Lameness in dairy cattle is a prevalent issue that significantly impacts both animal welfare and farm productivity. Traditional lameness detection methods often rely on subjective visual assessment, focusing on changes in locomotion and back curvature. However, these methods can lack consistency and accuracy, particularly for early-stage detection. Typically, lameness is classified using the Locomotion Scoring System (LCS), which grades severity based on observable changes in movement and posture. This study presents an objective analysis of cow back shape in a sample of 260 Holstein-Friesian cows to identify specific regions associated with varying levels of lameness. A keypoint detection algorithm was employed to map 12 keypoints along the cow's back, which was divided into three regions: cranial, middle, and caudal. Curvature analysis was performed by angles at each keypoint, enabling the extraction of relevant kinematic features, as back posture can be reliably captured with a single side-view camera and may reflect early signs of lameness. These features were subsequently input into a deep learning model to classify cows based on their locomotion scores. The model achieved a high classification accuracy of 97 % in distinguishing lame from non-lame cows. While the cranial region contributed minimally to lameness detection (η2 = 0.02), the middle (η2 = 0.14) and caudal (η2 = 0.068) regions were critical, especially for identifying more severe cases. These findings suggest that analyzing back shape characteristics, particularly in the middle and caudal regions, provides valuable indicators for detecting lameness severity and may enhance the accuracy of automated lameness assessment in dairy cattle.
Narlı et al. (Sun,) studied this question.