Robust abnormal Wireless Capsule Endoscopy frames detection based on least squared density ratio algorithm

Abstract

Wireless Capsule Endoscopy (WCE) constitutes a recent technological breakthrough that enables the observation of the gastrointestinal tract (GT) and especially the entire small bowel in a non-invasive way compared to the traditional imaging techniques. A primary difficulty with the management of WCE videos is that reviewing capsule endoscopic video is a labour intensive task and very time consuming. Also the diagnosis process by WCE videos is not real-time. In order to address those difficulties and limitations, we propose a new framework by defining Frame Abnormality Index (FAI) using the ratio of training and testing data densities, where training dataset only consist of normal samples and testing dataset consist of both normal and abnormal samples. In this paper, we use Least Square-based algorithm to estimate density ratio parameters without involving density estimation. Actual clinical patient frames including various abnormal frames are used to evaluate the performance of the proposed method. Experiments show that our proposed method is efficient and effective to detect abnormal frames in WCE videos.