Bridging the Semantic Gap: Synergistic Feature Fusion and Multi-Scale Adaptation for Medical Image Segmentation

Medical image segmentation is pivotal for clinical diagnosis but faces a systemic bottleneck in handling complex scenarios due to the inherent trade-offs in standard architectures. Existing methods often struggle with the "perception-representation dilemma" at the input stage and a "fusion-adaptation misalignment" during feature aggregation. To dismantle these interconnected challenges, we propose a novel Synergistic Fusion and Refinement Network (SFR-Net). Specifically, we design a Local-Regional Feature Perception (LRFP) module to couple fine-grained details with global context from the outset. To bridge the semantic gap, we introduce a Channel Refinement and Enhancement Module (CREM) as an intelligent gatekeeper in skip connections, alongside a Feature Mixing Module (FMM) to dynamically adapt to multi-scale targets at the bottleneck. Extensive experiments on four diverse datasets (CVC-ClinicDB, ISIC 2017, TN3K, and MICCAI Tooth) demonstrate that SFR-Net effectively overcomes these systemic limitations, achieving state-of-the-art performance in terms of accuracy and robustness.