PSXII-4 Using deep learning to evaluate the relationship between gait and dressage performance in sport horses.
Authors: A. Cooper, Samantha A Brooks, K. Allen, L. Dewberry, Cecelia Minner
Journal: Journal of Animal Science
Summary
# Editorial Summary Researchers used deep learning computer vision to quantify relationships between objective gait parameters and dressage competition scores in sport horses, analysing 194 trot videos collected during three-day eventing competitions across the southeastern United States between 2019 and 2022. A custom DeepLabCut model tracked 22 anatomical keypoints per horse, generating seven standardised gait metrics (stride length, duty factor, fetlock angle range, and limb swing and travel distances) which were then correlated against dressage scores achieved 3–4 days later. Stride length emerged as the strongest predictor of dressage performance, with a moderately strong positive correlation (p=0.00213 via generalised linear mixed modelling; p=0.000361 via the more sophisticated generalised additive model), whilst duty factor—the proportion of the stride cycle during which a limb remains in contact with the ground—approached significance and suggested a weaker but meaningful relationship to scoring. For farriers, veterinarians and coaching professionals, these findings offer an evidence-based framework for understanding how measurable movement characteristics drive competitive outcomes in dressage, potentially shifting subjective soundness assessments towards more objective criteria and creating opportunities to identify and breed for advantageous gait traits.
Read the full abstract on the publisher's site
Practical Takeaways
- •Stride length is an objective, measurable gait parameter that correlates with dressage performance—video gait analysis could provide more standardized soundness evaluations than subjective visual assessment
- •Deep learning-based gait analysis offers potential to reduce bias and increase repeatability in pre- and post-competition fitness evaluations ('jogs')
- •Quantifying gait quality through biomechanical parameters may eventually enable genetic selection for horses with advantageous movement characteristics for sport
Key Findings
- •Stride length showed a moderately strong positive correlation with dressage score (p=0.00213) using GLMM analysis
- •GAM analysis confirmed stride length as the strongest gait predictor of dressage performance (p=0.000361)
- •Duty Factor showed weaker but significant correlation with dressage score (p=0.06299)
- •Deep learning analysis of 22 anatomical keypoints from 194 trot videos identified objective gait parameters that predict dressage scoring outcomes