Segmentation and Modeling of Large-Scale Microvascular Networks: A Survey

Helya Goharbavang, Artem T Ashitkov, Joshua D. Wythe, Guoning Chen, David Mayerich

The figure illustrates the challenge of hollow vessels, a frequent issue observed in Light Sheet microscopy imaging, which can mislead segmentation and skeletonization methods. Different segmentation techniques, including 2D (image-based) and 3D (volume-based) Otsu, Frangi, Beyond Frangi, OOF filters, and U-Net are compared against the ground truth segmentation (top left). The skeletonization results using Lee's method on each segmentation outcome are shown in blue, while the main centerline of the large vessel in the ground truth (bottom left) is highlighted in green, underscoring its deviation from other results. These hollow structures often cause incomplete segmentation, leading to inaccurate or fragmented skeletons, as seen in the generated centerlines.

Abstract

While many algorithms exist for segmenting and skeletonizing vessel-like structures in small-scale images, they are not designed and have not been tested on gigavoxel-scale 3D images. We propose a comprehensive yet compact survey of available algorithms. We focus on essential features for microvascular analysis, including extracting vessel surfaces and the network's associated connectivity. Algorithms were selected based on popularity and availability and provide a thorough quantitative analysis of their performance on data sets acquired using emerging techniques.

Visualization

Segmentation

Ground Centerline

Kerautret's

Slicer3D

Segmentation
Centerline

Resources

Paper: Our paper is submitted for publication.

Code: Our code is available on Github.

Data: Our data (KESM/LSFM/MICRO-CT ) is available upon request.