SAM-DCE: Decoupled Context Enhancement for Prompt-Free Medical Image Segmentation
SAM-DCE is a prompt-free framework that improves medical image segmentation using SAM by addressing token uniformity and semantic over-smoothing. It introduces a dual module, ML-DCE, which combines high-level semantic information from deep layers (MCC) with spatial details from shallow features (ICC). This refinement enhances boundary accuracy and category consistency compared to standard SAM.