Title: retinalysis-vascx: An explainable software toolbox for the extraction of retinal vascular biomarkers

Abstract:

The automatic extraction of retinal vascular biomarkers from color fundus images (CFI) is a critical component in conducting large-scale investigations into the retinal vasculature. To address this need, we introduce VascX, an open-source Python-based toolbox designed to derive biomarkers directly from artery-vein segmentations of CFIs. The VascX pipeline begins with vessel segmentation masks to extract skeletons, constructs both undirected and directed vessel graphs, and consolidates individual vessel segments into longer, continuous vessels. From this structure, a wide array of biomarkers is generated, encompassing vascular density, central retinal equivalents (CREs), and tortuosity. Additionally, the tool allows for the calculation of spatially localized biomarkers by applying grids positioned relative to the optic disc and fovea.

VascX is accessible through GitHub and PyPI, accompanied by thorough documentation and illustrative examples. A test-retest reproducibility assessment, utilizing repeat imaging of the same eyes across various devices, indicates that the majority of VascX biomarkers demonstrate moderate to excellent agreement (ICC > 0.5), though significant variations exist regarding the robustness of specific biomarkers. Further analysis of biomarker sensitivity to image perturbations and heuristic parameter values elucidates these discrepancies and provides deeper characterization of the metrics.

Ultimately, VascX serves as an explainable and readily modifiable feature-extraction framework that works in tandem with segmentation to yield dependable retinal vascular biomarkers. Its graph-based computation stages facilitate reproducible, region-specific measurements ideal for epidemiological and clinical studies at scale. By streamlining the extraction of established biomarkers and accelerating the experimentation with novel ones, VascX advances oculomics research. The tool’s computational efficiency and robustness enable scalable deployment within large databases, while its open-source nature reduces adoption hurdles for clinicians and ophthalmic researchers.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC