Reliability, agreement and variability of a markerless computer vision algorithm for equine gait analysis under field...

Area of Science:

• Equine biomechanics
• Veterinary orthopedics
• Computer vision applications

Background:

• Computer vision algorithms offer accessible equine gait analysis.
• Thorough assessment under diverse conditions is crucial for these tools.
• Evaluating algorithm reliability is key for practical application.

Purpose of the Study:

• To assess a vision-based algorithm's reliability in measuring vertical displacement signals (VDS) at the eye, withers, and croup.
• To evaluate the algorithm's groundline estimation capabilities.
• To test the algorithm under field conditions with horses trotting on straight lines and circles.

Main Methods:

• A cross-sectional comparative study design was employed.
• 67 iPhone recordings from 37 horses were analyzed using a markerless computer vision algorithm.
• 2D keypoints were generated for groundline estimation, VDS, and stride-based vertical differences, compared against manual annotations using MSE, MAE, and Bland-Altman plots.

Main Results:

• Stride-level mean absolute errors for vertical displacements were low (overall 4.3 mm).
• The eye keypoint showed the lowest error (2.9 mm Maxdiff, 3.0 mm Mindiff), withers at 5.5 mm, and croup at 4.3-4.4 mm.
• Trial-level analysis indicated consistent performance across multiple strides, with lower absolute differences (Eye: 2.3 mm, Withers: 3.7 mm, Croup: 2.7 mm).

Conclusions:

• The computer vision algorithm demonstrated robust measurement of vertical displacements under varied field conditions.
• Further clinical validation against established gait analysis systems is recommended.
• The algorithm shows promise for accessible equine gait analysis.