Convolutional neural network for early detection of lameness and irregularity in horses using an IMU sensor
Authors: Benoit Savoini, J. Bertolaccini, Stéphan e Montavon, Michel Deriaz
Journal: 2025 International Conference on Advanced Machine Learning and Data Science (AMLDS)
Summary
# Editorial Summary Lameness detection remains challenging in equine practice because traditional visual assessment is subjective and prone to missing early-stage irregularities, whilst existing AI-based systems often demand expensive multi-sensor setups or laboratory conditions that limit real-world application. Savoini and colleagues developed a stride-level classification system using a single inertial measurement unit (IMU) paired with a one-dimensional convolutional neural network, trained specifically to distinguish sound from lame horses at the trot. Testing under practical field conditions yielded 90% session-level accuracy with zero false positives—a critical safeguard for clinical confidence. The elegance of this approach lies in its simplicity: one affordable, non-intrusive sensor replaces cumbersome force plates or video systems, making objective lameness screening genuinely deployable at yard or competition level. For practitioners, this technology offers a pathway to earlier intervention before subclinical gait deviations progress to overt lameness, supporting both welfare outcomes and athletic longevity whilst removing the guesswork inherent in visual diagnosis.
Read the full abstract on the publisher's site
Practical Takeaways
- •This IMU-based system offers a practical, affordable field tool for objective lameness detection that could reduce reliance on subjective visual assessment
- •Early detection capability may help catch subtle gait changes before they develop into clinical lameness, supporting preventive management
- •The single-sensor design makes this technology potentially deployable at yard level without specialized equipment or expertise
Key Findings
- •A single IMU sensor with 1D CNN achieved 90% session-level accuracy for detecting lameness in trotting horses with no false positives
- •The system successfully differentiated between sound and lame horses under real-world field conditions
- •Single-sensor approach significantly reduces hardware complexity and cost compared to multi-sensor or force plate systems