Gastrointestinal tract bleeding detection from wireless capsule endoscopy videos

Abstract

Wireless Capsule Endoscopy (WCE) videos reviewing for the presence of disease signs is a very time-consuming and presents a burdensome for the clinicians. Hence, there is an urgent need to develop algorithms to automatically identify clinically important frames in WCE videos. In this paper, we present a method for automatic detection of the bleeding region in WCE images. We first enhance the WCE images before employing a colour based segmentation method. Then, we propose to extract texture and statistical features from the segmented regions to avoid false detections. Finally, support vector machine is used for region classification. The results showed that the capability of detecting bleeding was much improved and the proposed approach could obtain classification accuracy up to 94.4%.