Integrating Machine Learning with Musculoskeletal Simulation Improves OpenCap Video-Based Dynamics Estimation.

Area of Science:

• Biomechanics
• Kinetics and Kinematics
• Machine Learning in Healthcare

Background:

• Musculoskeletal dynamics are crucial for understanding and treating movement disorders.
• Current methods for estimating whole-body dynamics from accessible tools like smartphone videos lack accuracy and physical realism.
• Existing physics-based and machine learning (ML)-based dynamic prediction approaches have limitations in achieving high precision.

Purpose of the Study:

• To develop and validate a hybrid ML-simulation framework for enhanced estimation of musculoskeletal dynamics from smartphone video kinematics.
• To improve the accuracy of predicting ground reaction forces, joint moments, and joint contact forces.
• To enable more accessible and scalable assessments of musculoskeletal dynamics.

Main Methods:

• Utilized ML models to predict ground forces and centers of pressure from video-based kinematics.
• Developed a hybrid framework integrating ML predictions with dynamic simulation for consistency.
• Compared the hybrid model against kinematic-tracking simulations and ML-only inverse dynamics using lab-based marker and force plate data from walking individuals.

Main Results:

• The hybrid and ML approaches reduced vertical ground force error by 40-44% compared to simulation alone.
• The hybrid model demonstrated superior accuracy in joint moment estimation (29-45% improvement) and joint contact force estimation (12-13% improvement).
• Significant improvements were observed in peak medial knee contact force (49%) and knee adduction moment impulse (30%).

Conclusions:

• The hybrid ML-simulation model significantly enhances the accuracy of musculoskeletal dynamics estimation from smartphone videos during walking.
• This novel framework outperforms simulation-only and ML-only methods for predicting ground forces, joint moments, and joint contact forces.
• The developed methods offer a scalable solution for out-of-lab musculoskeletal assessments, advancing precision treatment for gait-related conditions.