VAST: Vascular Flow Analysis and Segmentation for Intracranial 4D Flow MRI

Area of Science:

• Medical Imaging
• Biophysics
• Computational Fluid Dynamics

Background:

• Four-dimensional (4D) Flow MRI offers noninvasive cerebrovascular hemodynamics measurement.
• Clinical underuse stems from manual segmentation and velocity field sensitivity to noise and artifacts.
• Phase aliasing is a significant challenge in 4D Flow MRI data.

Purpose of the Study:

• To present VAST (Vascular Flow Analysis and Segmentation), an automated pipeline for intracranial 4D Flow MRI.
• To couple unsupervised vessel segmentation with physics-informed velocity reconstruction.
• To enhance the clinical utility of 4D Flow MRI for cerebrovascular assessment.

Main Methods:

• VAST uses iterative fusion of magnitude- and phase-based statistics for vessel mask derivation.
• Physics-informed velocity reconstruction includes continuity-constrained phase unwrapping and outlier correction.
• Low-rank denoising reduces noise and aliasing, promoting mass-consistent flow fields.
• Processing is automated and completes within minutes on a standard CPU.

Main Results:

• VAST achieves near quarter-voxel surface accuracy on synthetic data.
• Velocity root-mean-square error is reduced up to fourfold under degraded conditions.
• In vitro segmentation accuracy is within half a voxel of expert annotations.
• In vivo results show close matching with expert masks and reduced divergence residuals (~30%).

Conclusions:

• VAST automates intracranial 4D Flow MRI processing, overcoming limitations of manual methods.
• The pipeline enforces basic flow physics, yielding more self-consistent flow fields.
• VAST facilitates the routine quantitative use of 4D Flow MRI in cerebrovascular assessment.