C-TUnet: Cross-Type TransUnet for Colon Polyp Segmentation

Checking for colorectal polyps is an effective way to detect colon cancer. Therefore, it is important to identify and segment polyps from colonoscopy images in clinical experiments. However, the segmentation of polyps is a challenging task. There are two main reasons. The first is that the boundary between polyps and surrounding human tissue is not clear, and the second is that it is difficult to distinguish the texture features between polyps and human tissue. To address these issues, we propose a novel architecture, Cross-type TransUnet. We use long jump links to ensure the connection of texture features at different levels after convolution, and then design a texture enhancement module to extract texture features at different levels into semantic features. We use the Kvasir-SEG data set to conduct comparative experiments with multiple classic methods. The experimental results show that our methods are better than them.