SFS-Net: A Method for Medical Image Segmentation Based on Shallow Feature Supplementation and Enhancement

Medical image segmentation serves as a fundamental prerequisite for clinical diagnosis and treatment planning. Despite their remarkable success, existing methods often disproportionately focus on high-level semantic information. Consequently, these approaches tend to neglect shallow features, such as fine-grained tissue textures and precise boundary cues, resulting in suboptimal segmentation for small organs and ambiguous regions. To address this issue, we propose SFS-Net, a novel framework that explicitly preserves and utilizes shallow features. The encoder incorporates a designed shallow feature enhancement (SFE) module aimed at fortifying the representation of fine-grained image details to prevent the loss of critical information during the segmentation process.We also devise a shallowfeature supplementation (SFS) module within the decoder to effectively restore missing features in ambiguous regions, targeting areas with indistinct details or sparse information. Extensive evaluations on four datasets across three distinct tasks demonstrate that our method outperforms existing state-of-the-art (SOTA) segmentation approaches and yields significant performance improvements.