TaG-Net: Topology-Aware Graph Network for Centerline-Based Vessel Labeling

Area of Science:

• Medical Imaging
• Computer Vision
• Neuroscience

Background:

• Accurate anatomical labeling of head and neck vessels is crucial for diagnosing cerebrovascular diseases.
• Challenges in automated labeling include vessel tortuosity, branching complexity, and proximity to other vasculature in computed tomography angiography (CTA).

Purpose of the Study:

• To develop a novel topology-aware graph network (TaG-Net) for accurate and automated anatomical labeling of head and neck vessels in CTA.
• To combine volumetric segmentation and centerline labeling approaches for enhanced vessel labeling.

Main Methods:

• Extraction of vessel centerlines and construction of a vascular graph.
• Application of TaG-Net incorporating topology-preserving sampling, feature grouping, and multi-scale graph representation for vessel labeling.
• Utilizing the labeled vascular graph to refine volumetric segmentation through vessel completion.

Main Results:

• The proposed TaG-Net method demonstrated superior performance in vessel segmentation and labeling compared to state-of-the-art techniques.
• Experiments conducted on CTA images from 401 subjects validated the method's effectiveness.
• The system successfully labeled 18 segments of head and neck vessels.

Conclusions:

• TaG-Net effectively addresses the challenges of automated vessel labeling in CTA by integrating voxel and line space information.
• The method offers a significant advancement in anatomical labeling for cerebrovascular disease diagnosis.
• This approach shows promise for improving the accuracy and efficiency of medical image analysis in neurovascular applications.