Scholarlink

<p>Semi-supervised learning has achieved significant success in the field of medical image segmentation. However, overfitting to erroneous pseudo-labels can lead to cognitive biases in models, a persistent issue in semi-supervised learning that undermines its performance. To address the issues above, we propose a novel semi-supervised segmentation method named as dual task correction framework (DTCF). More specifically, we propose a dual-task collaborative review (DTCR) module, a spatial perception module (SPM), and a high-fidelity pseudo label generation (HFPLG) strategy. The DTCR divides the image into potential error regions and pseudo-correction areas based on segmentation predictions. Subsequently, it guides the network to perform pixel-level or task-level corrections specifically tailored to these different regions. The SPM effectively models long-range relationships between 3D object voxels by capturing contextual dependencies among multiple 2D slices, further enhancing spatial awareness between voxels. The HFPLG dynamically selects high-fidelity pseudo labels by integrating the analysis results of heterogeneous dual subnets. Experiments results on three available benchmark datasets (left atrium, pancreas and brain tumor) reveal that the proposed method yields better segmentation performance compared with the state-of-the-art methods in semi-supervised medical image segmentation. The code is available from <a href="https://github.com/Forrits/DTCF">https://github.com/Forrits/DTCF</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

DTCF: dual task correction framework for semi-supervised medical image segmentation

  • 期刊论文

  • 发表日期：

    2026年2月3日

查看全文
  • Shangwang Liu,
  • Defu Wan

摘要

Semi-supervised learning has achieved significant success in the field of medical image segmentation. However, overfitting to erroneous pseudo-labels can lead to cognitive biases in models, a persistent issue in semi-supervised learning that undermines its performance. To address the issues above, we propose a novel semi-supervised segmentation method named as dual task correction framework (DTCF). More specifically, we propose a dual-task collaborative review (DTCR) module, a spatial perception module (SPM), and a high-fidelity pseudo label generation (HFPLG) strategy. The DTCR divides the image into potential error regions and pseudo-correction areas based on segmentation predictions. Subsequently, it guides the network to perform pixel-level or task-level corrections specifically tailored to these different regions. The SPM effectively models long-range relationships between 3D object voxels by capturing contextual dependencies among multiple 2D slices, further enhancing spatial awareness between voxels. The HFPLG dynamically selects high-fidelity pseudo labels by integrating the analysis results of heterogeneous dual subnets. Experiments results on three available benchmark datasets (left atrium, pancreas and brain tumor) reveal that the proposed method yields better segmentation performance compared with the state-of-the-art methods in semi-supervised medical image segmentation. The code is available from https://github.com/Forrits/DTCF.