GI Bleeding Detection in WCE Images Using E-ORB and ME-DEEP CAPSNET

Abstract

Wireless Capsule Endoscopy (WCE) is utilized in the detection of several anomalies like bleeding, ulcers, polyps, and tumors in the gastrointestinal (GI) tract. As a huge number of images are produced by WCE, the manual examination becomes much more tedious, time-consuming, and furthermore increasing the possibility of human errors. Therefore, a new automated scheme to detect bleeding region in WCE images by means of deep learning technique is proposed in this approach. Initially, the WCE input images are pre-processed by means of distribution linearization and linear filtering. An Adaptive dense U-Net based segmentation approach is employed for the segmentation of pre-processed image. The feature point extraction is dome using Enhanced Oriented fast and Rotated BRIEF (ORB) algorithm. The detection process and the classification of detected region as bleeding and non-bleeding region is carried by Multi- Enhanced Deep CapsNet classification model. The performance assessment is carried in terms of accuracy, sensitivity, specificity, precision, recall, F-measure, FNR, and FPR, and the outcomes acquired are compared with existing methods to validate the improvement of proposed scheme