Tumor detecting in colonoscopic narrow-band imaging data

In recent years, colorectal cancer is the most common type of cancer worldwide. However, this disease can be prevented by detection and removal of precursor adenomatous polyps during an optical colonoscopy course. In an optical colonoscopy course, the endoscopist looks for colon polyps. Hyperplastic polyp is benign lesion; adenomatous polyp is likely to become cancerous. Hence, it is common practice to remove all of the identified polyps and send them to subsequent histological analysis. But, removing hyperplastic polyps will pose unnecessary risk and incur unnecessary costs for histological analysis. In this paper, we pose a novel optical biopsy application based on narrow band imaging (NBI) in the first part. A common barrier to an automatic system is that require manual segmentations of the polyps in NBI data, therefore, we automatically segment polyps in NBI data. We propose an algorithm, Classification of Regional Feature (CoRF), that is an extension of the sparse matrix and vector quantization algorithms, a state of the art algorithm for feature detection and segmentation. CoRF solves the intrinsic block selection problem of vector quantization by including training codebook about the shape of the regional feature. CoRF outperforms previous methods with a better polyp region segmentation - traditional clustering algorithm, LBG or Kmean clustering algorithms will disperse the energy to other similar block regions, reducing accuracy of analysis in polyp detection.