A survey of deep learning algorithms for colorectal polyp segmentation

Abstract

Early detecting and removing cancerous colorectal polyps can effectively reduce the risk of colorectal cancer. Computer intelligent segmentation techniques (CIST) can improve the detection rate of polyp by drawing the boundaries of colorectal polyps clearly and completely. Four challenges that encountered in deep learning methods for the task of colorectal polyp segmentation are considered, including the limitations of classical deep learning (DL) algorithms, the impact of data set quantity and quality, the diversity of intrinsic characteristics of lesions and the heterogeneity of images in different center datasets. The improved DL algorithms for intelligent polyp segmentation are detailed along with the key neural network modules being designed to deal with above challenges. In addition, the public and private datasets of colorectal polyp images and videos are summarized, respectively. At the end of this paper, the development trends of polyp segmentation algorithm based on deep learning are discussed.