Austin Veith (1), Aaron B. Baker (1,2,3,4)
Cold Spring Harbor Laboratory Press, April 2020. DOI: 10.1101/2020.04.06.028555

Keywords

image analysis

Abstract

The quantitative analysis of blood vessel networks is an important component in the understanding and analysis of vascular disease, ischemia, cancer and many other disease states. However, many imaging techniques used for imaging vascular networks are time consuming, prevent further analyses or are technically challenging. Here, we describe a nondestructive technique for imaging the vessels in harvested tissue samples that relatively rapid and effective visualization of in situ vasculature. Furthermore, this technique allows for further analysis of the sample using histochemical staining, immunostaining and other techniques that can be performed on fixed tissue. The method uses ex vivo permeation of the tissue with an iodine solution and micro CT imaging to visualize the vascular network. Using a deep learning algorithm, we trained a convolutional neural network to automatically segment blood vessels from the surrounding tissue for analysis and create a map of the vasculature. While this method cannot achieve the resolution obtainable through destructive techniques, it is a simple and quick method for visualizing and quantifying vascular networks in three dimensions prior to analysis with conventional histological techniques. The global visualization of the three-dimensional vascular network provided by this method gives additional insights into complex changes in the vascular structure and can guide further histological analyses to specific regions of the vascular network. Overall, this method is a simple technique that can be added to most conventional tissue analyses to enhance the quantification and information derived from animal models with vascular network remodeling.

How Our Software Was Used

Dragonfly was used to develop a binary segmentation model to segment the blood vessels of the brain, the heart and the sciatic nerve tissues following a standard U-Net architecture.

Author Affiliation

(1) Department of Biomedical Engineering, University of Texas at Austin, Austin, TX.
(2) Institute for Cellular and Molecular Biology, University of Texas at Austin, Austin, TX.
(3) The Institute for Computational Engineering and Sciences, University of Texas at Austin, Austin, TX.
(4) Institute for Biomaterials, Drug Delivery and Regenerative Medicine, University of Texas at Austin, Austin, TX.