Scholarlink

<p>Medical image segmentation plays a crucial role in clinical diagnosis and treatment planning. However, labeled data for medical images is scarce, and foreground–background class imbalance poses a significant challenge. To address these issues, we propose a semi-supervised learning framework that combines an imbalanced pseudo-labeling method with an enhanced Big-nnU-Net. Our approach filters unreliable pseudo-labels iteratively, generating a reliable pseudo-label dataset. Experiments on lower extremity vessel and abdominal organ segmentation datasets demonstrate that our method improves Dice similarity coefficient scores, effectively handling class imbalance and limited labels. Our framework shows promise in enhancing segmentation performance in real-world medical applications. The source code is available at <a href="https://github.com/knowleton/IPL-Net">https://github.com/knowleton/IPL-Net</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

Optimizing semi-supervised medical image segmentation with imbalanced filtering and nnU-Net enhancement

  • 期刊论文

  • 发表日期：

    2025年2月13日

查看全文
  • Yuanchuan Duan,
  • Peng Wang,
  • Yan Huang,
  • Yuxin Hang,
  • Qi Sun,
  • Haibo Shao,
  • Jinzhu Yang

摘要

Medical image segmentation plays a crucial role in clinical diagnosis and treatment planning. However, labeled data for medical images is scarce, and foreground–background class imbalance poses a significant challenge. To address these issues, we propose a semi-supervised learning framework that combines an imbalanced pseudo-labeling method with an enhanced Big-nnU-Net. Our approach filters unreliable pseudo-labels iteratively, generating a reliable pseudo-label dataset. Experiments on lower extremity vessel and abdominal organ segmentation datasets demonstrate that our method improves Dice similarity coefficient scores, effectively handling class imbalance and limited labels. Our framework shows promise in enhancing segmentation performance in real-world medical applications. The source code is available at https://github.com/knowleton/IPL-Net.