Accuracy and reliability of mastitis detection with electrical conductivity and milk colour measurement in automatic milking

Abstract

Detection of mastitis and reliability of alarms were evaluated in eight herds with automatic milking. Herds were visited once to gather milk samples and the farmers followed up alarms for 15 to 30 days. Sensitivity of detecting quarters with SCC over the threshold value of 200,000 or 1,000,000 cells/ml was 18.0 to 42.9% in Group A, based on electrical conductivity (EC) and 5.5 to13.0% in Group B, based on EC and milk colour. Specificity was over 97.0%. In Group A, 6/17 of the EC alarms were false positives. None of the seven EC alarms in Group B was false positive. On the day that the farmer observed clinical mastitis, more than half of the quarters received alarms, but only milk from some quarters was automatically diverted. In total 58.4% (Group A) and 26.8% (Group B) of cows had alarms based on EC, and 84.1% (<20% true positives) and 12.0% (60–90% true positives) had alarms based on milk yield or milk colour during the follow-up period. The number of alarms was halved in Group A when a running average was applied to the algorithm of EC. The proportion of true positives simultaneously increased from 30% to 60%. In Group B, the number of alarms was almost doubled with algorithm manipulation, but the proportion of true positives declined from >80% to >70%.