Diverse Attentions Feature Fusion Network for Polyp Segmentation

Accurate polyp segmentation is crucial for the early diagnosis and prevention of colorectal cancer. However, precise segmentation of polyps remains challenging due to their ambiguous boundaries and diverse shapes and sizes. To address these issues, we propose a novel diverse attention feature fusion network for polyp segmentation. Specifically, we first introduce a Boundary Generation Module, which learns boundary features by integrating low-level and high-level features. Additionally, a Scale-perception Convolution Module (SCM) is proposed to learn multi-scale features through different dilation rates in convolutional layers, enabling the segmentation network to obtain various receptive fields and thus recognize polyps of different sizes and shapes. Furthermore, we propose a Scale-level module to fuse Scale-level features while simultaneously exploring local and global contextual information. We also introduce a Diverse Attentions module that provides three distinct types of attention, namely boundary, foreground, and background attention, which facilitates providing different attentions to the input SFM features. We evaluate our model on two datasets, and the experimental results demonstrate the effectiveness of our proposed model.