Project Leader: Stephanie Buijs
Project code: D-101-19
This project will use an automated thermography system that was recently installed in AFBI’s milking parlour. The +/- 240 dairy cows that are milked in this parlour will pass the system on their way back to the pasture or barn after milking (thus, twice daily). This will be combined with highly regular and systematic observation of clinical signs of disease and injury by trained staff, and with behavioural assays of stress/fear/anxiety. Cross-comparison of thermal data and clinical examination will be used to identify thermal markers for different conditions (phase 1) and to test the reliability of these markers as early warning signals (phase 2). In the last phase we will evaluate if adding information from other sensor technology improves the sensitivity and specificity of the system (phase 3).
Our overall objective is to evaluate the capacity of automated thermography to improve the detection of reduced dairy cow welfare, by identification of diseased, injured and stressed individuals at the earliest possible stage. Early detection is crucial, as it allows intervention before problems progress to a stage of (extensive) antibiotic treatment, severely reduced welfare and/or major production losses. Good stockmanship and veterinary care are indispensable for early detection. The objective is not to replace these, but to allow these people to focus their efforts on those individuals that are more likely to be affected.
Thermography creates images of surface temperature. This allows detection of many clinical signs of disease and injury (fever, inflammation, bruising), as these affect skin temperature, increase heat radiating from the skin (bald patches) or cool the skin with moisture (discharge from the nose, eye or vulva). Stress leading to an increased eye temperature can also be detected.
Phase 1 aims to determine thermal markers associated with specific clinical signs of disease, injury and stress. To this end, thermal images of cows with and without a specific clinical sign will be compared. These animals will be identified by regular health checks and behavioural tests. Thermal markers may also be identified by comparing thermographs of different areas of affected animals (e.g., left vs. right leg) and from animals before, during and after the occurrence of the clinical sign. Once markers are identified, algorithms allowing their automatic detection will be created.
Phase 2 aims to assess how accurately the thermal markers detect the clinical signs. We will analyse which percentage of animals with a clinical sign are classified by the system as affected and which percentage of the animals without the clinical sign are classified as not affected. This last is important because many ‘false-alarms’ will decrease the likelihood that will act upon alerts if the system is implemented in practice. Furthermore, we will determine if the system can detect clinical signs ahead of routine care (i.e., outside of the regular health checks and tests). This will provide important information on the expected improvement in speed of detection in practice.
The objective of Phase 3 is to see if additional information on behaviour (feeding, ruminating, standing, lying, walking) can improve the accuracy of the system. Decreasing the threshold for a thermal marker if behaviour has changed, or blocking alerts unless behaviour has changed, may improve system accuracy.