Application of hyperspectral imaging and band selection for supervised classification of small intestinal tissues

Abstract

Acute mesenteric ischemia is a severely underestimated vascular and gastrointestinal disease today and is particularly frequent in neonates and the elderly. Few methods are currently used for noninvasive and objective identification of ischemic necrosis of small intestinal tissue. For this reason, we used a self-built hyperspectral imaging system to identify and classify small intestinal tissues in Japanese Large-Eared White rabbits. A model of small intestinal tissue ischemic necrosis was developed by an experienced physician who blocked part of the superior mesenteric artery of the small intestinal tissue, which was compared with normal blood supplying small intestinal tissue sites. We obtained hyperspectral data of small intestinal tissues from six Japanese large-eared white rabbits in the band between 376-1038 nm with a spectral resolution of 4.69 nm. To reduce band redundancy and a large amount of data, a genetic algorithm was used for band selection of the hyperspectral data. Our results show that the genetic algorithm is suitable for band selection of small intestine tissue hyperspectral data. The band-selected data is classified using the KNN supervised classification algorithm, which has higher classification accuracy and a smaller computational cost. This technique could help physicians objectively identify normal and necrotic areas of small bowel tissue.