Improving gait classification in horses by using inertial measurement unit (IMU) generated data and machine learning.

For centuries humans have been fascinated by the natural beauty of horses in motion and their different gaits. Gait classification (GC) is commonly performed through visual assessment and reliable, automated methods for real-time objective GC in horses are warranted. In this study, we used a full body network of wireless, high sampling-rate sensors combined with machine learning to fully automatically classify gait. Using data from 120 horses of four different domestic breeds, equipped with seven motion sensors, we included 7576 strides from eight different gaits. GC was trained using several machine-learning approaches, both from feature-extracted data and from raw sensor data. Our best GC model achieved 97% accuracy. Our technique facilitated accurate, GC that enables in-depth biomechanical studies and allows for highly accurate phenotyping of gait for genetic research and breeding. Our approach lends itself for potential use in other quadrupedal species without the need for developing gait/animal specific algorithms.

Lateral movement of the saddle relative to the equine spine in rising and sitting trot on a treadmill.

Saddle slip, defined as a progressive lateral displacement of the saddle during ridden exercise, has recently been given attention in the scientific press as a potential sign of lameness. The aim of this study was to objectively quantify the normal lateral movement (oscillations) of the saddle relative to the horse in non-lame horses, and associate this movement to the movements of the horse and rider. Data from seven Warmblood dressage horses competing at Grand Prix (n = 6) or FEI Intermediate (n = 1) level, ridden by their usual riders, were used. Simultaneous kinetic, kinematic and saddle pressure measurements were conducted during sitting and rising trot on a force-measuring treadmill. The maximum lateral movement of the caudal part of the saddle relative to the horse's spine (MAX) was determined for each diagonal step. A mixed model was applied, with MAX as outcome, and T6 and S3 vertical position, rigid body rotation angles (roll, pitch, yaw) of the horse's and rider's pelvis, vertical ground reaction forces, saddle force, and rider position (rising in rising trot, sitting in rising trot or sitting in sitting trot) as explanatory variables. The least square means for MAX were 14.3 (SE 4.7) mm and 23.9 (SE 4.7) mm for rising and sitting in rising trot, and 20.3 (SE 4.7) mm for sitting trot. A 10 mm increase in maximum pelvic height at push off increased MAX by 1.4 mm (p<0.0001). One degree increase in rider pelvis roll decreased MAX 1.1 mm, and one degree increase in rider pelvis yaw increased MAX 0.7 mm (both p<0.0001). The linear relationships found between MAX and movements of both horse and rider implies that both horse and rider movement asymmetries are reflected in the lateral movements or oscillations of the saddle in non-lame horses.

Vertical movement symmetry of the withers in horses with induced forelimb and hindlimb lameness at trot.

BACKGROUND: The main criteria for lameness assessment in horses are head movement for forelimb lameness and pelvic movement for hindlimb lameness. However, compensatory head nod in horses with primary hindlimb lameness is a well-known phenomenon. This compensatory head nod movement can be easily misinterpreted as a sign of primary ipsilateral forelimb lameness. Therefore, discriminating compensatory asymmetries from primary directly pain-related movement asymmetries is a prerequisite for successful lameness assessment. OBJECTIVES: To investigate the association between head, withers and pelvis movement asymmetry in horses with induced forelimb and hindlimb lameness. STUDY DESIGN: Experimental study. METHODS: In 10 clinically sound Warmblood riding horses, forelimb and hindlimb lameness were induced using a sole pressure model. The horses were then trotted on a treadmill. Three-dimensional optical motion capture was used to collect kinematic data from reflective markers attached to the poll, withers and tubera sacrale. The magnitude and side (left or right) of the following symmetry parameters, vertical difference in minimum position, maximum position and range-up were calculated for head, withers, and pelvis. Mixed models were used to analyse data from induced forelimb and hindlimb lameness. RESULTS: For each mm increase in pelvic asymmetry in response to hindlimb lameness induction, withers movement asymmetry increased by 0.35-0.55 mm, but towards the contralateral side. In induced forelimb lameness, for each mm increase in head movement asymmetry, withers movement asymmetry increased by 0.05-0.10 mm, in agreement with the head movement asymmetry direction, both indicating lameness in the induced forelimb. MAIN LIMITATIONS: Results must be confirmed in clinically lame horses trotting overground. CONCLUSIONS: The vertical asymmetry pattern of the withers discriminated a head nod associated with true forelimb lameness from the compensatory head movement asymmetry caused by primary hindlimb lameness. Measuring movement symmetry of the withers may, thus, aid in determining primary lameness location.

Head, withers and pelvic movement asymmetry and their relative timing in trot in racing Thoroughbreds in training.

BACKGROUND: Horses show compensatory head movement in hindlimb lameness and compensatory pelvis movement in forelimb lameness but little is known about the relationship of withers movement symmetry with head and pelvic asymmetry in horses with naturally occurring gait asymmetries. OBJECTIVES: To document head, withers and pelvic movement asymmetry and timing differences in horses with naturally occurring gait asymmetries. STUDY DESIGN: Retrospective analysis of gait data. METHODS: Head, withers and pelvic movement asymmetry and timing of displacement minima and maxima were quantified from inertial sensors in 163 Thoroughbreds during trot-ups on hard ground. Horses were divided into 4 subgroups using the direction of head and withers movement asymmetry. Scatter plots of head vs. pelvic movement asymmetry illustrated how the head-withers relationship distinguishes between contralateral and ipsilateral head-pelvic movement asymmetry. Independent t test or Mann-Whitney U test (P<0.05) compared pelvic movement asymmetry and timing differences between groups. RESULTS: The relationship between head and withers asymmetry (i.e. same sided or opposite sided asymmetry) predicts the relationship between head and pelvic asymmetry in 69-77% of horses. Pelvic movement symmetry was significantly different between horses with same sign vs. opposite sign of head-withers asymmetry (P<0.0001). Timing of the maximum head height reached after contralateral ('sound') stance was delayed compared to withers (P = 0.02) and pelvis (P = 0.04) in horses with contralateral head-withers asymmetry. MAIN LIMITATIONS: The clinical lameness status of the horses was not investigated. CONCLUSION: In the Thoroughbreds with natural gait asymmetries investigated here, the direction of head vs. withers movement asymmetry identifies the majority of horses with ipsilateral and contralateral head and pelvic movement asymmetries. Withers movement should be further investigated for differentiating between forelimb and hindlimb lame horses. Horses with opposite sided head and withers asymmetry significantly delay the upward movement of the head after 'sound' forelimb stance.