Real-Time near Infrared Spectral Monitoring of Mammary Gland for Inflammation Diagnosis in Dairy Cows

Abstract

The diagnosis of cow mastitis was performed using shortwave near infrared (NIR) spectroscopy of cow udder tissue collected at morning milking over a two-year period in order to investigate the influence of various factors such as season, udder position, cow identity etc. The NIR spectra were acquired using a portable NIR spectrometer and quarter foremilk was collected and analysed for milk components as reference data. The somatic cell count (SCC) was mainly used for mastitis diagnosis. The analysis of in vivo shortwave NIR spectral data of udder tissue was carried out using the soft independent modelling of class analogy (SIMCA) classification method. First, the influence of the difference in udder location on the spectral data were investigated. The SIMCA calibration revealed that spectra for different udder locations, such as front vs rear and right vs left udder quarter, enabled different prediction accuracy. SIMCA classification for mastitis diagnosis based on the SCC value was performed for the whole udder and for the front and rear quarter spectra, separately. In order to find the accurate threshold for mastitis using only spectral data, a single threshold value of SCC for recognition of mastitis was increased in steps of 10,000 cells mL−1 from 100,000 to 200,000 cells mL−1 and then SIMCA was applied for spectral data analysis. As a result, the best SIMCA prediction for the whole udder, 60.9% of sensitivity and 68.6% of specificity, was obtained for the SCC threshold of 110,000 cells mL−1. In contrast, the prediction result for front and rear quarter spectra showed different classification results. For the front udder quarter, a best prediction accuracy of 75.0% of sensitivity and 84.6% of specificity was obtained when using 160,000 cells mL−1 of SCC as the threshold, while the rear quarter spectra showed 100.0% of sensitivity and 84.7% of specificity with the same threshold for SCC as for front udder (160,000 cells mL−1). The highest classification accuracy of mammary gland inflammation diagnosis was obtained when front and rear udders were analysed separately.