Automatic Polyp Detection in Wireless Capsule Endoscopy Images using Hybrid Patch Extraction and Supervised Classification

Wireless Capsule Endoscopy (WCE) is a non-invasive and painless novel technique for diagnosing gastrointestinal disease. Computer Aided Diagnosis system is a solution to reduce the great burden of a doctor to examine images frame by frame to locate abnormalities. In this paper an automated computer aided detection system with image analysis and supervised learning method is proposed to detect polyp from WCE images. Textural features are extracted not only from single key point, by utilizing single scale-invariant feature also from neighborhood key points. Haralick texture features are extracted from each of patch size of 16*16 around the key points. After acquiring the different texture features, we develop a strategy to integrate these features with Key point descriptor. For the best classification performance, SIFT is integrated with 22 Haralick textural features. The supervised classification is performed using a Multilayer perceptron Neural Network. In our proposed method, Feature based classification performance is better than other classifiers and correctly identifies polyp images with accuracy about 98.8%.