PLS-DA and Vis-NIR spectroscopy based discrimination of abdominal tissues of female rabbits

Abstract

Using Vis-NIR spectroscopy to distinguish gestational sac from other abdominal tissues is the key to diagnosing female rabbits' pregnancy by optical means. This study aims to demonstrate the gestational sac and other abdominal tissues (hair, skin, breast, muscle, cecum, small intestine) of rabbits can be identified using Vis-NIR spectroscopy in vitro. These tissues' raw NIR spectra were recorded in the Vis-NIR range (490-940 nm) with interactive mode. The raw spectra of tissues were analyzed by the principal component analysis (PCA), and were pre-processed using five spectral pre-processing techniques (moving average filter (MF), De-trending (DT), first-order derivative (D1), Multivariate scattering correction (MSC), and standard normal variate (SNV)) to reduce signal noises. The raw and pre-processed spectra were classified using partial least squares discrimination analysis (PLS-DA). Two-way and multi-way PLS-DA model was conducted to understand the classification of each tissue from the gestational sac and to understand the classification of all tissues from the gestational sac, respectively. SNV-PLS-DA model had the best performance, and its multi-way accuracy (Ac), determination coefficients (R²), and Q² were 0.89, 0.91, 0.77, respectively. The successive projection algorithm (SPA) and competitive adaptive reweighted sampling (CARS) were used to select characteristic wavelengths (CWs). The SNV-SPA-PLS-DA model with eighteen CWs was better than the SNV-CARS-PLS-DA model. The results showed that Vis-NIR spectroscopy technology combined with PLS-DA could discriminate the gestational sac from the abdominal tissues. This study may help develop an optical diagnosis system for pregnant rabbits.

Keywords: Female rabbits; PLS-DA; Pregnancy diagnosis; Tissues discrimination; Vis-NIR spectroscopy.