Title: retinalysis-vascx: An explainable software toolbox for the extraction of retinal vascular biomarkers
ArXiv ID: 2602.08580
Date: 2026-02-09
Authors: ** 논문에 명시된 저자 정보가 제공되지 않았습니다. **
📝 Abstract
The automatic extraction of retinal vascular biomarkers from color fundus images (CFI) is essential for large-scale studies of the retinal vasculature. We present VascX, an open-source Python toolbox designed for the automated extraction of biomarkers from artery and vein segmentations. The VascX workflow processes vessel segmentation masks into skeletons to build undirected and directed vessel graphs, which are then used to resolve segments into continuous vessels. This architecture enables the calculation of a comprehensive suite of biomarkers, including vascular density, bifurcation angles, central retinal equivalents (CREs), tortuosity, and temporal angles, alongside image quality metrics. A distinguishing feature of VascX is its region awareness; by utilizing the fovea, optic disc, and CFI boundaries as anatomical landmarks, the tool ensures spatially standardized measurements and identifies when specific biomarkers are not computable. Spatially localized biomarkers are calculated over grids relative to these landmarks, facilitating precise clinical analysis. Released via GitHub and PyPI, VascX provides an explainable and modifiable framework that supports reproducible vascular research through integrated visualizations. By enabling the rapid extraction of established biomarkers and the development of new ones, VascX advances the field of oculomics, offering a robust, computationally efficient solution for scalable deployment in large-scale clinical and epidemiological databases.
💡 Deep Analysis
📄 Full Content
Artificial intelligence (AI) has rapidly changed ophthalmic research by enabling the quantitative, automated analysis of imaging at scale. Modern deep learning (DL) applied to color fundus imaging (CFI) has demonstrated the ability to automatically detect ophthalmic disease. In recent years the field of oculomics has taken this a step further by utilizing the eye to understand systemic health [1][2][3] . Non-invasive CFI allows for time and cost-efficient examination of the arteriolar and venular retinal vasculature, also opening the door for large-scale analysis of the vasculature in existing population-based cohorts and clinical databases. Today, substantial evidence links key retinal vascular features to hypertension, kidney disease, stroke subtypes, and overall cardiovascular risk 2 .
Relevant retinal vascular biomarkers include the central retinal arteriolar / venular equivalents (CRAE/CRVE), artery-vein ratio (AVR), bifurcation angles, tortuosity/fractal dimension and bifurcation counts. These features capture changes in the vasculature due to hypertension, hemodynamic status, and network remodeling/efficiency, and have each been linked to target organ damage and incident vascular events 2,4 .
Prior work on the technical challenge of automatically extracting biomarkers from CFIs has laid a strong foundation. PVBM provides a modular Python toolbox for computing a diverse set of vascular biomarkers from pre-segmented vessel maps, including branching angles, endpoints and intersections 5 . AutoMorph delivers a comprehensive end-to-end DL pipeline that includes standardized morphological measurements like caliber, tortuosity, and measures of vascular complexity 6 . Together, these tools have made vascular analysis on fundus image more accessible.
Nevertheless, important technical gaps remain. Most notably PVBM and AutoMorph do not use the location of the fovea for feature computation, preventing the extraction of more localized biomarkers computed over regions or grids defined relative to the optic-discfovea axis. The topological data representation in both tools is limited to an undirected graph, preventing the extraction of topological features that rely on the directedness of the vessel tree graph. Furthermore, both tools lack utilities for easily visualizing the computed features, which limits interpretability.
VascX addresses these needs with a graph-based feature extraction toolbox that utilizes AI disc segmentations and fovea locations for localized feature extraction 7 . The pipeline transforms image-level vessel masks into four different representations sequentially: skeletons, undirected graphs, directed trees with the optic disc as the root, and resolved vessel trees. Biomarkers are then computed from the most appropriate representation stage (e.g., mask-based density, node-based bifurcation counts/angles, segment-based diameter/length/tortuosity, and OD-fovea-aligned spatial features such as CRE and temporal arcade angles). Critically, VascX computes local features using grids and regions defined relative to the fovea and optic disc landmarks, facilitating harmonized reporting and region-specific analyses relevant to clinical translation. The VascX pipeline has already been applied in a multi-cohort study on the phenotypic and genetic characteristics of retinal vascular parameters and their association with diseases 4 .
In this paper, we present a transparent, well-documented, and easily modifiable Python toolbox designed to accelerate methodological innovation and large-scale, reproducible oculomics research. Our contributions are the following:
• We provide an easy-to-use toolbox for analyzing the retinal vasculature. Our toolbox can compute a comprehensive catalog of morphology, topology, caliber, and spatially localized biomarkers, including OD-fovea-aligned and ETDRS-style grid features, from artery-vein-resolved segmentations.
• We present results on the reproducibility of VascX biomarkers across images of the same eye.
• VascX enables rapid experimentation with improved retinal biomarkers via a clear, modular graph-based pipeline and well-documented APIs.
•
The toolbox is open source and easily accessible in Python via PyPI.
Our package was developed entirely in Python using standard data science and image manipulation packages: numpy, opencv, Pillow, scikit-learn, skimage. The package has only open-source dependencies. In this work, we present a first public version of the package. We will follow semantic versioning for future versions.
The VascX feature extraction pipeline operates on image segmentations (usually generated by AI models) and outputs a CSV file with features or biomarkers computed from them. The previous step of model segmentation has been addressed in previous work [6][7][8] . VascX operates on image files individually or in bulk. It accepts all input image formats readable by Pillow.