Artificial Intelligence for Lameness Detection in Horses-A Preliminary Study.
Authors: Feuser Ann-Kristin, Gesell-May Stefan, Müller Tobias, May Anna
Journal: Animals : an open access journal from MDPI
Summary
# Editorial Summary Lameness remains one of the most significant welfare and performance concerns in equine practice, yet current detection methods rely heavily on subjective clinical observation by owners and veterinarians, limiting consistency and early identification. Researchers developed an artificial intelligence-based lameness detection system using pose estimation technology, tracking 58 anatomical landmarks across the head, limbs and pelvis to automate gait analysis without requiring invasive instrumentation or specialist equipment. The system successfully identified forelimb lameness through characteristic movement patterns of the head and forelimbs, whilst hindlimb lameness detection proved most reliable using stifle position data—though tuber coxae tracking did not yield clinically useful information. These preliminary findings suggest pose-estimation technology could offer a practical, objective screening tool for lameness detection across diverse settings, from performance yards to clinical environments, potentially enabling earlier intervention and more standardised assessment protocols. Wider application will require training the algorithm on substantially larger datasets and further validation across different horse populations, gaits and environmental conditions before integration into routine practice.
Read the full abstract on PubMed
Practical Takeaways
- •Pose estimation technology offers a non-invasive alternative to subjective lameness assessment, potentially reducing observer bias in gait evaluation
- •Current method is more reliable for detecting forelimb lameness; hindlimb lameness detection still needs refinement before clinical application
- •Simple setup with easily detectable anatomical landmarks suggests future potential for widespread practical use, though this is preliminary work requiring validation on larger populations
Key Findings
- •Pose estimation using 58 anatomical reference points enables non-invasive gait analysis for lameness detection
- •Forelimb lameness is detectable via trajectories of head and forelimb reference points
- •Stifle joint showed promising results for hindlimb lameness detection, while tuber coxae was unsuitable
- •System demonstrates feasibility but requires larger dataset for further development