pith. sign in

arxiv: 2602.08580 · v2 · submitted 2026-02-09 · 🧬 q-bio.TO · cs.CV

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

Jose D. Vargas Quiros , Michael J. Beyeler , Sofia Ortin Vela , EyeNED Reading Center , Sven Bergmann , Caroline C.W. Klave , Bart Liefers , VascX Research Consortium This is my paper

Pith reviewed 2026-05-16 03:52 UTC · model grok-4.3

classification 🧬 q-bio.TO cs.CV
keywords retinal vascular biomarkerscolor fundus imagesvessel segmentationgraph-based analysisopen-source toolboxreproducibilityoculomicsartery vein classification
0
0 comments X

The pith

VascX is an open-source Python toolbox that extracts reproducible retinal vascular biomarkers by building graphs from artery-vein segmentations.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces VascX, an open-source Python toolbox that takes vessel segmentation masks from color fundus images and computes biomarkers such as vascular density, central retinal equivalents, and tortuosity. It extracts skeletons, constructs undirected and directed vessel graphs, resolves segments into longer vessels, and supports spatially localized calculations over grids relative to the fovea and optic disc. This graph-based approach enables explainable and modifiable feature extraction that complements segmentation tools. A sympathetic reader would care because the resulting measurements are reproducible across devices and suited for large-scale clinical and epidemiological research in oculomics.

Core claim

VascX starts from vessel segmentation masks, extracts their skeletons, builds undirected and directed vessel graphs, and resolves vessel segments into longer vessels. A comprehensive set of biomarkers is derived, including vascular density, central retinal equivalents (CREs), and tortuosity. Spatially localized biomarkers may be calculated over grids placed relative to the fovea and optic disc. Test-retest reproducibility analysis on repeat imaging of the same eye by different devices shows that most VascX biomarkers have moderate to excellent agreement (ICC > 0.5), with important differences in the level of robustness of different biomarkers. Our analyses of biomarker sensitivity to image

What carries the argument

Vessel graph construction from skeletons followed by heuristic resolution of segments into longer vessels, enabling region-aware biomarker computation.

If this is right

  • Most biomarkers achieve moderate to excellent test-retest agreement (ICC > 0.5) across repeat scans from different devices.
  • Biomarker robustness differs by type, as confirmed by sensitivity to image perturbations and changes in heuristic parameters.
  • The computational efficiency supports scalable deployment across large clinical databases.
  • Open-source availability with documentation lowers barriers for researchers and clinicians to extract and experiment with retinal biomarkers.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Pairing VascX with automated segmentation models could create end-to-end pipelines for retinal vascular analysis without manual intervention.
  • The modifiable code structure allows quick addition of new biomarkers tailored to specific diseases or research questions.
  • Region-aware outputs may reveal localized vascular patterns linked to particular retinal pathologies.

Load-bearing premise

The input artery-vein segmentation masks are accurate enough for downstream graph construction, and the heuristic rules for resolving vessel segments remain robust across different imaging devices, image qualities, and parameter choices.

What would settle it

A test-retest study on images from a new device or lower-quality scans showing ICC values below 0.5 for most biomarkers would demonstrate that the reliability claims do not hold.

read the original abstract

Automatic extraction of retinal vascular biomarkers from color fundus images (CFI) is crucial for large-scale studies of the retinal vasculature. We present VascX, an open-source Python toolbox that extracts biomarkers from CFI artery-vein segmentations. VascX starts from vessel segmentation masks, extracts their skeletons, builds undirected and directed vessel graphs, and resolves vessel segments into longer vessels. A comprehensive set of biomarkers is derived, including vascular density, central retinal equivalents (CREs), and tortuosity. Spatially localized biomarkers may be calculated over grids placed relative to the fovea and optic disc. VascX is released via GitHub and PyPI with comprehensive documentation and examples. Our test-retest reproducibility analysis on repeat imaging of the same eye by different devices shows that most VascX biomarkers have moderate to excellent agreement (ICC > 0.5), with important differences in the level of robustness of different biomarkers. Our analyses of biomarker sensitivity to image perturbations and heuristic parameter values support these differences and further characterize VascX biomarkers. Ultimately, VascX provides an explainable and easily modifiable feature-extraction toolbox that complements segmentation to produce reliable retinal vascular biomarkers. Our graph-based biomarker computation stages support reproducible, region-aware measurements suited for large-scale clinical and epidemiological research. By enabling easy extraction of existing biomarkers and rapid experimentation with new ones, VascX supports oculomics research. Its robustness and computational efficiency facilitate scalable deployment in large databases, while open-source distribution lowers barriers to adoption for ophthalmic researchers and clinicians.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The paper presents VascX, an open-source Python toolbox for extracting retinal vascular biomarkers (vascular density, central retinal equivalents, tortuosity) from color fundus image artery-vein segmentations. It extracts skeletons, constructs undirected and directed vessel graphs, resolves segments via heuristics, and computes biomarkers over fovea/optic-disc grids. The authors report moderate-to-excellent test-retest reproducibility (ICC > 0.5) on repeat imaging across devices, plus sensitivity analyses to image perturbations and heuristic parameters, claiming the toolbox provides explainable, reproducible, region-aware measurements for large-scale oculomics research.

