Kinematics to Kinetics: Evaluating OpenGRF for Scalable Ground Reaction Force Estimation in Healthy and Pathological Gait

Abstract

Ground reaction forces (GRFs) are important markers of gait impairment and rehabilitation, but direct measurement with force plates is expensive and limited to laboratory settings. OpenGRF is an open-source framework that estimates GRFs from motion-capture data using musculoskeletal modeling, yet its validity in pathological gait remains uncertain. This study evaluated the accuracy of OpenGRF for estimating three-dimensional GRFs during gait in healthy adults and in people with Parkinsons disease (PD), stroke, hip osteoarthritis (HOA), and total hip arthroplasty, and assessed its ability to reproduce clinically relevant GRF peak magnitudes and timings. OpenGRF estimates were compared with force-plate measurements in healthy participants and in cohorts with PD (OFF/ON medication), stroke, and HOA before surgery (M0) and 6 months after surgery (M6). Accuracy was quantified using root mean square error (RMSE), normalized RMSE (NRMSE), and Pearson correlation coefficients (PCC). Peak analyses examined biases in anterior-posterior (AP) braking and propulsive peaks, first and second vertical peaks (V1, V2), and the main mediolateral peak, as well as timing shifts across the gait cycle. One-dimensional statistical parametric mapping tested waveform differences (p = 0.01). OpenGRF reproduced overall GRF profiles across groups. Vertical GRF showed the best agreement (PCC 0.84-0.94; NRMSE 0.14-0.23), whereas mediolateral GRF showed low absolute error (RMSE 1.45-1.95 %BW) and moderate-to-good agreement (PCC 0.70-0.82). AP GRF was least accurate (PCC 0.61-0.73), especially during propulsion in PD (NRMSE up to 0.39). OpenGRF can reasonably estimate GRFs in healthy and pathological gait and may support kinetic gait assessment when force plates are unavailable.