Skip to content
Open access

DSFE-Net: A Decoupled Semantic-Guided Network With Region-Boundary Refinement for Medical Image Segmentation

2026 · IEEE Access · Vol 14, pp. 165575-165591 · 0 citations · 51 references

Abstract

Medical image segmentation is essential support clinical decision-making, surgical planning, and therapeutic assessment. However, there are two challenges for accurate medical image segmentation. First, blurred target boundaries in medical images hinder accurate target delineation. Second, target-like background structures may be mis-segmented as regions of interest, leading to false-positive errors. To address these challenges, we propose a Decoupled Semantic-guided Feature Enhancement Network (DSFE-Net). Specifically, we introduce a Multi-scale-Fusion-based Semantic Decoupling (MFSD) module to fuse multi-scale encoder features and decouple them into foreground, background, and uncertainty regions. Furthermore, we design a Semantic-Guided Feature Enhancement (SGFE) module to exploit the aforementioned foreground and background features for highlighting salient target regions and suppressing background regions. Finally, we propose a Region-Boundary Refinement Decoder (RBR-Decoder), which exploits the consistency and discrepancy of adjacent-scale features to enhance target region and boundary sensitivity, thereby improving target localization and boundary delineation accuracy. Extensive experiments on six medical image segmentation tasks demonstrate that the proposed method achieves competitive segmentation performance across diverse medical imaging scenarios.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.