Portable near-infrared spectroscopy and linear support vector machine classifiers for rapid screening of subclinical inflammation in dairy goat milk

Abstract

We evaluated the applicability of portable near-infrared spectroscopy (NIRS) to the rapid screening of goat milk for alterations related to mammary inflammation using the mastitis marker cathelicidin (Cath) as an inflammation-related reference outcome and SCC as the conventional routine reference. Diffuse-reflectance spectra (900-1,700 nm) were acquired directly through milk tubes (n = 228), preprocessed by standard normal variate (SNV), and reduced to 5 chemically interpretable spectral windows. An unsupervised principal component analysis was computed on the full SNV preprocessed spectral dataset, and samples were assigned to training (n = 171) and independent validation (n = 57) sets by principal component 1-quantile stratified randomization. Linear-kernel support vector machine (SVM) classifiers were trained to predict Cath status (positive defined as the presence of a reactive band at 15 kDa in the lane corresponding to the sample) and SCC class (positive defined as SCC >1,000,000 cells/mL). On validation, NIRS demonstrated associations with both reference outcomes. The Cath status model and the SCC class model yielded sensitivity of 0.69 versus 0.52, specificity of 0.90 versus 0.72, and area under the receiver operating characteristics curve of 0.80 versus 0.62, respectively. This was in line with the biological reliability of Cath as an inflammation marker, although paired bootstrap comparisons did not show statistically significant differences on the restricted validation set. Overall, portable NIRS combined with SVM showed preliminary promise as a rapid, nondestructive field triage approach for screening milk alterations associated with mammary inflammation in dairy goats.

Abstract as supplied by the indexing source. © the authors / publisher.