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

We present CaravelMetrics, a computational framework for automated cerebrovascular analysis that models vessel morphology through skeletonization-derived graph representations. The framework integrates atlas-based regional parcellation, centerline extraction, and graph construction to compute fifteen morphometric, topological, fractal, and geometric features. The features can be estimated globally from the complete vascular network or regionally within arterial territories, enabling multiscale characterization of cerebrovascular organization. Applied to 570 3D TOF-MRA scans from the IXI dataset (ages 20-86), CaravelMetrics yields reproducible vessel graphs capturing age- and sex-related variations and education-associated increases in vascular complexity, consistent with findings reported in the literature. The framework provides a scalable and fully automated approach for quantitative cerebrovascular feature extraction, supporting normative modeling and population-level studies of vascular health and aging.

💡 Research Summary

CaravelMetrics is a fully automated computational framework designed to extract quantitative cerebrovascular features from binary vessel masks derived from any 3‑D neurovascular imaging modality (e.g., TOF‑MRA, CTA). The pipeline begins with a 26‑connectivity skeletonization of the mask, converting each voxel of the one‑voxel‑thick skeleton into a graph node and linking neighboring voxels with weighted edges based on geodesic versus Euclidean distance. Two post‑processing steps improve graph fidelity: (1) edges whose geodesic‑to‑Euclidean distance ratio exceeds 2 are discarded, eliminating low‑intensity spurious connections; (2) orphan voxels not captured by the skeleton are re‑attached to the nearest skeleton point if they satisfy a distance (>5 mm) and branching‑angle (≤30°) constraint, thereby maximizing vascular coverage. The resulting graph may consist of several disconnected components; for each component three root nodes are defined (the two largest‑diameter end‑points and the largest‑diameter bifurcation, or the largest end‑point if no bifurcation exists). Shortest‑path extraction from each root to all reachable end‑points yields vessel segments that serve as the basis for feature computation.

Fifteen vascular descriptors are calculated, grouped into four categories: (i) morphometric – total vessel length and vascular volume; (ii) topological – bifurcation count, bifurcation density, loop number, number of connected components, and proportion of abnormal‑degree nodes; (iii) fractal – box‑counting fractal dimension and lacunarity, quantifying self‑similarity and spatial heterogeneity; (iv) geometric – geodesic length, curvature, mean curvature, average tortuosity, curvature variability, and arc‑over‑chord ratio, all derived from cubic‑spline parameterisation of centreline segments. Importantly, each metric can be aggregated globally across the whole‑brain graph or locally within 30 arterial territories defined by a publicly available MNI‑aligned arterial atlas. This dual‑scale capability enables whole‑brain as well as territory‑specific analyses.

Implementation relies on Python 3.10 with NumPy, SciPy, scikit‑image, NetworkX, and vedo. After skeletonisation, a two‑iteration Laplacian smoothing (α = 0.8) and removal of components with fewer than seven nodes further reduce noise. Atlas registration follows a multi‑step protocol: brain extraction (FSL BET), reorientation to MNI space, atlas‑to‑T1 registration using FLIRT with mutual information, and finally T1‑to‑TOF‑MRA alignment.

The framework was evaluated on 570 healthy participants (age 20–86) from the IXI dataset, using deep‑learning‑generated vessel masks from the VesselVerse repository. Age‑related trends were examined with Spearman correlations and quartile‑based ANOVA, while sex, body‑mass index (BMI), and education effects were assessed with t‑tests or ANOVA as appropriate. All tests were corrected for multiple comparisons using Benjamini‑Hochberg FDR (q < 0.05).

Key findings include: (1) Strong negative correlations between age and total vessel length (r ≈ ‑0.55), bifurcation count (r ≈ ‑0.55), and fractal dimension (r ≈ ‑0.55), indicating a substantial reduction in network extent and complexity across the lifespan; total length decreased by roughly 20 % from young adulthood to old age. (2) Positive age correlations for curvature (r ≈ +0.20) and lacunarity (r ≈ +0.50), suggesting that the remaining vasculature becomes more tortuous and spatially heterogeneous with aging. (3) Sex differences revealed higher fractal dimension and bifurcation density in females, implying a denser, more intricate vascular architecture. (4) Higher educational attainment was associated with longer vessels, higher fractal dimension, and lower lacunarity, supporting the notion that cognitive reserve may be reflected in cerebrovascular organization. (5) BMI showed a dose‑response effect: normal‑weight individuals possessed longer, more branched networks with lower curvature and lacunarity compared with overweight and obese participants, highlighting the vascular impact of adiposity. Regional analyses demonstrated that age effects varied across arterial territories, with cortical regions showing stronger declines in length and fractal dimension, whereas deep nuclei exhibited opposite patterns.

The authors emphasize that CaravelMetrics offers three major advantages: modality‑agnostic graph construction, a comprehensive suite of 15 reproducible vascular metrics, and scalability to large cohorts. Limitations include dependence on the quality of the input binary mask and potential loss of fine vessels during skeletonisation. Future work should test robustness across heterogeneous acquisition protocols, extend validation to pathological populations, and explore the predictive value of the extracted metrics for clinical outcomes.

In summary, CaravelMetrics provides an open‑source, fully automated solution for large‑scale cerebrovascular quantification, enabling researchers to model normative vascular aging, investigate demographic influences, and lay the groundwork for biomarker discovery in neurovascular disease.