A geometric flow for segmenting vasculature in proton-density weighted MRI

Area of Science:

• Neurosurgery
• Medical Image Analysis
• Computational Geometry

Background:

• Modern neurosurgery relies on magnetic resonance imaging (MRI) for surgical planning and guidance.
• Dual echo MRI acquisitions provide proton-density (PD) and T2-weighted images, useful for evaluating peritumoral edema.
• Segmenting vasculature from PD images is clinically valuable due to widespread acquisition and non-invasive nature.

Purpose of the Study:

• To develop a novel geometric flow method for segmenting vasculature in PD MRI images.
• To enable vasculature segmentation from PD data, potentially eliminating the need for additional contrast-based sequences.
• To validate the developed method qualitatively and quantitatively on various MRI datasets.

Main Methods:

• Utilized Frangi's vesselness measure to identify putative vessel centerlines and radii.
• Developed a vector field from the vesselness measure to guide a flux maximizing flow algorithm.
• Applied the geometric flow algorithm to segment vessel boundaries from PD, MR angiography, and Gadolinium-enhanced MRI data.

Main Results:

• Qualitative validation performed on PD, MR angiography, and Gadolinium-enhanced MRI volumes.
• Quantitative validation on a single-subject dataset showed high recovery of vasculature compared to ground truth.
• Multi-subject validation on 19 PD datasets from a digital brain phantom also demonstrated significant vasculature recovery (>=90%).

Conclusions:

• The novel geometric flow method effectively segments vasculature from PD MRI images.
• The approach shows high accuracy and potential to streamline preoperative imaging by reducing the need for supplementary sequences.
• This method offers a new visualization technique for segmentations using masked projections.