Cerebral MRA Dataset Annotations
This page provides manual annotations of the publicly available UNC dataset, supporting research in vascular landmark detection and semantic segmentation. The dataset comprises 109 MRA scans from healthy volunteers. The annotations contain 19 bifucation landmarks and 20 vascular segments in the Circle of Willis (CoW) region. Detailed definition of the annotations are shown as below. Notably, physiological structural variations (missing vascular segments) are observed in certain samples.
Download
Two versions of annotations are available for download for non-commercial research and educational purposes:
Origin Data: Landmark annotations in the origin data: Landmark-Annotation.zip
The corresponding origin images can be found on the UNC website:
UNC-dataset
After Pre-processing:
Intensity-based rigid registration was performed on all scans by taking a sample as the template. The scans were cropped to 192 × 160 × 96 with a voxel spacing of 0.513 × 0.513 × 0.8 mm3.
Images after pre-processing and annotations of landmarks and semantic segmentations:
Annotation-aftpreprocess.zip
Reference
Research papers that used this database should cite the following paper:
Zimeng Tan, Jianjiang Feng, Wangsheng Lu, Yin Yin, Guangming Yang, and Jie Zhou, “Multi-Task Global Optimization-Based Method for Vascular Landmark Detection”, Computerized Medical Imaging and Graphics (CMIG), 2024
Contact
Contact Dr. Jianjiang Feng for further information.
