Machine Learning-Based Estimation of Knee Joint Mechanics from Kinematic and Neuromuscular Inputs: A Proof-of-Concept...

Area of Science:

• Biomechanics
• Biomedical Engineering
• Machine Learning

Background:

• Estimating tibiofemoral joint contact forces is crucial for understanding knee joint health and disease.
• Traditional musculoskeletal models can be computationally intensive and require extensive input data.

Purpose of the Study:

• To investigate the feasibility of using deep learning models to estimate in vivo tibiofemoral joint contact forces.
• To compare the performance of different deep learning architectures (biLSTM-MLP and TCN) for knee contact force prediction.

Main Methods:

• Developed and evaluated biLSTM-MLP and TCN models using the CAMS-Knee dataset.
• Employed a leave-one-subject-out validation strategy for robust performance assessment.
• Conducted leave-one-feature-out analyses to identify key predictive variables.

Main Results:

• The biLSTM-MLP model achieved high accuracy in predicting total knee contact forces (Ftot) during walking (RMSE: 0.16 BW, R: 0.98).
• Lower-limb kinematics and ground reaction forces were identified as primary predictors, while EMG data had minimal impact.
• The TCN model exhibited more variable performance compared to the biLSTM-MLP.

Conclusions:

• Deep learning models show significant promise for accurate and scalable estimation of knee contact forces.
• This approach provides a reliable alternative to conventional methods for personalized knee load assessment.
• Findings support further research with larger, diverse populations to refine deep learning applications in knee biomechanics.