Explainable deep learning and biomechanical modeling for TMJ disorder morphological risk factors

Area of Science:

• Biomedical Engineering
• Computational Biology
• Musculoskeletal Research

Background:

• Understanding multifactorial musculoskeletal disorders like temporomandibular joint (TMJ) disorders is crucial for effective prevention and treatment.
• Deep learning excels at identifying risk factors but lacks mechanistic insights for clinical application.
• Multiscale biomechanical modeling provides mechanistic understanding within physiological contexts.

Purpose of the Study:

• To develop and apply a hybrid approach integrating 3D explainable deep learning and multiscale biomechanical modeling.
• To investigate the etiology of TMJ disorders by identifying risk factors and their mechanistic links to joint biomechanics.
• To enhance the clinical applicability of deep learning in etiological investigations.

Main Methods:

• A 3D convolutional neural network was employed to identify patient-specific morphological features associated with TMJ disorders.
• Explainable deep learning outputs were used to drive multiscale biomechanical models.
• Biomechanical models simulated joint forces, tissue nutrient availability, cellular ATP production, and disc strain energy density.

Main Results:

• The deep learning model accurately recognized TMJ disorder patients based on condylar, ramus, and chin morphology.
• Small mandibular size and flat condylar shape were identified as risk factors for TMJ disorders.
• These morphological factors were mechanistically linked to increased joint force, reduced nutrient/ATP levels, and elevated disc strain energy density.

Conclusions:

• The hybrid approach successfully combined deep learning's risk identification with biomechanical modeling's mechanistic explanations.
• This integration addresses limitations in deep learning's clinical translational confidence for etiological studies.
• The method enhances accessibility for analyzing smaller clinical datasets by incorporating essential biomechanical context.