Linear Discriminant Analysis for Investigating Differences in Upper Body Movement Symmetry in Horses before/after Diagnostic Analgesia in Relation to Expert Judgement.

Abstract

Simple SummaryIdentifying the anatomical structures involved in causing pain and an associated lameness in a horse typically requires assessment in straight lines and circles and using regional administration of local anesthetic drugs (diagnostic analgesia). Visual assessment of changes in movement are affected by bias, i.e., expected changes influence decisions. Quantitative measurements with inertial sensors aim at removing this bias. The current study is aimed at investigating how a specific data-driven method, linear discriminant analysis (LDA), may be useful for aiding veterinary decision making about perceived changes in lameness. Changes in movement data after diagnostic analgesia and expert judgements from 53 lame horses were used to study (a) the accuracy of LDA-based decision making, (b) differences between straight-line and circular movement and (c) which commonly used movement features are most useful in this context. Accuracy was comparatively low and varied considerably between 36% and 57%, indicating considerable overlap between movement symmetry data of the diagnostic analgesia categories. The best data-driven separation between categories was observed when the limb in which perineural anaesthesia had been performed was on the inside of the circle (on hard ground for forelimb and on soft ground for hindlimb diagnostic analgesia). Movement features of all three landmarks (head, withers, pelvis) were important for data-driven classification, emphasizing the complexity of the movement pattern changes after diagnostic analgesia observed in lame horses.Diagnostic analgesia and lunging are parts of the equine lameness examination, aiding veterinarians in localizing the anatomical region(s) causing pain-related movement deficits. Expectation bias of visual assessment and complex movement asymmetry changes in lame horses on the lunge highlight the need to investigate data-driven approaches for optimally integrating quantitative gait data into veterinary decision-making to remove bias. A retrospective analysis was conducted with inertial sensor movement symmetry data before/after diagnostic analgesia relative to subjective judgement of efficacy of diagnostic analgesia in 53 horses. Horses were trotted on the straight and on the lunge. Linear discriminant analysis (LDA) applied to ten movement asymmetry features quantified the accuracy of classifying negative, partial and complete responses to diagnostic analgesia and investigated the influence of movement direction and surface type on the quality of the data-driven separation between diagnostic analgesia categories. The contribution of movement asymmetry features to decision-making was also studied. Leave-one-out classification accuracy varied considerably (38.3–57.4% for forelimb and 36.1–56.1% for hindlimb diagnostic analgesia). The highest inter-category distances (best separation) were found with the blocked limb on the inside of the circle, on hard ground for forelimb diagnostic analgesia and on soft ground for hindlimb diagnostic analgesia. These exercises deserve special attention when consulting quantitative gait data in lame horses. Head and pelvic upward movement and withers minimum differences were the features with the highest weighting within the first canonical LDA function across exercises and forelimb and hindlimb diagnostic analgesia. This highlights that movement changes after diagnostic analgesia affect the whole upper body. Classification accuracies based on quantitative movement asymmetry changes indicate considerable overlap between subjective diagnostic analgesia categories.