Graph-Based Classification of Penetrating Arteries and Veins in Whole-Mouse Vasculature Datasets

Inside the brain, a vast and intricate network of tiny tubes delivers life-sustaining blood and carries away waste. This network is not a random tangle; it is a highly organized system where arteries bring oxygen-rich blood deep into the tissue, and veins carry the used blood back to the surface. For scientists studying how the brain works, or how diseases like stroke or dementia disrupt it, knowing which tube is an artery and which is a vein is essential. Without this distinction, researchers cannot accurately simulate blood flow or understand the specific roles different parts of the network play. However, while modern imaging technology can now capture the entire vascular map of a mouse brain in stunning detail, most of these massive digital maps come without labels. The tubes are visible, but their identity is hidden, leaving a critical gap in our ability to analyze them.

The challenge is that these networks are too large and complex for a human to label by hand. A single mouse brain contains millions of vessel segments, and the surface vessels, which one might try to trace to find the source, are often distorted by the very process of imaging them. To solve this, a team of researchers developed a new way to automatically sort these vessels into arteries and veins using the shape and structure of the network itself, rather than relying on special chemical dyes that are not always available. They focused on the "penetrating trees"—the main branches that dive down from the brain's surface into the deep tissue. By treating these branching structures as mathematical graphs, the team taught a computer to recognize the unique architectural signatures of arteries versus veins.

The researchers trained their computer system using data from two different types of high-resolution microscopes. One set of data came from images where arteries were highlighted with a specific stain, and another came from smaller, manually labeled samples. The computer learned to look for specific patterns: how long the segments between branch points were, how wide the tubes were, and how the branches twisted and turned. They discovered that arteries tend to have longer, straighter segments and larger diameters, acting like efficient highways designed to push blood under pressure to specific areas. In contrast, veins showed a more complex, dense branching pattern, resembling a collection system designed to gather fluid from a wide area at lower pressure.

To test if this method worked, the team applied it to a large, unlabelled dataset of a whole mouse brain that had never been stained for arteries or veins. The computer successfully assigned labels to the vessels, creating a complete map of the brain's blood supply. The system proved remarkably accurate, correctly identifying the vessel type in 96% of cases when tested on independent data. Crucially, the researchers found that they did not even need to measure the width of the vessels to get good results; the way the branches connected and the overall shape of the tree were enough to tell the difference. This means the method can work on data from different types of microscopes and imaging setups without needing to be retrained, making it a versatile tool for future research.

The study also revealed that the differences between arteries and veins are not just about size, but about the fundamental topology, or the arrangement of connections, of the network. Arteries exhibit a structure where a few main branches persist over long distances, while veins have a more uniform distribution of branches that spread out evenly. These structural signatures remained consistent regardless of how the images were taken, suggesting they are a genuine biological feature of how the brain organizes its blood flow. By validating their predictions against known anatomical landmarks, such as the major artery supplying the brain, the team confirmed that their automated labels matched the expected biological reality.

This work provides a powerful new foundation for cerebrovascular research. It allows scientists to take existing, unlabelled datasets from experiments around the world and instantly add the missing artery and vein information. This capability opens the door to large-scale studies of how blood flow changes in different disease models or between different strains of mice, without the need for new, time-consuming experiments. The researchers suggest that while their current system is highly effective, future improvements in image quality and segmentation could make the classification even more precise. Ultimately, this approach turns raw, unlabelled images into rich, informative maps, enabling a deeper understanding of the brain's vascular architecture and its role in health and disease.