A statistical feature based novel method to detect bleeding in wireless capsule endoscopy images

Wireless capsule endoscopy (WCE) is a recently developed technology to detect small intestine diseases, such as bleeding. In this paper, a scheme for automatic bleeding detection from WCE video is proposed based on different statistical measures computed from a new red to green (R/G) pixel ratio intensity plane of RGB color images. Different statistical parameters, namely mean, mode, maximum, minimum, skewness, median, variance, and kurtosis are used to extract variation in spatial characteristics in R/G intensity plane of bleeding and non-bleeding WCE RGB images. Depending on the ability to provide significantly distinguishable characteristics, in the proposed feature vector, median, variance, and kurtosis of R/G ratio values corresponding to a WCE image are considered. For the purpose of classification, K-nearest neighbor (KNN) classifier is employed. From extensive experimentation on several WCE videos collected from a publicly available database, it is observed that the proposed method can successfully detect bleeding and non-bleeding images with high level of accuracy, sensitivity and specificity in comparison to that of some of the existing methods.