Lameness remains one of the most costly and persistent health issues in the dairy sector. It reduces milk production, increases treatment costs, and causes long periods of discomfort for cows. Early detection is critical, yet current assessment methods often give inconsistent results.
New research from the UBC Animal Welfare Program has tested a different way to evaluate lameness—one that may help farms catch problems earlier while improving the data used to develop future automated detection tools.
Why Current Scoring Falls Short
Traditional gait scoring asks trained assessors to assign each cow a numerical score based on how she walks. This system demands time, farm visits, and consistent judgment. In practice, results often vary:
-
Trained assessors frequently disagree with each other.
-
The same assessor may give different scores on different days.
-
Subtle signs often go unnoticed until the cow is clearly lame.
These inconsistencies also limit progress in developing automated detection systems, since poor-quality training data leads to weak computer-vision models.
A Simpler Question: Which Cow Looks More Lame?
UBC researchers tested a new approach called a pairwise lameness assessment. Instead of scoring one cow at a time, assessors watch two cows walking side by side and choose which one looks more lame.
People tend to make comparisons more accurately than they score things in isolation. The team collected:
-
Videos of 30 cows
-
435 pairwise comparisons
-
Assessments from expert evaluators
-
Assessments from more than a dozen “crowd workers” with no specialized training
Using these comparisons, the researchers applied the Elo-rating system—a ranking method commonly used in chess and online games—to order all cows from least to most lame.
What the Study Found
The pairwise approach produced more consistent, detailed rankings:
-
Agreement among expert assessors was much higher than with traditional gait scoring.
-
The method detected a wider range of subtle lameness differences.
-
Crowd workers, even without training, produced rankings that closely matched those of experts.
Traditional scoring identified only 3 of the 30 cows as clinically lame. The ranking system revealed many more early or mild cases that conventional scoring tended to miss.
Why This Matters for Dairy Farms
If scaled up, this method could:
-
Lower the cost of lameness assessments
-
Improve consistency across assessors
-
Help identify early cases before they worsen
-
Produce higher-quality data for training AI-based monitoring tools
-
Speed up processing by using online workers to review large numbers of videos
Better early detection can support improved welfare, shorter recovery times, and fewer chronic cases.
Next Steps in Development
The research team is now working to automate parts of the process. Their next goal is to collect and sort video footage from multiple farms more efficiently, which will help build a large, reliable dataset. This type of dataset is essential for developing computer-vision tools that can eventually monitor cows in real time.
While the method still needs to be tested on larger groups and across different farm conditions, the results show strong potential for a more reliable and scalable way to assess lameness.
