MSCFF-Net : Multi-scale context feature fusion network for polyp segmentation

Abstract Accurate segmentation of polyps in colonoscopic images is a key step in computer-aided diagnosis of colorectal cancer. Although the current segmentation algorithm has some achievements in the field of polyp segmentation, there are still some challenges. The size and shape of the polyp area are different, and the boundary with the background is not obvious. In order to solve the above problems, we propose a new multi-scale context information fusion network(MSCFF-Net). Specifically, the network first uses pyramid transformer (PVTv2) as the encoder, and designs a feature interactive decoder (FID) to obtain a rough location map of the polyp area. Then, four multi-stage feature fusion modules (MSFF) are designed to realize the interaction of multi-stage feature information and enrich the scale diversity of polyp features. Finally, multi-scale attention (MSA) is introduced behind the multi-stage fusion module to improve the attention of the model to polyp features. Experiments on two public polyp datasets show that MSCFF-Net is superior to other advanced polyp segmentation methods.