Automatic tracking of vessel-like structures from a single starting point.

Area of Science:

• Medical Image Analysis
• Computational Imaging
• Vascular Biology

Background:

• Tracking complete vascular networks is challenging, often requiring significant computational resources and user input.
• Existing methods primarily focus on single vessel tracking, neglecting complex network structures.

Purpose of the Study:

• To develop an iterative method for tracking entire vascular networks using minimal user interaction.
• To improve the efficiency and accuracy of vascular network segmentation in medical imaging.

Main Methods:

• A novel approach utilizing a cloud of sampling points on concentric spherical layers.
• A proposed vessel model and fitting metric integrated into a min-cost flow problem.
• An optimization scheme for iterative tracking, inherently handling bifurcations and paths.

Main Results:

• Achieved maximum accuracies exceeding 98% on synthetic blood vessel datasets.
• Demonstrated robustness to parameter variations through sensitivity analysis on synthetic data.
• Successfully segmented vascular structures in coronary, carotid, and pulmonary real image datasets, and nerve fiber networks.

Conclusions:

• The method provides accurate and robust vascular network segmentation across diverse datasets.
• Extracted topological information offers potential for computer-aided diagnosis and surgical planning.
• The modular design allows for future problem-specific adaptations and enhancements.