Automated bleeding detection in capsule endoscopy videos using statistical features and region growing

Wireless Capsule Endoscopy (WCE) is a technology in the field of endoscopic imaging which facilitates direct visualization of the entire small intestine. Many algorithms are being developed to automatically identify clinically important frames in WCE videos. This paper presents a supervised method for automated detection of bleeding regions present in WCE frames or images. The proposed method characterizes the image regions by using statistical features derived from the first order histogram probability of the three planes of RGB color space. Despite being inconsistent and tiresome, manual selection of regions has been a popular technique for creating training data in the studies of capsule endoscopic images. We propose a semi-automatic region-annotation algorithm for creating training data efficiently. All possible combinations of different features are exhaustively analyzed to find the optimum feature set with the best performance. During operation, regions from images are obtained by applying a segmentation method. Finally, a trained neural network recognizes the patterns of the data arising from bleeding and non-bleeding regions.