Journal of Animal Sciences and Livestock Production 开放获取

抽象的

Development of Sow Lameness Classification Trees Using an Embedded Microcomputer-based Force Plate in a Commercial Setting

KJ Stalder, BM McNeil, JA Calderon Diaz, JD Stock, TD Parsons, DL Beam, AK Johnson, CE Bruns and JB Niemi

Background and Objectives: The objectives of this study were: 1) to examine the relationship between forces applied by each leg as measured by the force plate and the degree of visually assessed lameness under conditions applicable to a commercial herd, and 2) to develop an automated lameness detection algorithm based on the force plate output.

Methods and Findings: The microcomputer-based embedded force plate system provides an objective approach to lameness detection by measuring the force generated by each individual limb. The force plate device was installed within an Electronic Sow Feeder (ESF) and used to monitor a subset of the 120 multiparous gestating sows housed in a dynamic group over a 21 day period. Each day sows entered the ESF station one at a time to eat. At times when the sow stood squarely and applied pressure to all quadrants of the device, the force applied by each foot was recorded once per second. Sows were visually scored for the presence of lameness using a four-point scale (0=normal to 3=severely lame) on a weekly basis and classified based on this visual assessment as non-lame (score ≤ 1) or lame (score ≥ 2). An ensemble learning method called Random Forest was used to identify the optimal decision tree for classifying the force plate data into similar categories of non-lame and lame. A Kappa Statistics test was used to measure the level of agreement between the visual scoring and force plate results. Changes in lameness status, as well as the first day of lameness identification for each detection method, were also analyzed. Seven variables were included in the classification tree with the most weight given to the difference between the forces applied to the 2 hind legs. The two lameness detection methods assigned the same lameness classification in 95% of cases and had substantial agreement (Kappa Statistic=0.79; P<0.05). However, the classification tree algorithm detected lameness almost 5 days earlier than the visual scoring system (P<0.001). Additionally, comparing lameness of sows from the time of entry into the group, showed an increase in lameness after the first week regardless of the lameness scoring method.

Conclusions: Lameness detection typically is based on subjective visual evaluation, which requires time, training, and can be biased between and within individuals. Results demonstrate that under conditions applicable to a commercial herd, the force plate can accurately detect lameness sooner than a weekly visual lameness assessment.

Keywords: Sow; Lameness detection; Force plate; Weight distribution