Significance. If the reliability claims hold after addressing validation gaps, VascX would be a useful contribution as an open-source, modifiable complement to segmentation tools, enabling scalable biomarker extraction with graph-based reproducibility and region-aware localization. The release via GitHub/PyPI with documentation supports adoption in clinical and epidemiological studies.

major comments (1)
  1. [Reproducibility and sensitivity analyses] Reproducibility and sensitivity analyses (abstract and associated sections): The claim that graph-based stages produce 'reliable' biomarkers rests on test-retest ICC > 0.5 and perturbation checks, but no controlled experiment perturbs the input artery-vein masks (e.g., boundary erosion, label flips, or realistic segmentation noise) to quantify resulting biomarker drift. This propagation analysis is load-bearing for the reliability assertion, as the toolbox assumes accurate input masks whose errors could affect skeletonization, graph construction, and downstream metrics.
minor comments (2)
  1. [Abstract] Abstract: Specific numerical ICC values, confidence intervals, and exact methods for the test-retest analysis are not reported, only the threshold ICC > 0.5; providing these would strengthen the reproducibility claims.
  2. [Methods] Methods: The heuristic rules for vessel segment resolution and parameter values are described as modifiable, but the paper should include a table or explicit list of default heuristic parameters and their sensitivity ranges for full reproducibility.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the detailed and constructive review. We agree that the current sensitivity analyses focus on image-level perturbations and heuristic parameters but do not directly quantify error propagation from noisy artery-vein input masks. Below we address the major comment and outline the revisions we will make.

read point-by-point responses

  1. Referee: [Reproducibility and sensitivity analyses] Reproducibility and sensitivity analyses (abstract and associated sections): The claim that graph-based stages produce 'reliable' biomarkers rests on test-retest ICC > 0.5 and perturbation checks, but no controlled experiment perturbs the input artery-vein masks (e.g., boundary erosion, label flips, or realistic segmentation noise) to quantify resulting biomarker drift. This propagation analysis is load-bearing for the reliability assertion, as the toolbox assumes accurate input masks whose errors could affect skeletonization, graph construction, and downstream metrics.

    Authors: We acknowledge this is a valid gap. The existing perturbation analyses apply noise to the final images or vary heuristic thresholds, but do not systematically degrade the input artery-vein segmentation masks themselves. In the revised manuscript we will add a dedicated propagation study: we will generate controlled perturbations to the input masks (boundary erosion/dilation at multiple radii, random label flips at 5-15% rates, and simulated realistic segmentation noise derived from inter-observer variability), recompute all biomarkers, and report the resulting ICCs and percentage drifts. These results will be presented in a new subsection of the sensitivity analyses and will directly support (or qualify) the reliability claims for the graph-based stages. revision: yes

Circularity Check

0 steps flagged

No circularity in derivation chain

full rationale

The paper describes a software toolbox that ingests artery-vein segmentation masks, extracts skeletons, constructs graphs, applies heuristic segment resolution, and computes biomarkers such as density, CREs and tortuosity over spatial grids. All reliability statements rest on external empirical evidence (test-retest ICC values across devices and sensitivity analyses to perturbations and parameter choices) rather than any equation, fitted parameter, or self-citation that reduces the output to the input by construction. No load-bearing self-referential definitions, uniqueness theorems, or ansatzes appear in the text.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The toolbox relies on standard computer vision methods for skeletonization and graph construction from vessel masks. Free parameters exist in the heuristics for segment resolution and biomarker calculation, as evidenced by the sensitivity analysis mentioned. No new physical entities are introduced.

free parameters (1)
  • heuristic parameter values
    The abstract explicitly references analyses of biomarker sensitivity to heuristic parameter values, indicating that parameters controlling vessel graph construction and biomarker derivation are set heuristically or tuned to data.
axioms (2)
  • domain assumption Input artery-vein segmentation masks accurately separate arteries from veins
    The entire pipeline begins from these masks; any systematic error in segmentation would propagate to all downstream biomarkers.
  • domain assumption Undirected and directed vessel graphs correctly capture retinal vascular topology
    The toolbox builds these graphs from skeletons and resolves segments into longer vessels, assuming the graph model reflects true anatomical connectivity.

