Answers are grounded in this paper and cite their sources.
Remarkable progress has recently been made in semi-supervised medical image segmentation. Nevertheless, issues related to unreliable pseudo-labels and insufficient feature discriminability remain unresolved. To mitigate the aforementioned limitations, a multi-level collaborative structural perception learning network (SHAPE-Net) is proposed based on a teacher-student framework from the perspective of structural consistency, enabling the unified and coordinated optimization of input perturbations, semantic representations and discriminative learning. Specifically, SHAPE-Net introduces Confidence-Guided Patch Displacement at the input layer to alleviate semantic drift caused by strong augmentations, enhances cross-sample semantic interactions through Multi-view Prototype Learning at the semantic level, and strengthens boundary-region representations by integrating Hard-aware Contrastive Learning with Feature-level Consistency Alignment at the feature level. As a result, the robustness of pseudo-labels and the representation capability of lesion regions are synergistically improved, thereby yielding more robust and precise segmentation results in complex medical scenarios. Extensive experiments conducted on three public medical image datasets validate the effectiveness of SHAPE-Net, which consistently achieves consistently improved segmentation performance across various annotation ratios, especially under limited-label settings and in anatomically complex boundary regions. In particular, under the 10% labeled-data setting, Dice scores of 92.15%, 79.14% and 86.22% were achieved on the LA, Pancreas-CT and BraTS-2019 datasets, respectively. Furthermore, under the 20% labeled-data setting, robust and stable segmentation performance was consistently achieved, surpassing that of existing state-of-the-art semi-supervised methods. In addition, ablation studies further verify the effectiveness and complementarity of each module and demonstrate the applicability and robustness of the proposed method in complex medical image segmentation tasks.
No takes yet. Share an insight, caveat, or question.
Shen et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: