Related Experiment Video
Updated: Jul 11, 2025
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
6.8K
Development and validation of a fully automated 2-dimensional imaging system generating body condition scores for
N Siachos1, M Lennox2, A Anagnostopoulos1
1Department of Livestock and One Health, Institute of Infection, Veterinary and Ecological Sciences, University of Liverpool, Leahurst Campus, CH64 7TE, United Kingdom.
Journal of Dairy Science
|November 17, 2023
Summary
A new machine learning system accurately estimates dairy cow body condition score (BCS) using 2D imaging. This automated tool shows performance comparable to human experts, aiding in herd health management.
Area of Science:
- Animal Science
- Machine Learning
- Veterinary Medicine
Background:
- Body condition score (BCS) is crucial for assessing dairy cow energy reserves and health.
- Current manual BCS assessment relies on experienced personnel, which can be subjective and time-consuming.
- Developing automated methods for BCS evaluation is essential for efficient herd management.
Purpose of the Study:
- To develop and evaluate a fully automated 2D imaging system with a machine learning algorithm for real-time BCS prediction in dairy cows.
- To compare the accuracy and reliability of the automated system against manual BCS assessments by experienced veterinarians.
- To assess the system's ability to track changes in BCS over time (ΔBCS) and its correlation with backfat thickness.
Main Methods:
- Utilized ordinal regression and deep learning on a large dataset (34,150 manual BCS) for algorithm development.
- Validated the system on a separate dataset (9,657 BCS) using experienced human assessors as ground truth.
- Employed statistical analyses including weighted kappa (κw), percentage agreement (PA), Bland-Altman plots, and Passing-Bablok regressions.
Main Results:
- The automated system demonstrated substantial agreement with manual BCS (κw = 0.69), with high percentage agreement (94.8% within ±0.50 units).
- Repeatability of the system was almost perfect (κw = 0.99).
- The system showed strong correlation with backfat thickness (ρ = 0.75), comparable to manual scoring (ρ = 0.91).
Conclusions:
- The developed 2D imaging and machine learning system provides accurate and reliable automated BCS predictions for dairy cows.
- The system's performance is comparable to that of trained human scorers, offering a valuable tool for real-time herd health monitoring.
- This technology has the potential to enhance dairy farm management by enabling objective and efficient assessment of cow energy status.
