Abstract: Colorectal polyp is an abnormal tissue growing in the gastrointestinal tract with the potential to develop into colorectal cancer. Therefore, early detection and removal of colorectal polyps are crucial for preventing colorectal cancer. In recent years, deep learning technology has made significant strides in the field of colonic polyp image segmentation, substantially enhancing both the accuracy and automation levels of segmentation. This paper focuses on research related to deep learning in colorectal polyp image segmentation. Firstly, it introduces various imaging techniques for colonic polyps and commonly used datasets, including both image and video datasets, and elaborates on the characteristics of these datasets. Subsequently, the deep learning-based segmentation methods are summarized, covering fully convolutional networks, Mask R-CNN, generative adversarial networks, U-Net, Transformer, and multi-network fusion models. Particular emphasis is placed on the application of U-Net and its variants in colonic polyp image segmentation, analyzing their structural improvements, performance enhancements, and practical application outcomes. Furthermore, the review comprehensively compares the main improvements, advantages, disadvantages, and segmentation results of each network model. Finally, it points out the main challenges currently faced by deep learning in this field and provides an outlook on future research directions.

Key words: colonic polyp segmentation, deep learning, medical imaging, convolutional neural networks, U-Net

摘要： 结肠息肉是一种可能发展为结直肠癌的胃肠道异常生长组织，因此，早期检测和切除结肠息肉对预防结直肠癌具有重要意义。近年来，深度学习技术在结肠息肉图像分割领域中的应用取得了显著进展，大幅提高了分割的准确性和自动化水平。针对深度学习在结肠息肉图像分割中的研究展开综述，介绍了多种结肠息肉成像方式及包括图片和视频在内的常用数据集，并详细说明了这些数据集的特点。深入阐述了基于深度学习的结肠息肉分割方法，涵盖了全卷积网络、Mask R-CNN、生成对抗网络、U-Net、Transformer以及多网络融合模型，其中重点强调了U-Net及其变体在结肠息肉图像分割中的应用，分析了其结构改进、性能提升和实际应用效果。同时，综合对比了各网络模型的主要改进思路、优缺点及其分割结果。指出了当前深度学习在该领域面临的主要挑战，并对未来的研究方向进行了相应的展望。

关键词: 结肠息肉分割, 深度学习, 医学图像, 卷积神经网络, U-Net