pith-pipeline@v0.9.0 · 5612 in / 1550 out tokens · 78749 ms · 2026-05-16T03:52:39.853978+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

21 extracted references · 21 canonical work pages

  1. [1]

    If conversion factors are available, VascX supports an optional explicit scaling_factor argument, provided per sample

    In practice this means that the ETDRS grid in VascX is defined relative to optic-disc fovea distance rather than fixed distances from the optic disc. If conversion factors are available, VascX supports an optional explicit scaling_factor argument, provided per sample. When provided, this factor will be used for both scaling of the ETDRS grid (overriding t...

    work page 1990

  2. [2]

    The spline smoothing parameter, which controls the smoothness of the splines used to model vessel centerlines, for example, has negligible influence on computation

    showed that most vascx biomarkers exhibit, in most cases, little systematic bias with respect to their optional arguments. The spline smoothing parameter, which controls the smoothness of the splines used to model vessel centerlines, for example, has negligible influence on computation. Some parameters defining computation regions and thresholds, such as ...

    work page 2025

  3. [3]

    Oculomics: A Crusade Against the Four Horsemen of Chronic Disease

    Patterson EJ, Bounds AD, Wagner SK, Kadri-Langford R, Taylor R, Daly D. Oculomics: A Crusade Against the Four Horsemen of Chronic Disease. Ophthalmol Ther. 2024;13(6):1427-1451. doi:10.1007/s40123-024-00942-x

    work page doi:10.1007/s40123-024-00942-x 2024

  4. [4]

    Oculomics: Current concepts and evidence

    Zhu Z, Wang Y, Qi Z, et al. Oculomics: Current concepts and evidence. Prog Retin Eye Res. 2025;106:101350. doi:10.1016/j.preteyeres.2025.101350

    work page doi:10.1016/j.preteyeres.2025.101350 2025

  5. [5]

    Insights into Systemic Disease through Retinal Imaging-Based Oculomics

    Wagner SK, Fu DJ, Faes L, et al. Insights into Systemic Disease through Retinal Imaging-Based Oculomics. Transl Vis Sci Technol. 2020;9(2):6. doi:10.1167/tvst.9.2.6

    work page doi:10.1167/tvst.9.2.6 2020

  6. [6]

    doi:10.1101/2023.07.07.23292368

    work page doi:10.1101/2023.07.07.23292368 2023

  7. [7]

    An Automated Image Segmentation and Useful Feature Extraction Algorithm for Retinal Blood Vessels in Fundus Images

    Abdulsahib AA, Mahmoud MA, Aris H, Gunasekaran SS, Mohammed MA. An Automated Image Segmentation and Useful Feature Extraction Algorithm for Retinal Blood Vessels in Fundus Images. Electronics. 2022;11(9). doi:10.3390/electronics11091295

    work page doi:10.3390/electronics11091295 2022

  8. [8]

    doi:10.1177/11206721211033488

    Retinal vascular geometry detection as a biomarker in diabetes mellitus. doi:10.1177/11206721211033488

    work page doi:10.1177/11206721211033488

  9. [9]

    LUNet: deep learning for the segmentation of arterioles and venules in high resolution fundus images

    Fhima J, Van Eijgen J, Billen Moulin-Romsée MI, others. LUNet: deep learning for the segmentation of arterioles and venules in high resolution fundus images. Physiological Measurement. 2024;45(5):10.1088/1361-6579/ad3d28. doi:10.1088/1361-6579/ad3d28

    work page doi:10.1088/1361-6579/ad3d28 2024

  10. [10]

    AutoMorph: Automated Retinal Vascular Morphology Quantification Via a Deep Learning Pipeline

    Zhou Y, Wagner SK, Chia MA, et al. AutoMorph: Automated Retinal Vascular Morphology Quantification Via a Deep Learning Pipeline. Translational Vision Science & Technology. 2022;11(7):12. doi:10.1167/tvst.11.7.12

    work page doi:10.1167/tvst.11.7.12 2022

  11. [11]

    VascX Models: Deep Ensembles for Retinal Vascular Analysis From Color Fundus Images

