CBISC: A Novel Approach for Colon Biopsy Image Segmentation and Classification

Abstract

The morphology of epithelial cells plays a vital role in distinguishing malignant colon tissues from the normal ones. Epithelial cells have near elliptic shape in normal colon tissues, whereas they deform into an amorphous shape in malignant tissues. The information about the morphology of epithelial cells may be incorporated in order to obtain an effective segmentation of colon biopsy images. In this research study, we propose a novel colon biopsy image segmentation and classification (CBISC) technique that does so. The proposed CBISC technique comprises two main modules, namely, segmentation and classification. The segmentation module exploits the background information about morphology of epithelial cells, and detects elliptic and nearly elliptic epithelial cells in four orientations. It further calculates three novel features, namely, semi-major axis, direction, and area occurrence for each image pixel. Finally, it grows and merges regions based on these features, and demarcates final region boundaries. Genetic algorithm has been employed to optimize several parameters used in the segmentation process. A dataset comprising 300 colon biopsy images has been used for the evaluation of proposed segmentation module, and improved performance has been observed compared to previously reported techniques. To validate the effectiveness of segmentation, moments of gray-level histogram and gray-level co-occurrence matrix-based features have been extracted from 710 segmented patches of the images, and have been used for the classification of segmented regions into normal and malignant classes. Radial basis function kernel of support vector machines has been used for classification, and reasonable classification results have been obtained.