Answers are grounded in this paper and cite their sources.
Semi-supervised medical image segmentation methods have drawn wide attention as they reduce reliance on heavily annotated data. However, existing models suffer from confirmation bias with limited annotations, and structural or parameter coupling hinders self-correction, especially for medical images with ambiguous boundaries, low contrast and complex backgrounds. To address these issues, we propose an uncertainty-guided decoupling and complementary network (UGDC-Net) for binary medical image segmentation. It uses a contrastive mechanism to filter high-uncertainty regions of sub-networks for avoiding model collapse, and a dynamic competition mechanism to tackle the weight coupling bottleneck in teacher-student frameworks. Reliable pseudo-labels from high-performance sub-networks and supplementary ones from inconsistent high-uncertainty regions are integrated to provide diverse supervision, enhancing the model's exploration of small targets. Experiments on left atrium, Pancreas-CT and ISIC datasets validate that UGDC-Net outperforms state-of-the-art methods on binary segmentation tasks.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: