Texture based classification of hyperspectral colon biopsy samples using CLBP

Computer aided diagnosis (CAD) is aimed at supporting the pathologists in their diagnosis. In this paper, we present an algorithm for texture-based classification of colon tissue patterns. In this method, a single band is selected from its hyperspectral cube and spatial analysis is performed using circular local binary pattern (CLBP) features. A novel method for feature selection is presented resulting in the best feature set without actually running the classifier. Classification results using Gaussian kernel SVM, with an accuracy of 90%, demonstrate that texture analysis based on CLBP features is able to distinguish the benign and malignant patterns.