Bleeding Fragment Localization Fusion Time Domain Characteristics For WCE Sequence

Small intestine bleeding is one of the most common diseases in human gastrointestinal tract. With the progress and development of the Wireless Capsule Endoscopy (WCE), it is already widely used in a variety of disorder analyses, especially for the detection of gastrointestinal bleeding. One WCE video sequence usually contains around 50,000-60,000 frames. Doctors have to detect all frames one by one to inspect patients’ gastrointestinal tract. Obviously, it is a tremendous project for doctors to deal with. Therefore, high efficiency and high accuracy computer aided diagnosis methods are demanded.In this paper, a new time domain based method was proposed to locate bleeding fragment in WCE video sequence. We combined with image preprocessing, super pixel segmentation, color model, feature fusion and time domain model to optimize bleeding location detection problem of wireless capsule endoscopy sequence. First, data set of wireless capsule endoscopy sequence was established. Secondly, we analysed and compared the existing feature extraction algorithms of bleeding detection. This paper focus on CBRF algorithm, which we removed the background, the wrinkled and black holes interfere in enteric cavity respectively in each wireless capsule endoscopy video sequence frame. Also, each sequence frame was segmented into small patches by gSLIC algorithm which greatly improve the processing speed. At the same time, to against the old hemorrhage in video sequence segment, we combined a color model with SVM learning algorithm to detect pre-optimized. In our algorithm, the red ratio feature and the temporal a feature were extracted. Finally, bleeding fragment was detected based on the temporal red ratio feature and the temporal a features fusion. Then, compared with the state of art WCE bleeding detection algorithm which based on a single image, this paper proposed a algorithm used information between the consecutive frames to optimize the detecting process.Since video sequence in the context of changes in the intestinal wall, time series can provide more information. Compared with traditional algorithms, algorithm which we based on video sequence in detecting bleeding can get better results. The experimental results show the proposed method has a promising performance.