FDAPNet: Feature Denoising and Aggregation Prediction Network for Polyp Segmentation

Automatic polyp segmentation can assist doctors in screening colonoscopy images, which is crucial for preventing colorectal cancer. Although existing methods have achieved promising results, they are susceptible to interference from surrounding tissues and lack adaptability to different sizes of polyps. To address these issues, we propose a feature denoising and aggregation prediction network (FDAPNet) for polyp segmentation. Specifically, we first design a feature denoising module to eliminate the background noise in shallow features, allowing the network to focus more on the target area. Then, a channel attention module is proposed to automatically emphasize relevant features while suppressing irrelevant ones, which can enhance the network’s ability to capture important information. Moreover, we propose a multi-scale feature aggregation prediction module to better deal with polyps of various sizes. We conduct extensive experiments on five public datasets. The quantitative and qualitative results demonstrate that our FDAPNet outperforms other state-of-the-art polyp segmentation methods.