Knowledge-Based Deep Learning for Time-Efficient Inverse Dynamics.

Area of Science:

• Biomechanics
• Computational Biology
• Neuroscience

Background:

• Accurate muscle activation and force estimation is crucial for neuro-rehabilitation and musculoskeletal disorder treatments.
• Computational musculoskeletal modeling is a key non-invasive tool, but traditional inverse dynamics methods are computationally intensive.
• Existing methods often require extensive labeled data and significant computation time, limiting their clinical applicability.

Purpose of the Study:

• To develop a time-efficient knowledge-based deep learning framework for inverse dynamic analysis.
• To enable direct prediction of muscle activation and forces from joint kinematic data without requiring labeled training data.
• To improve the speed and accessibility of musculoskeletal modeling for clinical applications.

Main Methods:

• A knowledge-based deep learning framework utilizing a Bidirectional Gated Recurrent Unit (BiGRU) neural network was developed.
• Physical knowledge from forward dynamics and physiological criteria were integrated into the loss function to guide network training.
• The model was trained and validated on upper and lower limb movement datasets from healthy subjects.

Main Results:

• The proposed BiGRU model demonstrated superior performance compared to other neural network architectures.
• The framework achieved time-efficient inverse dynamic analysis, accurately predicting muscle activation and forces.
• Integration of prior physical knowledge significantly enhanced the model's effectiveness and robustness.

Conclusions:

• The developed knowledge-based deep learning framework offers a robust and efficient solution for musculoskeletal modeling.
• This approach accelerates the estimation of muscle activation and forces, supporting advancements in neuro-rehabilitation and musculoskeletal disorder treatment.
• The findings highlight the potential of integrating prior physical knowledge into deep learning for complex biomechanical analyses.