This blog post was written by Anna Bradtmueller
Lameness is a prevalent issue within the dairy industry that has serious financial and welfare implications. Farmers and researchers have often relied on methods of visual locomotion scoring to identify lameness. However, these approaches are relatively subjective in comparison to the more detailed, objective measures of locomotion recorded through multiple novel technologies that have become available for automated gait assessment purposes. One such technology is kinematic gait analysis systems, which typically consist of a camera and motion tracking software. Kinematics is a subdivision of biomechanics involving aspects of motion that can be visually observed in space and time. This makes kinematics a good “starting point” for automating - and eventually minimizing the need for - traditionally used visual locomotion scoring of dairy cows.
As part of my master’s degree, I conducted a validation study using our new kinematic room. Our study aimed to use our updated kinematic system to see if a machine learning approach could be used to predict a locomotion score based on a commonly used visual scoring system from the collected kinematic data. The locomotion scoring system used was developed by Flower and Weary (2006) and consisted of a 1 – 5 scale with 0.5 intervals, where a score of 3 or higher represented a clinically lame cow. This study was conducted in collaboration with the Bioinformatics Lab at the Université du Québec à Montréal, who had worked with our lab on a previous study using similar kinematic data to develop a model that could predict lame vs. non-lame classifications of cows. In our current study, we were particularly interested in identifying cows who exhibited more subtle signs of locomotor impairment or gait abnormalities (locomotion scores of 2 or 2.5) but whose gait was not yet at the severity level of clinical lameness (locomotion scores 3+). Identifying cows at this level of locomotor impairment is of great interest to both farmers and researchers, as detecting more minor changes or abnormalities in gait allows for earlier intervention and is imperative to preventing more severe cases of lameness from developing.
The new kinematic room/system consisted of 6 overhead cameras mounted on the ceiling with additional lighting, a 7-meter walkway upon which the cows were filmed walking, a software used to simultaneously record video from all 6 camera angles, and a software used to track the motion of the cow to acquire 3D-coordinates of specific joints.
We conducted data collection in January-February of 2021 and completed digitization and processing of our recorded videos in March-April. A visual observer used recorded video to conduct locomotion scoring of individual “passages” (in which a cow completed walking the length of the designated walkway). The 3D trajectories of 20 markers attached to specific joints on the cow and overall locomotion scores assigned for individual “passages” were then used for training and testing of two types of artificial neural networks.
While overall locomotion scores could not be successfully predicted through this approach, the next steps will aim to make predictions of more specific types of gait attributes, such as tracking up of hind limbs or back arch of the cow while walking. These more specific aspects of gait may be less subjective than an “overall” locomotion score to visual observers who are conducting scoring, and therefore may be more easily reflected in kinematic data. The next steps will also test using machine learning approaches to make predictions based on kinematic data from fewer, specific joints on the cow which may show a greater difference between cows of varying locomotor ability levels.
The continuation of this research in our lab will aim to identify patterns in cow gait which could be linked to specific causes or reasons behind different types of gait abnormalities. For example, changes in certain aspects of gait such as stride length or joint range of motion could be associated with specific types of hoof pathologies. Additionally, an artificial neural network capable of predicting locomotor ability levels of cows - either in terms of “overall” gait or through more specific gait attributes - could be applied to study environmental or management factors which may affect locomotion, such as housing type, flooring surfaces, or access to exercise. This research will ultimately help to improve the detection of locomotion abnormalities at an earlier stage, identify changes that could help minimize the occurrence of pathologies or injuries contributing to locomotor impairment, and better address the overall health and welfare of cows. Lastly, I was happy to present my findings to the Animal Science Department at the beginning of December 2021 as part of my Results Seminar, which successfully marked the conclusion of my Master's.