Endoscopic Visual and near Infrared Diffuse Reflectance Spectral Analysis for the Recognition of Healthy Porcine Gastric and Small Intestinal Wall—A Feasibility Study

Abstract

Recent investigations have proved the usefulness of spectroscopy in endoscopic examinations of the gastrointestinal tract in human medicine. However, current commercially available spectroscopy methods are relatively expensive. Near infrared (NIR) diffuse reflectance spectroscopy is a more cost-effective method. To assess its applicability for gastrointestinal endoscopy in veterinary medicine, in a pilot study we tested its feasibility for the differentiation of parts of the healthy gastrointestinal wall of pigs. Both white light standard endoscopy and NIR endoscopy were performed on the corpus, antrum and duodenum of seven clinically healthy pigs. General and gastrointestinal health was assessed by history, clinical examination and post-mortem examination. The spectral values of NIR endoscopy were obtained at intervals of about 0.3 nm in the range of 195–1118 nm. The method for the analysis of endoscopically taken NIR spectra consisted of pre-processing, feature extraction and classification. After smoothing, down-sampling and feature computing, spectral data were analysed using the principal component analysis (PCA) and partial least squares (PLS) techniques. The best algorithm performance was achieved using the spectra-PCA-SVC, giving a classification rate of 0.76. Further research is needed to assess the applicability of NIR endoscopy in other species and in animals showing gastrointestinal pathology. To improve the classification rate, probe calibration must be optimised and cleanness of the gastrointestinal tract ensured.