Automatic polyp detection for wireless capsule endoscopy images

Wireless capsule endoscopy (WCE) opens a new stage for diagnosing gastrointestinal tract diseases since it enables direct visualization of the small intestine for the first time. However, it requires a clinician’s long time inspection due to a great number of images produced by the procedure. Therefore, it may be beneficial to devise an automatic detection system to help clinicians identify problematic images. In this work, we attempt to design a computerized scheme aiming for polyp WCE image recognition though polyp in WCE images show great variations in appearance. This scheme utilizes a new texture feature to characterize WCE images, which integrates advantages of wavelet transform and uniform local binary pattern. With support vector machine (SVM) as a classifier, extensive experiments on our present image data, which consists of 600 normal WCE images and 600 polyp WCE images chosen from 10 patients, verify that it is promising to utilize the proposed scheme to detect polyp WCE images.