Scholarlink

Obtaining accurate 3D blood vessel segmentation masks is difficult, as automatic methods often produce incomplete (false-negative) results. This paper presents an interactive framework to identify and adjust these errors. The objective is to obtain a reference-quality mask by combining automatic anomaly detection with expert correction. Our method automatically detects discontinuities by first determining the median lines (centerlines), then labeling the resulting connected components, identifying their extremities, and finally determining their 3D orientation. Based on this analysis, the system proposes possible connections. These are presented to the user, who can interactively edit them. The result is a complete centerline, which serves as the basis for a topologically correct segmentation mask.

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

A Human-in-the-Loop Framework for Topological Completion of 3D Vascular Segmentations

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Codruț-Georgian Artene,
  • Nicolae-Alexandru Botezatu,
  • Paul-Corneliu Herghelegiu

摘要

Obtaining accurate 3D blood vessel segmentation masks is difficult, as automatic methods often produce incomplete (false-negative) results. This paper presents an interactive framework to identify and adjust these errors. The objective is to obtain a reference-quality mask by combining automatic anomaly detection with expert correction. Our method automatically detects discontinuities by first determining the median lines (centerlines), then labeling the resulting connected components, identifying their extremities, and finally determining their 3D orientation. Based on this analysis, the system proposes possible connections. These are presented to the user, who can interactively edit them. The result is a complete centerline, which serves as the basis for a topologically correct segmentation mask.