VesselBoost: A Python Toolbox for Small Blood Vessel Segmentation in Human Magnetic Resonance Angiography Data

Area of Science:

• Neuroimaging
• Medical image analysis
• Vascular biology

Background:

• Ultra-high magnetic field Magnetic Resonance Angiography (MRA) offers mesoscopic views of the living human brain's arteries.
• Understanding brain vasculature is crucial for insights into blood supply and small vessel diseases.
• Accurate, detailed segmentation of small vessels is needed for quantitative characterization and blood-flow simulations.

Purpose of the Study:

• To explore the application of deep learning for segmenting high-resolution MRA data.
• To address the challenge of limited accurately labeled training data in medical imaging.
• To introduce VesselBoost, a novel package for accurate vasculature segmentation.

Main Methods:

• Utilized deep learning techniques for vessel segmentation on high-resolution MRA datasets.
• Developed VesselBoost, a package designed to handle imperfect training labels.
• Implemented an innovative data augmentation strategy leveraging the inherent similarities in vascular structures.

Main Results:

• Achieved accurate and detailed segmentation of brain vasculature.
• Demonstrated the effectiveness of VesselBoost in segmenting small vessels.
• Showcased the utility of deep learning with imperfect labels for MRA data.

Conclusions:

• Deep learning, particularly with VesselBoost, enables precise segmentation of brain angioarchitecture from MRA.
• The method overcomes limitations of data availability and labeling accuracy.
• Facilitates advanced quantitative analysis and simulation of cerebral blood flow.