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

Overview

• This study examines two statistical methods to estimate the forces exerted by the forelimbs of foals (young horses) on the ground during walking and trotting.
• It aims to improve understanding of equine biomechanics by providing accessible ways to measure ground reaction forces (GRFs) without complex direct measurements.

Background and Motivation

• Equine models are valuable in biomechanics research because horses share musculoskeletal similarities with humans.
• Foals’ rapid skeletal growth allows for studying developmental changes in biomechanics (ontogeny).
• Measuring ground reaction forces during locomotion is challenging in large animal models, limiting the availability of biomechanical data such as joint torques.
• Direct force measurements often need expensive force plates or instrumented treadmills, which may not always be accessible.

Research Objectives

• To evaluate and compare two statistical approaches – linear regression and machine learning models – in estimating the forelimb GRF of foals during walking and trotting.
• To assess these models’ ability to estimate key force parameters including peak vertical, braking, and propulsion forces, as well as continuous force profiles.
• To normalize gait speed differences between individual foals by employing the dimensionless Froude number (Fr), allowing comparison across ages and sizes.

Methodology

• Subjects: Three foals were studied longitudinally across different ages.
• Data Collection:
  • Motion capture system recorded limb movements during walking and trotting.
  • Ground reaction force data collected for direct comparison.
  • Subject mass was measured to calculate the Froude number (Fr = v^2 / (g*l), where v is velocity, g is gravity, l is leg length), providing a body-size normalized measure of gait speed.
• Gait Transition Analysis:
  • The walk-to-trot transition was consistently observed within a Froude number range of ~0.37 to 0.69.
  • Speed at transition varied from 1.75 m/s to 2.15 m/s but was stable across different ages of foals.
• Modeling:
  • Linear regression model: a straightforward statistical approach estimating GRF variables based on input parameters.
  • Machine learning model: potentially more flexible and accurate, particularly with larger datasets, but computationally more complex.

Key Findings

• Both statistical models produced comparable accuracy when estimating peak forces and continuous GRF profiles.
• Linear regression provided a simpler, less resource-intensive option suitable for small datasets.
• Machine learning showed promise for better performance as dataset size increases, suggesting scalability for future research.
• The models can potentially enable estimation of GRF data in naturalistic field settings where direct force measurement equipment is unavailable.

Significance and Applications

• Estimating GRFs accurately is critical for understanding limb loading, joint stresses, and overall locomotor biomechanics in foals and adult horses.
• The ability to reliably estimate these forces from simpler data (motion capture and mass) will support more widespread biomechanical studies without reliance on specialized force platforms.
• Forelimb force profiles estimated by these methods could facilitate early detection of gait abnormalities, inform rehabilitation strategies, and optimize training in equine veterinary medicine.
• The methodology and code have been made openly accessible on GitHub (github.com/TBL-UIUC/Equine-GRF-EstimationTools), encouraging community use and further refinement.

Limitations and Future Directions

• The sample size was small (n=3 foals), which may limit the generalizability and robustness of the models.
• Further data collection with larger and more diverse populations would enhance model accuracy and allow for tuning of machine learning approaches.
• Incorporation of additional gait types, other limb forces, or different species could expand the utility of these estimation techniques.
• Integration with real-time motion capture and wearable sensor technology could make GRF estimation more accessible outside laboratory environments.