Human Milk Lactation Phases Evaluation Through Handheld Near-Infrared Spectroscopy and Multivariate Classification

Currently, many newborns are consuming human milk (HM) donated by nursing mothers, especially those in the neonatal intensive care unit. However, a frequent problem in HM banks is the lack of information about the lactation phase of donated milk (colostrum, transition, or maturity), which makes it more difficult to supply children according to their particular needs. This research aims to develop models based on handheld near-infrared spectroscopy (NIR) coupled with partial least squares with discriminant analysis (PLS-DA) to individually classify HM in its different phases (colostrum—class 1; transition—class 2; mature—class 3). A total of 198 samples were pasteurized before spectra collection. Subsequently, several preprocessing techniques were investigated in NIR spectra using principal component analysis: multiplicative scatter correction (MSC), derivatives, and combinations between them. The NIR spectra preprocessed through MSC were employed in PLS-DA models, where sensitivity and specificity obtained were considered satisfactory for classifying mainly colostrum (87.5% and 90.3%, respectively) and mature (93.8% and 93.8%, respectively) phases. VIP scores showed that for the modeling of the colostrum phase the important regions are related to protein, while for the mature phase the lipids are more important. This plot also confirmed that the transition phase shares properties of the colostrum and mature phases, which probably contributed to its poor classification (rates ranging from 56.3 to 71.9%). Therefore, handheld NIR coupled with PLS-DA is suitable for the classification of HM in different phases of lactation in a low-cost and quick analysis. The proposal can be used in regulated banks to classify samples with unknown lactation phase before supplying them to newborns and can help to provide the ideal HM phase for their needs.