New brain-vessel atlas maps vascular aging and points to stroke recovery risks (IMAGE)
Caption
Schematic overview of the MARVAL platform. (a) MARVAL dataset construction. 150 TOF-MRA scans (100 healthy participants and 50 stroke patients) were sampled from the PRECISE and CNSR-III cohorts, respectively. Ground-truth labels were generated manually via a two-tier expert consensus process. (b) Automated vascular segmentation. Preprocessing includes brain extraction, denoising, vessel enhancement, and histogram standardization. Vascular segmentations are predicted by a neural network pre-trained on public datasets and fine-tuned on the MARVAL dataset. Predicted segmentations are refined by post-processing. (c) IDP extraction. For each scan, five quantitative feature maps (diameter, segment length, curvature, tortuosity, and torsion) are derived from the vascular segmentation. A normative template is generated by registering individual segmentations to a common space using MultiMorph. (d) Group analysis. Two strategies are implemented: scale-based analysis, which stratifies IDPs by vascular scale hierarchically; and template-based analysis, which registers phenotype maps to the template space. Full details about the pipeline are provided in Method and Supplementary Methods online.
Credit
©Science Bulletin
Usage Restrictions
use with credit.
License
Original content