多中心标注的CoW数据集
收藏资源简介:
该数据集由中国科学院团队构建,是首个大规模多中心标注的Circle of Willis(CoW)血管分割专用数据集,整合了来自ADAM、BraVa等14个公开源的1341例时间飞跃法磁共振血管造影(TOF-MRA)影像,覆盖不同扫描设备和人群特征。数据经过标准化预处理,包含20类CoW动脉结构的精细标注,特别针对小血管(如后交通动脉)进行了半径信息标注。通过模板配准和ROI裁剪实现了跨中心数据空间对齐,为血管拓扑连续性研究和多类分割算法开发提供了重要基准。数据集主要应用于脑血管疾病诊断、阿尔茨海默病血管形态学生物标志物发现等神经影像研究领域。
This dataset was constructed by a team from the Chinese Academy of Sciences, and it is the first large-scale multi-center annotated dedicated dataset for Circle of Willis (CoW) vascular segmentation. It incorporates 1341 time-of-flight magnetic resonance angiography (TOF-MRA) images from 14 public sources including ADAM and BraVa, covering diverse scanning equipment and population characteristics. The data has undergone standardized preprocessing, with fine-grained annotations for 20 types of CoW arterial structures; specifically, radius information annotations are provided for small vessels such as the posterior communicating artery. Spatial alignment of cross-center data is realized through template registration and ROI cropping, providing an important benchmark for vascular topological continuity research and the development of multi-class segmentation algorithms. This dataset is primarily applied in neuroimaging research fields such as cerebrovascular disease diagnosis and the discovery of vascular morphological biomarkers for Alzheimer's disease.

- 1AG-TAL: Anatomically-Guided Topology-Aware Loss for Multiclass Segmentation of the Circle of Willis Using Large-Scale Multi-Center Datasets中国科学院·自动化研究所; 中国科学院·脑认知与脑机智能技术国家重点实验室; 中国科学院大学·人工智能学院; 中国科学院大学·未来技术学院 · 2026年



