Multi-Scale Attention Fusion Supervised Network

Medical image segmentation aims to analyze tissue structures, providing support for lesion localization and assisted diagnosis. Polyp segmentation is a challenging task due to the similarity in color between polyps and the background, which imposes high demands on feature extraction and segmentation accuracy. To address the issue of blurred boundaries, this paper proposes a Multi-Scale Attention Fusion Supervised Network (MSAFS-Net). The framework is based on a U-Net architecture and incorporates two key modules: the Contextual Information Capture Module (CICM) and the Contextual Information Fusion Module (CIFM). CICM consists of global and local information capture modules to effectively extract multi-scale contextual information. CIFM optimizes and integrates global and local features while improving segmentation accuracy through deep supervision. Experimental results demonstrate that MSAFS-Net outperforms existing models on the Kvasir-SEG dataset, exhibiting outstanding performance.