Veterinary Science Collected Works - Theses
Permanent URI for this collection
Search Results
1 - 1 of 1
-
ItemThe applications of sensor-based monitoring systems for diagnosing disease in peripartum dairy cattle managed in pasture-based dairy systems in south-west Victoria( 2019)The objectives of this thesis were: (1) to quantify the association between rumination time and physical activity in pastured dairy cattle, as measured by the SCR HR-LD collar (SCR Engineers, Netanaya, Israel) and the presence of peripartum health disorders; and (2) to assess the predictive potential of measurements generated by these devices in the predictive diagnosis of postpartum disease. An observational cohort study of 148 primiparous and multiparous Holstein dairy cows fitted with the SCR HR-LD device was undertaken at a commercial dairy farm in south western Victoria. Sensor derived, 2-hourly logs records of rumination and physical activity were collected from 10 days before parturition to 14 days in milk. The results of a physical examination performed by a veterinarian on each animal at 6±3 days in milk were also recorded. These data were used to construct multivariable linear, mixed-effects models to determine the effect of common peripartum health disorders on daily rumination time (DRT) and daily physical activity (ACT). Postpartum DRT was lower in animals affected by LDA and RFM and postpartum ACT lower in animals with subclinical ketosis or left displaced abomasum. Heifers had lower levels of postpartum DRT and higher levels of ACT. To determine the predictive potential of sensor measurements, bihourly rumination and activity data from the 48 hours prior to clinical examination were used to construct classification models for LDA, subclinical ketosis and undifferentiated metritis using three algorithmic classification techniques: extreme gradient boosting, neural network analysis and the random ferns technique. Fixed-effects logistic regression models were used to describe the risk relationships between sensor measurements and health disorders for the 48 hours prior to diagnosis. Declines in rumination and activity were associated with an increased odds of being diagnosed with any health disorder. Classification models had poor sensitivity and specificity for identifying all conditions. Optimal combinations of sensitivity and specificity were for the identification of measurements associated with LDA. The utility of algorithmic classifications of health status was hampered by poor combinations of sensitivity and specificity. Cow level sensing technologies may have a role in monitoring peripartum disease in pasture-based dairies but should be viewed as a screening device to indicate which cows require further assessment before interventions are taken.