    Vargas Quiros J, Liefers B, van Garderen KA, Vermeulen JP, Klaver C. VascX Models: Deep Ensembles for Retinal Vascular Analysis From Color Fundus Images. Trans Vis Sci Tech. 2025;14(7):19. doi:10.1167/tvst.14.7.19

    work page doi:10.1167/tvst.14.7.19 2025

  12. [12]

    Rotterdam artery-vein segmentation (RAV) dataset

    Quiros JV, Liefers B, Garderen K van, Vermeulen J, Center ER, Klaver C. Rotterdam artery-vein segmentation (RAV) dataset. arXiv. Preprint posted online February 18, 2026:arXiv:2512.17322. doi:10.48550/arXiv.2512.17322

    work page doi:10.48550/arxiv.2512.17322 2026

  13. [13]

    PVBM: A Python Vasculature Biomarker Toolbox Based on Retinal Blood Vessel Segmentation

    Fhima J, Eijgen JV, Stalmans I, Men Y, Freiman M, Behar JA. PVBM: A Python Vasculature Biomarker Toolbox Based on Retinal Blood Vessel Segmentation. In: 29 Karlinsky L, Michaeli T, Nishino K, eds. Computer Vision – ECCV 2022 Workshops. Springer Nature Switzerland; 2023:296-312. doi:10.1007/978-3-031-25066-8_15

    work page doi:10.1007/978-3-031-25066-8_15 2022

  14. [14]

    Revised formulas for summarizing retinal vessel diameters

    Knudtson MD, Lee KE, Hubbard LD, Wong TY, Klein R, Klein BEK. Revised formulas for summarizing retinal vessel diameters. Current Eye Research. 2003;27(3):143-149. doi:10.1076/ceyr.27.3.143.16049

    work page doi:10.1076/ceyr.27.3.143.16049 2003

  15. [15]

    retinalysis-fundusprep: A python package for robust color fundus image bounds extraction

    Quiros JV, Liefers B, Garderen K van, Center ER, Klaver C. retinalysis-fundusprep: A python package for robust color fundus image bounds extraction. arXiv. Preprint posted online December 17, 2025:arXiv:2512.16044. doi:10.48550/arXiv.2512.16044

    work page doi:10.48550/arxiv.2512.16044 2025

  16. [16]

    2015;10(9):e0138701

    PLoS One. 2015;10(9):e0138701. doi:10.1371/journal.pone.0138701

    work page doi:10.1371/journal.pone.0138701 2015

  17. [17]

    Deep learning aging marker from retinal images unveils sex-specific clinical and genetic signatures

    Trofimova O, Böttger L, Bors S, et al. Deep learning aging marker from retinal images unveils sex-specific clinical and genetic signatures. medRxiv. Preprint posted online October 5, 2025:2025.07.29.25332359. doi:10.1101/2025.07.29.25332359

    work page doi:10.1101/2025.07.29.25332359 2025

  18. [18]

    Comparing tangible retinal image characteristics with deep learning features reveals their complementarity for gene association and disease prediction

    Beyeler MJ, Trofimova O, Bontempi D, et al. Comparing tangible retinal image characteristics with deep learning features reveals their complementarity for gene association and disease prediction. medRxiv. Preprint posted online January 13, 2026:2024.12.23.24319548. doi:10.1101/2024.12.23.24319548

    work page doi:10.1101/2024.12.23.24319548 2026

  19. [19]

    RetFit: A Novel Deep Learning-Based Biomarker of Cardiorespiratory Fitness Derived From the Retina

    Quintas I, Bontempi D, Bors S, et al. RetFit: A Novel Deep Learning-Based Biomarker of Cardiorespiratory Fitness Derived From the Retina. medRxiv. Preprint posted online January 26, 2026:2025.09.29.25336874. doi:10.1101/2025.09.29.25336874

    work page doi:10.1101/2025.09.29.25336874 2026

  20. [20]

    Artificial Intelligence-Based Uveitis Diagnosis Through Retinal Vasculature Analysis: A Paradigm Shift in Ocular Tuberculosis

    Putera I, Quiros JDV, Rombach SM, Dik WA, van Hagen PM, La Distia Nora R. Artificial Intelligence-Based Uveitis Diagnosis Through Retinal Vasculature Analysis: A Paradigm Shift in Ocular Tuberculosis. Ophthalmol Ther. 2025;14(4):717-732. doi:10.1007/s40123-025-01103-4

    work page doi:10.1007/s40123-025-01103-4 2025

  21. [21]

    Optional biomarker arguments Table

    doi:10.21203/rs.3.rs-8176394/v1 30 Appendix A. Optional biomarker arguments Table

    work page doi:10.21203/rs.3.rs-8176394/v1