Scalable robust graph and feature extraction for arbitrary vessel networks in large volumetric datasets

Area of Science:

• Biomedical imaging
• Computer vision
• Scientific visualization

Background:

• 3D imaging advances offer detailed insights but challenge automated analysis due to large datasets.
• Current automated vessel network analysis often ignores memory constraints and produces spurious branches.
• Existing methods are frequently limited to tree topologies or specific image modalities.

Purpose of the Study:

• To develop a scalable and robust pipeline for automated vessel network analysis.
• To extract an annotated abstract graph representation from vessel segmentations.
• To overcome limitations of existing methods regarding memory, topology, and image modality.

Main Methods:

• A scalable iterative pipeline is proposed for vessel network analysis.
• The pipeline extracts an annotated abstract graph from foreground segmentations.
• A novel iterative refinement process is controlled by a single, dimensionless parameter.

Main Results:

• The pipeline handles vessel networks of arbitrary topology and shape.
• It is scalable in terms of computational cost, memory, and robustness.
• Analysis of 1 TB volumes on commodity hardware is achieved for the first time.

Conclusions:

• The proposed pipeline offers improved robustness against surface noise, vessel shape deviation, and anisotropic resolution.
• It enables topological analysis of large-scale 3D datasets on standard hardware.
• An implementation is available in the Voreen engine.