Scholarlink

<p>The extraction of liver vessel centerlines (LVC) and topology (LVT) is of significant clinical value in the diagnosis and treatment of liver vascular diseases. However, extracting LVC and LVT is an extremely challenging and arduous task, with very few quantitative studies available. In this paper, we propose an algorithm for the automatic extraction of LVC and LVT based on multi-scale vascular features, requiring only initial seed points. First, a vessel feature extractor based on the Hessian filter is designed and implemented. The obtained features are mapped through Hessian-Sphere and guided by DBSCAN to determine the next step of the tracker. A bifurcation recognition module is also designed to achieve the extraction of LVT and multi-branch tasks. Additionally, the initial seed point consists of a position point and a direction point. The proposed method has been evaluated and validated on a liver vessel CTA dataset provided by a certain hospital, achieving 90% OV, with LVC reaching 100% OV and OF in single-branch cases. The tracked LVC is close to the reference standard and shows good stability, with AI maintained at ± 1.0mm.</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

Automatic extraction of liver vessel centerlines and topology guided by multi-scale vascular features

  • 期刊论文

  • 发表日期：

    2025年8月5日

查看全文
  • Pengjin Liu,
  • Keyuan Qiu,
  • Zixuan Wang,
  • Quan Qi

摘要

The extraction of liver vessel centerlines (LVC) and topology (LVT) is of significant clinical value in the diagnosis and treatment of liver vascular diseases. However, extracting LVC and LVT is an extremely challenging and arduous task, with very few quantitative studies available. In this paper, we propose an algorithm for the automatic extraction of LVC and LVT based on multi-scale vascular features, requiring only initial seed points. First, a vessel feature extractor based on the Hessian filter is designed and implemented. The obtained features are mapped through Hessian-Sphere and guided by DBSCAN to determine the next step of the tracker. A bifurcation recognition module is also designed to achieve the extraction of LVT and multi-branch tasks. Additionally, the initial seed point consists of a position point and a direction point. The proposed method has been evaluated and validated on a liver vessel CTA dataset provided by a certain hospital, achieving 90% OV, with LVC reaching 100% OV and OF in single-branch cases. The tracked LVC is close to the reference standard and shows good stability, with AI maintained at ± 1.0mm.