Automatic polyp recognition from Colonoscopy images based on bag of visual words

Abstract

Colorectal cancer (CRC) is one of the most popular cancer in the world. Adenoma and sessile serrated polyp precursor lesions claim over 95% of CRC. The incidence of CRC is reduced 76-90% through the early diagnosis and removal of colorectal polyps. Colonoscopy is the golden standard for the detection of colorectal polyps but about 25% of polyps were missed during colonoscopy examinations. In this study, we proposed a novel method to recognize polyps from colonoscopy images based on bag-of-visual-words (BoW) with extracted regions of interest. The proposed method generates a histogram of visual word occurrences to represent an image, and uses support vector machine (SVM) with error correcting output codes (ECOC) for the detection of polyps. A dataset composed of 131 cases' clinical data were used to train and test the proposed method. Validation demonstrates an average specificity of 97.8±1.5%, an average sensitivity of 97.2±1.7%, and an average accuracy of 97.5±1.0%.