Computer aided diagnosis system for ulcer detection in capsule endoscopy using optimized feature set

One of the most common lesions of the gastrointestinal tract (GIT) is an ulcer. Capsule endoscopy (CE) is a recent advancement in the field of gastroenterology for diagnosis of GIT abnormalities. However, CE video length ranges from 6 to 8 hours generating approximately 60000 images. For a medical expert, examination of such lengthy videos is time-consuming and tiresome. Also, the accuracy of diagnosis will largely depend upon individual expertise. A computer-aided diagnosis (CAD) system can significantly improve accuracy and reduce diagnosis time. In the proposed automated ulcer detection system, relevant features of the histogram of oriented gradients (HOG) and uniform local binary patterns (LBP) are optimally selected by high variance low correlation (HVLC) based novel feature selection technique and the classification task is performed using support vector machine (SVM). Proposed feature selection technique reduces the feature set by 96.53% and outperforms five other state of the art feature selection techniques. The performance of proposed system is compared with other systems and it performs with accuracy, F measure and sensitivity of 95%, 95.12%, and 97.5% respectively.