ABNORMALITY DETECTION FOR CAPSULE ENDOSCOPE IMAGES BASED ON SUPPORT VECTOR MACHINES

Wireless capsule endoscopy (WCE) represents a significant technical breakthrough for the investigation of intestines. It can be used to examine entire section of the intestines, including the blind section that is not reachable with a traditional endoscope. However, one problem with this new technology is that too many images need to be examined by eyes to detect the normal and/or abnormal images and it becomes a burden to physicians. This paper presents some potential methods for an automatic detection system to identify suspected capsule endoscope images containing either chyme blocked, suspected blood indicator, or white spot abnormality in order to reduce this burden. These methods use color and texture of images as recognition features for the classifiers such as support vector machines (SVM), imbalanced SVM, and total margin-based adaptive fuzzy SVM. For comparison, the nearest neighbor (NN) classifier is also considered. Experimental results, carried out on 10-runs of 5-fold cross validation, show that the combination of hue-saturation (HS) histogram using relevant features (64 bins), image downsampling factor by 1, and TAF-SVM performs the best. The resulting accuracy obtained is 98.13% and the computational time during the testing phase is below 0.5 seconds per image, which is useful for practical applications.