Estimation of dynamic spinal loads during manual lifting using smartphone-based markerless motion capture

Area of Science:

• Biomechanics
• Ergonomics
• Digital Health

Background:

• Accurate estimation of spinal loads is crucial for understanding and preventing work-related musculoskeletal disorders.
• Conventional methods for dynamic spinal load estimation are often time-consuming and require specialized equipment.

Purpose of the Study:

• To evaluate the feasibility of dynamic spinal load estimation using smartphone-based markerless motion capture.
• To compare markerless-driven musculoskeletal (MSK) modeling and a machine learning (ML) approach against a conventional marker-based MSK modeling workflow.

Main Methods:

• Simultaneously recorded marker-based and markerless kinematics during various lifting tasks.
• Developed a markerless-driven MSK modeling workflow and an ML model predicting spinal loads from MSK model outputs.
• Used OpenSim for optimization-based MSK modeling to obtain reference spinal load estimates.

Main Results:

• Both markerless-driven MSK modeling and the ML approach estimated compression forces with reasonable accuracy (nRMSE 12% and 9%, respectively).
• Shear force estimation showed larger errors, particularly with the markerless-driven MSK modeling approach.
• The ML-based approach demonstrated higher accuracy for compression force estimation.

Conclusions:

• Smartphone-based markerless kinematics can enable feasible spinal load estimation during manual lifting.
• A hybrid framework integrating data-driven algorithms with physics-based MSK modeling shows promise for efficient spinal load assessment.
• Future work should enhance markerless system accuracy and ML model generalizability, especially for shear force prediction.