Detection of bleeding patterns in WCE video using multiple features

Wireless capsule endoscopy (WCE) uses a pill size capsule which is easily swallowed by patients, that comes with wireless power transmission. WCE images are recorded into a hard disk to be later analyzed by a doctor. This paper studies various bleeding characteristics and proposes a computer- based procedure to reduce the analysis time for detecting suspected bleeding diseases in patients. A two-step process is proposed, the first to discriminate images with or without ´bleeding´, using a block-based color saturation method. The second step, to refine the initial classification and increase its reliability uses a pixel-based saturation-luminance analysis. The luminance-saturation relationship is explored to reduce the false positive detections resulting from Step 1. Another analysis tool is proposed, exploring the red color component, to analyze the presence of food remains or bubbles, partly occluding the tissues. Experimental results suggest that bleeding patterns can be effectively detected using the proposed method, to guide the specialist to images with bleeding patterns. This proposal is added into the previously proposed framework, Capsule Endoscopy Supporting Software (CESS), consisting of multiple tools to analyze WCE images.