💻 computer science

An Automated Framework for Large-Scale Graph-Based Cerebrovascular Analysis

This paper introduces CaravelMetrics, a fully automated framework that models cerebrovascular morphology as skeleton-derived graphs to extract multiscale morphometric and topological features from 3D TOF-MRA scans, successfully demonstrating reproducible detection of age-, sex-, and education-related vascular variations in a large-scale population study.

Original authors: Daniele Falcetta, Liane S. Canas, Lorenzo Suppa, Matteo Pentassuglia, Jon Cleary, Marc Modat, Sébastien Ourselin, Maria A. Zuluaga

Published 2026-05-20

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Daniele Falcetta, Liane S. Canas, Lorenzo Suppa, Matteo Pentassuglia, Jon Cleary, Marc Modat, Sébastien Ourselin, Maria A. Zuluaga

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine your brain's blood vessels as a massive, intricate city of roads. Some are wide highways, some are tiny alleyways, and they all connect in a complex web to deliver fuel (blood) to every neighborhood (brain region). As we age, this city changes: roads get shorter, some intersections disappear, and the remaining paths might get curvier or more twisted.

The paper introduces CaravelMetrics, a new "city planner's toolkit" designed to automatically map and measure this vascular city without a human needing to trace every single road by hand.

Here is how the framework works and what it found, explained simply:

1. The Toolkit: Turning a Picture into a Map

Instead of just looking at a blurry photo of blood vessels, CaravelMetrics does three main things:

Skeletonizing: Imagine taking a thick, 3D model of a tree and stripping away all the leaves and bark until you are left with just the bare branches. This creates a "skeleton" of the blood vessels.
Graphing: It turns that skeleton into a digital map made of dots (nodes) and lines (edges), similar to a subway map.
Measuring: It calculates 15 different "stats" about this map, such as:
- Morphometric: How long are the roads? How much total space do they take up?
- Topological: How many intersections (bifurcations) are there? Are there any loops?
- Fractal: How complex and self-similar is the pattern? (Think of how a small branch looks like a tiny version of the whole tree).
- Geometric: How curvy or twisted are the roads?

2. The Experiment: A Study of 570 People

The researchers tested this toolkit on 570 healthy people (ages 20 to 86) using brain scans. They wanted to see if their automated tool could spot the natural changes that happen as we get older, and if those changes were different for men, women, or people with different lifestyles.

The "Scanner" Problem:
First, they noticed that the type of MRI machine used mattered a lot. It was like taking photos of the same city with three different cameras; the pictures looked slightly different just because of the camera, not the city. They had to filter out the data from one specific hospital to get a clear picture of the actual biological changes.

3. What They Discovered

Once they cleaned up the data, the toolkit revealed some clear patterns about how our "vascular city" ages:

The City Shrinks and Simplifies: As people got older, the total length of the blood vessels decreased by about 20%. The number of intersections (branching points) also dropped. The complex web became simpler and less dense.
The Roads Get Curvier: While the network got smaller, the remaining roads became more twisted and winding (higher "tortuosity").
Gaps Appear: The space between the vessels became more uneven (higher "lacunarity"), meaning the network became spottier.

4. Lifestyle and Demographics

The study also found that these vascular changes aren't just about age; they are linked to other factors:

Body Size: Taller people and those with a lower Body Mass Index (BMI) tended to have longer, more complex vessel networks. Heavier individuals had simpler, shorter networks.
Gender: After accounting for height, women tended to have slightly more complex and compact vascular networks than men.
Education: This was a fascinating finding. People with higher levels of education had blood vessel networks that were longer and more complex. The researchers suggest this might be related to "cognitive reserve"—the idea that a more educated brain maintains a richer vascular structure, similar to how a well-maintained city keeps its roads in better shape.

Summary

In short, CaravelMetrics is an automated robot that can look at a brain scan, turn the blood vessels into a digital map, and instantly tell you if that map looks like a healthy young city or an aging one. It confirmed that as we age, our brain's road network gets shorter, simpler, and curvier, but it also showed that factors like education and body size play a role in how well that network is maintained.

The authors emphasize that this tool is fully automatic and can be used on different types of brain scans, making it a powerful way to study vascular health on a large scale without needing a human to draw every line.

Technical Summary: CaravelMetrics Framework for Cerebrovascular Analysis

Problem Statement
Cerebrovascular morphology undergoes significant changes throughout the lifespan, including declining vessel density, increased curvature, and reorganized branching patterns. Understanding these changes is critical for distinguishing pathological conditions from normal aging. However, large-scale population studies are currently hindered by a reliance on manual vessel tracing and a lack of fully automated tools capable of extracting complex biomarkers across diverse neurovascular imaging modalities. Existing approaches are often limited to simple volumetric measurements or are restricted to specific image types and semi-automatic workflows, failing to capture the full topological and geometric complexity of the vascular tree.

