The role of sensor measurements in treating mastitis on farms with an automatic milking system

Abstract

Mastitis detection is aimed at finding cows with clinical mastitis (CM) so they can be treated. Farmers with an automatic milking system (AMS) rely on mastitis alert lists, generated by the AMS, to detect cows with CM. These lists are based on sensor measurements collected during milking. Mastitis detection models with sufficiently high sensitivity and specificity will help to select the CM cases that need treatment. The current CM detection performance can be summarized with a sensitivity of 36.8% and a specificity of 97.9%. These values don’t meet the suggested requirements on CM detection performance (sensitivity >70%, specificity >99%). Recent research was able to improve detection performance by using better detection algorithms, adding non-sensor information, and using new or improved sensors. The achieved sensitivity and specificity, however, were still not perfect. This imperfection may be explained by the fact that the majority of the alerts are from cows with intramammary infections. Some of these cows will develop CM, but the majority will not become clinically infected. Therefore, reaching a very high specificity (e.g. 100%) for CM detection is very difficult, and it is expected that future CM detection models on an AMS will not be perfect. As a consequence, farmers with AMS have to accept that not all CM cases will be detected, that some cases will be detected late, and that there will be false-positive alerts. With this inevitable imperfect detection in mind, it is essential to start thinking about detection and treatment protocols for farmers with AMS. These protocols may include specific actions for new alerts and for alerts which are on the list for weeks. This paper can be seen as a starting point for a further debate on providing treatment protocols for farmers with AMS.