Colorectal Polyp Segmentation Based on Deep Learning Methods: A Systematic Review
Xin Liu, Nor Ashidi Mat Isa, Chao Chen, Fajin Lv
Abstract
Colorectal cancer is one of the three most common cancers worldwide. Early detection and assessment of polyps can significantly reduce the risk of developing colorectal cancer. Physicians can obtain information about polyp regions through polyp segmentation techniques, enabling the provision of targeted treatment plans. This study systematically reviews polyp segmentation methods. We investigated 146 papers published between 2018 and 2024 and conducted an in-depth analysis of the methodologies employed. Based on the selected literature, we systematically organized this review. First, we analyzed the development and evolution of the polyp segmentation field. Second, we provided a comprehensive overview of deep learning-based polyp image segmentation methods and the Mamba method, as well as video polyp segmentation methods categorized by network architecture, addressing the challenges faced in polyp segmentation. Subsequently, we evaluated the performance of 44 models, including segmentation performance metrics and real-time analysis capabilities. Additionally, we introduced commonly used datasets for polyp images and videos, along with metrics for assessing segmentation models. Finally, we discussed existing issues and potential future trends in this area.