Methodology
The authors present CaravelMetrics, a fully automated computational framework designed to extract fifteen quantitative vascular features from binary vessel masks. The framework is modality-agnostic, relying on a graph-based representation of the vascular network rather than specific image intensities.

Graph Construction:
- Skeletonization: A binary vascular mask ( $M$ ) is converted into a one-voxel-thick skeleton ( $S$ ) that preserves topological properties (connectivity and loops).
- Graph Generation: Skeleton points become nodes ( $V$ ) in an undirected weighted graph ( $G$ ). Edges ( $E$ ) connect nodes within a 26-neighborhood.
- Post-Processing: Spurious edges are removed based on geodesic vs. Euclidean distance thresholds. Orphan points (voxels in the mask but not the skeleton) are integrated via centroid connections under branching-angle and distance constraints.
- Segment Definition: Three root nodes are defined based on vessel diameter and bifurcation status to ensure representative coverage. Vessel segments are defined as the shortest paths between these roots and reachable endpoints.
Feature Extraction:
The framework computes 15 features across four categories (Table 1 in the paper):
- Morphometric: Total vessel length and volume.
- Topological: Bifurcation count, bifurcation density, number of loops, loop lengths, abnormal-degree nodes, and connected components.
- Fractal: Fractal dimension (via box-counting) and lacunarity (spatial heterogeneity).
- Geometric: Geodesic length, instantaneous and mean curvature, mean squared curvature, and arc-over-chord ratio (tortuosity).
Regional Analysis:
An arterial atlas partitions the brain into 30 territories. The framework supports both whole-brain analysis and region-specific analysis by intersecting the vascular mask with atlas-defined territories, allowing for multiscale characterization.
Implementation:
The pipeline is implemented in Python using libraries such as scikit-image, NetworkX, and vedo. Registration involves multi-step alignment (brain extraction, MNI reorientation, and co-registration) to align the atlas with TOF-MRA data.

Key Contributions

Automated Framework: CaravelMetrics provides a scalable, fully automated pipeline for extracting complex vascular features without manual intervention.
Modality Agnosticism: By operating on binary masks and graph structures, the tool is designed to work with any 3D neurovascular imaging modality.
Multiscale Characterization: The integration of an arterial atlas enables simultaneous global (whole-brain) and regional (territory-specific) analysis.
Comprehensive Feature Set: The extraction of 15 distinct features captures structural, topological, fractal, and geometric properties, offering a more complete characterization than simple volume or diameter metrics.

Experimental Results
The framework was applied to 570 healthy subjects (ages 20–86) from the IXI dataset, utilizing TOF-MRA scans and deep learning-generated vessel masks from the VesselVerse dataset.

Site Effects: Initial analysis revealed significant scanner-induced biases (particularly at the IOP site), necessitating the exclusion of this site to analyze 488 subjects from two homogeneous centers.
Age-Related Changes:
- Declines: Significant negative correlations were found between age and total vessel length ( $r \approx -0.55$ ), bifurcation count, fractal dimension, and volume. Total vessel length showed a ~20% reduction across the lifespan.
- Increases: Positive correlations were observed with age for curvature ( $r \approx +0.20$ ) and lacunarity ( $r \approx +0.50$ ), indicating increased tortuosity and sparsity.
- Regional Variability: Age effects varied across cortical territories, with deep nuclear regions showing different patterns compared to cortical areas.
Demographic Associations:
- BMI: Higher BMI was associated with structural simplification (reduced bifurcations and nodes) and increased curvature/lacunarity. Normal-weight individuals had longer vessels than overweight or obese groups.
- Sex: Post-bias removal, females exhibited higher fractal dimension and bifurcation density than males, suggesting more compact and complex networks.
- Education: Higher education levels were associated with increased vessel length, higher fractal dimension, and reduced lacunarity, suggesting a more complex and spatially homogeneous vascular organization.

Significance and Claims
The paper claims that CaravelMetrics successfully yields reproducible vessel graphs that capture known biological variations related to aging, sex, and education. The framework demonstrates that vascular complexity decreases and tortuosity increases with age, consistent with existing literature. Furthermore, it identifies associations between education and vascular complexity, potentially linking to cognitive reserve.

The authors position CaravelMetrics as a tool for normative modeling and population-level studies of vascular health. They emphasize that the framework supports the discrimination between healthy aging and pathological changes by providing demographic-specific reference ranges. The study concludes that the tool is effective for quantitative feature extraction but acknowledges limitations regarding sensitivity to mask quality and the need for future validation across pathological cohorts and diverse imaging protocols.