Reconstructing microvascular network skeletons from 3D images: What is the ground truth?

Area of Science:

• Medical imaging
• Computational biology
• Image processing

Background:

• Microvascular network changes are key disease markers (e.g., Alzheimer's, cancer).
• 3D imaging and computational models analyze these networks for functional simulations like blood flow.
• Extracting 3D networks involves segmentation and skeletonization, but skeletonization lacks robust validation metrics.

Purpose of the Study:

• To address the lack of validation metrics for 3D skeletonization algorithms.
• To introduce a novel metric for assessing skeletonization quality and optimizing algorithms.
• To demonstrate the metric's utility in selecting the best algorithm and guiding improvements.

Main Methods:

• Applied four common skeletonization algorithms to three 3D imaging datasets.
• Introduced a novel super metric evaluating skeleton volume, connectivity, medialness, bifurcation points, and homology.
• Validated the metric's ability to select optimal algorithms, tune parameters, and identify areas for algorithm improvement.

Main Results:

• Significant variability was observed between skeletons generated by different algorithms from the same dataset.
• This structural variability impacted simulated functional metrics, such as blood flow.
• The new super metric effectively compared skeleton quality to original data, enabling algorithm selection and optimization.

Conclusions:

• The developed super metric provides a fast, easy-to-compute solution for validating 3D skeletonization algorithms.
• It aids in selecting the best algorithm for specific datasets and optimizing their parameters.
• This metric is a valuable tool for understanding how structural variations in networks affect biological functions.