Classification of gastrointestinal diseases of stomach from WCE using improved saliency-based method and discriminant features selection

Wireless capsule endoscopy (WCE) is a new imaging procedure that is used to record internal conditions of gastrointestinal tract for medical diagnosis. However, due to the presence of bulk of WCE image data, it becomes difficult for the physician to investigate it thoroughly. Therefore, considering aforementioned constraint, lately gastrointestinal diseases are identified by computer-aided methods and with better classification accuracy. In this research, a new computer-based diagnosis method is proposed for the detection and classification of gastrointestinal diseases from WCE images. The proposed approach comprises of four fundamentalsteps:1) HSI color transformation before implementing automatic active contour segmentation; 2) implementation of a novel saliency-based method in YIQ color space; 3) fusion of images using proposed maximizing a posterior probability method; 4) fusion of extracted features, calculated using SVD, LBP, and GLCM, prior to final classification step. We perform our simulations on our own collected dataset - containing total 9000 samples of ulcer, bleeding and healthy. To prove the authenticity of proposed work, list of statistical measures is considered including classification accuracy, FNR, sensitivity, AUC, and Time. Further, a fair comparison of state-of-the-art classifiers is also provided which will be giving readers a deep inside of classifier's selection for this application. Simulation results clearly reveal that the proposed method shows improved performance in terms of segmentation and classification accuracy.