diagnosis is critical for preventing lifelong pain and deformity.
• Early detection of degenerative changes during growth is essential for effective management.

Purpose of the Study:

• To identify key clinical and imaging predictors of adolescent TMJ-OA.
• To evaluate the performance of multimodal machine learning models for diagnosing TMJ-OA.

Main Methods:

• Evaluated diagnostic utility in 79 adolescents (10-18 years) with TMJ pain using panoramic radiography (PR) and MRI.
• Developed three decision tree models: clinical-only, imaging-only, and combined clinical-imaging.
• Utilized nested 5-fold cross-validation for robust model assessment.

Main Results:

• The imaging-only model showed high specificity (0.77) and accuracy (0.59), with PR evidence of TMJ-OA as the strongest predictor.
• The combined model improved sensitivity (0.61) and identified PR_TMJ_OA, MRI_TMJ_ADD, VAS score, and age as key predictors.
• PR alone achieved perfect specificity (0.97) but low sensitivity (0.38); clinical-only models had limited predictive performance.

Conclusions:

• Panoramic radiography (PR) is a useful screening tool for adolescent TMJ-OA but insufficient alone.
• Integrating clinical data with MRI findings enhances diagnostic accuracy and provides valuable decision-support tools for TMJ-OA.