Lameness remains one of the most persistent and costly health challenges in dairy production. It affects cow comfort, milk production, fertility, and longevity—yet early detection continues to be a major hurdle. Traditional scoring systems rely on trained assessors watching individual cows and assigning a gait score, but research continues to show significant variability between assessors, between scoring sessions, and even within the same evaluator over time.
A new study from the UBC Animal Welfare Program proposes a different approach—one that could dramatically improve the consistency of lameness assessments while supporting the next generation of automated detection tools.
The Limitations of Traditional Gait Scoring
Producers and veterinarians know early lameness identification is essential, but current manual assessment systems come with several challenges:
-
Inconsistent scoring: Even experienced assessors may disagree on the same cow or give different scores for the same cow on different days.
-
Labour intensive: Skilled assessors must visit farms to evaluate cows one at a time.
-
Delayed identification: Visible lameness often appears after the cow has already been uncomfortable for days, making recovery slower and potentially less successful.
These inconsistencies make it difficult to generate reliable datasets. Since automated detection systems require high-quality training data, any variability in manual scoring ultimately affects the accuracy of new technologies.
A New Concept: Ranking Cows Through Pairwise Comparison
Instead of assigning an absolute gait score to an individual cow, researchers at UBC tested a simple comparative question:
When you watch two cows walk side by side, which one appears more lame?
This method—known as pairwise lameness assessment—leverages the natural human ability to make relative comparisons more consistently than assigning a numerical score in isolation.
How the method works:
-
Videos show two cows walking at the same time.
-
Assessors choose which cow appears more lame, and by what degree.
-
Every cow is compared with every other cow in the set.
-
An Elo-rating system (similar to chess ranking) creates a hierarchy from the least to the most lame.
The study also tested whether untrained “crowd workers” from online platforms could participate. The result: these untrained evaluators produced rankings that closely matched the assessments of trained experts.
What the Research Found
Researchers evaluated 30 cows using both the new ranking method and traditional 5-level gait scoring.
Key findings:
-
Stronger agreement among assessors:
-
Pairwise method ICC: 0.81
-
Traditional scoring ICC: 0.44
-
-
More consistent repeated assessments:
-
Even the same expert showed variation using traditional scoring (ICC: 0.62).
-
-
Greater sensitivity:
-
Traditional scoring identified three clinically lame cows.
-
Pairwise rankings detected finer differences across the group.
-
-
High agreement from untrained evaluators:
-
Crowd-sourced rankings ICC: 0.85, closely matching experts.
-
Only a small number of crowd assessments were needed for reliable results.
-
These results show that comparing cows directly offers a more intuitive and consistent method, producing cleaner data than traditional scoring.
Implications for Dairy Farms
While this research is still in development, the approach has several potential benefits for producers and advisors:
1. Earlier Detection
The method’s finer ranking detail may help identify cows showing subtle mobility changes before obvious lameness develops.
2. More Consistent Data
Improved agreement among assessors means more objective information to support decisions around treatment, hoof care, housing adjustments, or nutrition strategies.
3. Faster, More Scalable Assessments
Using untrained online workers greatly reduces the cost of large-scale evaluations and allows large numbers of videos to be processed quickly.
4. Stronger Foundation for AI-Based Technologies
Reliable manual data is essential for training accurate machine-learning models.
This ranking system offers a pathway toward building high-quality datasets for future automated lameness detection tools.
Next Steps in the Research
The UBC team is now working on the next phase of development, including:
-
Automating video collection and cow identification across multiple farms
-
Expanding the video library to include more animals and diverse environments
-
Integrating pairwise rankings into machine-learning systems for practical, on-farm tools
As the database grows, researchers expect to refine automated systems that could help make routine lameness monitoring more feasible.
Key Takeaways
Lameness continues to be a major challenge for the dairy sector, but new approaches are emerging that may lead to earlier, more reliable detection. Pairwise video comparisons offer a practical and consistent way to rank cows by mobility, and crowd-sourced assessments have shown strong alignment with expert evaluations. This method could help support improved cow welfare, more informed management decisions, and more accurate AI-based detection tools in the future.
