Statistical approaches for estimating forelimb ground reaction forces in foals during walking and trotting.

Abstract

Equine models are useful in biomechanics research due to their similarity in musculoskeletal tissue to humans, their athletic nature, and rapid skeletal development which permits ontogenetic studies. However, a continuing challenge in musculoskeletal models for large animal biomechanics is measuring the ground reaction force (GRF) during locomotion and therefore few reports of biomechanical measures such as joint torques. Here we evaluate two statistical approaches for estimating forelimb ground reaction forces in foals (n = 3). Longitudinal motion capture, GRF, and subject mass data during walking and trotting gaits. To account for differences in subject size, we calculated the dimensionless Froude number Fr=v²gl. The walk-trot transition occurred within the Fr range from 0.37 to 0.69 (v = 1.75-2.15 m/s) and was consistent across ages. Linear regression and machine learning models were used to estimate peak and continuous vertical, braking, and propulsion forces. Both models resulted in comparable performances when estimating peaks and continuous GRF profiles, with the linear regression model offering a simple and computationally inexpensive option and the machine learning model showing potential for improved performance with larger datasets. The models are available at github.com/TBL-UIUC/Equine-GRF-EstimationTools. Although the models would benefit from a larger sample size, our results highlight the potential to estimate GRF profiles in real-world settings and, when coupled with motion capture data, facilitate future studies of equine biomechanics.