Clinical and MRI Markers for Acute vs Chronic TMD Using a Machine Learning and Deep Learning Approach

Abstract

This study aimed to identify factors that significantly contribute to the chronicity of symptoms in patients with temporomandibular disorders (TMD). Statistical, machine learning, and deep learning models were used for analysis. The study include 239 patients with TMD (161 women and 78 men; mean age 35.60 ± 17.93 years), diagnosed using Diagnostic Criteria for TMD (Axis I). Participants were categorized into: acute TMD (< 6 months) and chronic TMD (≥ 6 months) (51.05%). Significance clinical findings revealed that temporomandibular joint (TMJ) noise and bruxism were more frequently reported in patients with chronic TMD than in those with acute TMD. The visual analog scale (VAS) score, reflecting subjective pain intensity, was significantly higher in chronic TMD than in acute TMD. Additionally, patients with chronic TMD had shorter average sleep durations than their acute counterparts. STOP-Bang total scores were higher in chronic TMD (3.02 ± 2.09 vs. 2.39 ± 1.73, p = 0.012) than in acute TMD. Magnetic resonance imaging revealed structural abnormalities in patients with chronic TMD, with significantly higher rates of anterior disc displacement (ADD), TMJ osteoarthritis, and joint space narrowing. Using logistic regression—a widely recognized machine learning model—the AUROC for predicting chronic TMD was 0.7550 (95% CI]: 0.6550–0.8550). Significant predictors of chronic TMD included TMJ noise, bruxism, VAS score, sleep disturbance, STOP-Bang total score ≥ 5 (high risk of obstructive sleep apnea), ADD, and joint space narrowing. Deep learning (multilayered perceptron) improved prediction performance by 3.99% over logistic regression (79.49% vs. 75.50%, p = 0.3067). These findings may help clinicians prevent symptom chronicity and mitigate the progression to chronic TMD, ultimately improving patient care.