Comparison of Knee Contact Force Estimation Errors Between Two OpenSim Musculoskeletal Models During Walking and Squatting

Abstract

Objective Accurate estimation of musculoskeletal forces is essential for understanding joint biomechanics, improving rehabilitation strategies, and optimizing implant design. Joint reaction forces can be estimated using biomechanical measurements combined with musculoskeletal modeling approaches. This study aimed to compare the accuracy of knee joint contact force estimation between the Gait2392 and Rajagopal musculoskeletal models implemented in OpenSim. Methods Data from six participants (five males and one female; age: 68 ± 5 years; body mass: 88 ± 12 kg; height: 173 ± 4 cm) were analyzed. Biomechanical data collected during walking and squatting tasks were used in this study. After scaling both musculoskeletal models, joint reaction force analyses were performed in OpenSim. Relative errors between predicted knee contact forces and in vivo implant-measured forces were calculated. Statistical Parametric Mapping (SPM) paired t-tests were used to compare modeling errors between the two models at a significance level of 0.05. Results SPM analysis revealed significantly greater estimation errors in the Rajagopal model during the stance phase of walking and across a substantial portion of the squat cycle, particularly in the vertical force component. The RMSE values for the Gait2392 model were 0.835 during walking and 1.536 during squatting