A Machine Learning Model to Predict Spinal Compression Force During Manual Lifting: Comparison of the Model Performance With Marker-Based Versus Markerless Kinematic Inputs

Musculoskeletal (MSK) modeling is widely used to estimate spinal loading. However, conventional modeling pipelines are computationally intensive and time-consuming, which limit their application for workplace injury risk assessment. This study developed a machine learning (ML) model to predict lumbosacral compression force using intermediate outputs from an MSK modeling workflow. Twenty-one participants completed lifting tasks while marker-based and markerless kinematics were recorded. Reference compression forces were obtained from completed OpenSim simulations, and the resulting dataset was used to train the ML model using three input feature sets. The best-performing model achieved mean root mean square errors of 3.0% with marker-based inputs and 9.9% with markerless inputs. These findings suggest that combining ML with conventional MSK modeling can enable rapid spinal load prediction and may support field deployment when coupled with markerless motion capture. However, given the higher errors observed with markerless input, future work is needed to improve the kinematics accuracy of the markerless system.