Ai-Assisted Prediction Model For Early Diagnosis Of IADD Progression

We aim to develop a diagnostic framework and predict the irreversible anterior disc displacement (IADD) in temporomandibular joint disorders (TMD), by collecting and analyzing Surface electromyography (sEMG), MRI, and synovial fluid(SF) from patients. This case-control study enrolled 120 subjects aged 20-60 (90 of IADD; 30 of controls), and TMD was clinically diagnosed by MRI. Exclusion criteria included pre-treatments, joint degeneration, and incomplete imaging. The sEMG signals from masticatory muscles were collected during chewing cycles and assessed for time-frequency entropy. The features of TMJ disc was extracted via PyRadiomics. SF were collected from patients for metabolomics and integrated with an machine-learning model. One neural network was used to integrate three individual factors to produce a final prediction. This network automatically focused on the most relevant data patterns to generate diagnoses, with decision thresholds optimized for >85% sensitivity. IADD in TMD was significantly related to sEMG instability (R=−0.68, p<0.001), while MRI showed a significant relation with synovial metabolomic biomarkers (pyruvate and seleno compounds, p<0.01). Attention mechanism analysis revealed balanced contributions from functional (33%), structural (42%), and molecular (25%) modalities. The tri-modal model exhibited better diagnostic performance (AUC=0.88, 95%CI 0.83-0.92) than single-modal approaches.The indicators suggested superior sensitivity (86%), specificity (82%), and accuracy (84%; F1-score=0.83) on the prediction of IADD in TMD. This diagnosis framework helps for early IADD risk prediction, offering clinically decision support to prevent irreversible TMJ degeneration.