Classification of Colon Biopsy Samples by Spatial Analysis of a Single Spectral Band from its Hyperspectral Cube

The histopathological analysis of colon biopsy samples is a very important part of screening for colorectal cancer. There is, however, significant inter-observer and even intra-observer variability in the results of such analysis due to its very subjective nature. Therefore, quantitative methods are required for the analysis of histopathological images to aid the histopatholgists in their diagnosis. In this paper, we exploit the shape and structure of the gland nuclei cells for the classification of colon biopsy samples using two-dimensional principal component analysis (2DPCA) and Support Vector Machine (SVM). We conclude that the use of textural features extracted from non-overlapping blocks of the histopathological images results in a non-linear decision boundary which can be efficiently exploited using a SVM with appropriate choice of parameters for its Gaussian kernel. The SVM classifier outperforms all the remaining methods by a clear margin.