Scholarlink

<p>Pulmonary segment segmentation is crucial for cancer localization and surgical planning. The voxel-wise annotation of pulmonary segments is laborious, as the boundaries between segments are indistinguishable. To this end, we propose Anatomy-Hierarchy Supervised Learning (AHSL), a novel weakly supervised learning method. It incorporates anatomical priors into pulmonary segment segmentation through hierarchical supervision. Specifically, AHSL adopts a dual-level supervision paradigm at both the lobe and segment levels. At the lobe level, supervision constrains pulmonary segments to remain within their corresponding lobes. At the segment level, sparse bronchovascular annotations are used to enforce spatial correspondence between each segment and its associated bronchovascular tree. In addition, we introduce a consistency loss based on the L1 norm of the Laplacian to encourage smooth segmental boundaries, along with a new evaluation metric designed to measure the smoothness of the boundaries. Furthermore, we introduce a two-stage segmentation strategy. Specifically, the first stage identifies bronchovascular information, which is then utilized in the second stage to guide the segmentation of the pulmonary segments. Quantitative evaluations on the internal dataset demonstrate that our method outperforms the state-of-the-art methods. AHSL achieves a Dice coefficient of 0.923/0.933 for the mapped artery and 0.924/0.934 for the mapped bronchus on computed tomography pulmonary angiography and non-contrast computed tomography, respectively. In the visual assessments on an independent external dataset, AHSL receives scores of 3.71 and 3.69 on a five-point rating system from two radiologists, respectively, indicating the accurate segmentation with excellent generalization and clinical validity. The proposed AHSL achieves superior performance in pulmonary segment segmentation by integrating hierarchical supervision, consistency loss, and a two-stage segmentation strategy, all without requiring voxel-wise pulmonary segment annotations.</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

Bronchovascular Tree-Guided Weakly Supervised Learning Method for Pulmonary Segment Segmentation

  • 期刊论文

  • 发表日期：

    2026年3月31日

查看全文
  • Ruijie Zhao,
  • Zuopeng Tan,
  • Xiao Xue,
  • Longfei Zhao,
  • Bing Li,
  • Zicheng Liao,
  • Ying Ming,
  • Jiaru Wang,
  • Ran Xiao,
  • Sirong Piao,
  • Rui Zhao,
  • Qiqi Xu,
  • Wei Song

摘要

Pulmonary segment segmentation is crucial for cancer localization and surgical planning. The voxel-wise annotation of pulmonary segments is laborious, as the boundaries between segments are indistinguishable. To this end, we propose Anatomy-Hierarchy Supervised Learning (AHSL), a novel weakly supervised learning method. It incorporates anatomical priors into pulmonary segment segmentation through hierarchical supervision. Specifically, AHSL adopts a dual-level supervision paradigm at both the lobe and segment levels. At the lobe level, supervision constrains pulmonary segments to remain within their corresponding lobes. At the segment level, sparse bronchovascular annotations are used to enforce spatial correspondence between each segment and its associated bronchovascular tree. In addition, we introduce a consistency loss based on the L1 norm of the Laplacian to encourage smooth segmental boundaries, along with a new evaluation metric designed to measure the smoothness of the boundaries. Furthermore, we introduce a two-stage segmentation strategy. Specifically, the first stage identifies bronchovascular information, which is then utilized in the second stage to guide the segmentation of the pulmonary segments. Quantitative evaluations on the internal dataset demonstrate that our method outperforms the state-of-the-art methods. AHSL achieves a Dice coefficient of 0.923/0.933 for the mapped artery and 0.924/0.934 for the mapped bronchus on computed tomography pulmonary angiography and non-contrast computed tomography, respectively. In the visual assessments on an independent external dataset, AHSL receives scores of 3.71 and 3.69 on a five-point rating system from two radiologists, respectively, indicating the accurate segmentation with excellent generalization and clinical validity. The proposed AHSL achieves superior performance in pulmonary segment segmentation by integrating hierarchical supervision, consistency loss, and a two-stage segmentation strategy, all without requiring voxel-wise pulmonary segment annotations.