A computer aided method to detect bleeding, tumor, and disease regions in Wireless Capsule Endoscopy

Abstract

Wireless Capsule Endoscopy (WCE) is a relatively new technology to record the entire gastrointestinal (GI) tract, in vivo. A large amount of images (frames) are captured during the WCE examination. Reviewing this number of images by a gastroenterologist would be time consuming and prone to human error. Therefore, a diagnostic computer-aided technique is essential to detect and segment regions of abnormalities. In this study, a novel method based on textural features (such as Gabor filters, local binary pattern, and Haralick) in HSV color space, Fisher score test, and neural networks is presented to detect and differentiate regions such as bleeding, tumor, and other types of gastric diseases including Crohn's, Lymphangectasia, Stenosis, Lymphoid Hyperslasia and Xanathoma. The experimental results indicate that this method is able to classify a lesion from a normal region in every single frame and group them into normal and abnormal frames to be considered for surgery/treatment planning by an expert.