Automatic informative tissue's discriminators in WCE

Wireless capsule endoscopy (WCE) is a new device which investigates the entire gastrointestinal (GI) and especially small bowel. About 55000 frames are recorded in an examination for a capsule which captures two frames per second. Thus, it is essential to find an automatic and intelligent method to help physicians. The WCE videos have lots of uninformative parts (such as extraneous matters, bubbled, and dark part), so preprocessing is necessary to separate these uninformative regions in a frame or reduce frames' numbers. In this paper, we introduce two novel methods to detect automatically uninformative parts. In order to achieve this goal, we use two Mathematical Morphological operations, sigmoid function as a method to segment regions, statistic features, Gabor filters, fisher score test to reduce number of features, neural network and discriminators in color space. Our experimental studies indicates that precision, sensitivity, accuracy, and specificity are respectively 96.13%, 95.30%, 96.35% and 97.00% in the first method, and 90.17%, 95.68%, 93.72%, and 92.71%, respectively in the second method.