Source-linked AI summary
Towards a glaucoma risk index based on simulated hemodynamics from fundus images
José Ignacio Orlando, João Barbosa Breda, Karel van Keer, Matthew B. Blaschko, Pablo J. Blanco, Carlos A. Bulant
TL;DR
Glaucoma-related ocular hemodynamic changes have been observed, but effects on retinal arteriolar function have not been studied. This paper simulates flow in patient-specific arterial trees from fundus images and represents the outcomes as fixed-length features; the classification experiment reports an AUC different from random for identifying glaucomatous patients.
Problem
Although glaucoma-related ocular hemodynamic changes have been observed, effects on retinal arteriolar function have not been studied.
Source: §1 · p. 2Method
The paper simulates steady, incompressible, non-Newtonian blood flow in patient-specific arterial networks extracted from fundus images and encodes outcomes as fixed-length features.
Source: §1 · p. 2Results
Takeaways & Limitations
The pilot study reports that glaucomatous patients exhibit hemodynamic-variable changes relative to controls, and releases its database and code.
Source: §1 · p. 2, §4 · p. 7The simulation scenarios use preset values rather than patient-specific parameters, including outlet-flow assumptions based on Murray’s law.
Source: §2.2 · p. 4
Abstract
from arXiv · showhide
Glaucoma is the leading cause of irreversible but preventable blindness in the world. Its major treatable risk factor is the intra-ocular pressure, although other biomarkers are being explored to improve the understanding of the pathophysiology of the disease. It has been recently observed that glaucoma induces changes in the ocular hemodynamics. However, its effects on the functional behavior of the retinal arterioles have not been studied yet. In this paper we propose a first approach for characterizing those changes using computational hemodynamics. The retinal blood flow is simulated using a 0D model for a steady, incompressible non Newtonian fluid in rigid domains. The simulation is performed on patient-specific arterial trees extracted from fundus images. We also propose a novel feature representation technique to comprise the outcomes of the simulation stage into a fixed length feature vector that can be used for classification studies. Our experiments on a new database of fundus images show that our approach is able to capture representative changes in the hemodynamics of glaucomatous patients. Code and data are publicly available in https://ignaciorlando.github.io.
1 Introduction
Although glaucoma-related ocular hemodynamic changes have been observed, analysis of retinal arterioles accessible through fundus photography has been limited. This study proposes simulation-based characterization using patient-specific arterial networks, a fixed-length representation of simulation outcomes, and a pilot database.
1 Introduction
- Retinal arteriolar and venular hemodynamics have received limited analysis despite their accessibility in fundus photography.Prior ocular hemodynamic measurements have focused on structures such as the central retinal artery, ophthalmic artery, and retinal capillary network.Source: §1 · pp. 1–2
- The study proposes patient-specific retinal microvasculature simulations to characterize hemodynamics in glaucomatous and control subjects.The arterial networks are extracted from fundus images and modeled as steady, incompressible, non-Newtonian flow in rigid domains.Source: §1 · p. 2
- A novel bag-of-words-inspired technique summarizes simulation outcomes in fixed-length feature vectors for classification.The representation is intended to support linear classifiers examining the relationship between features and disease presence.Source: §1 · p. 2
- The study presents a pilot finding of hemodynamic-variable changes in glaucomatous patients relative to controls.The contribution is framed as a first pilot study.Source: §1 · p. 2
- The authors release a new fundus-image database together with code and implementation details.The database is called LES-AV.Source: §1 · p. 2
2 Materials and Methods
The workflow extracts patient-specific arterial trees from fundus images, simulates retinal hemodynamics, and converts the resulting measurements into fixed-length vectors for classification. Simulations use a steady 0D model with prescribed inlet pressures and outlet flows, while the feature representation summarizes arterial segments, bifurcations, and terminals.
2 Materials and Methods
- The LES-AV database contains 22 color fundus images, mostly at 30° field of view and 1444×1620 pixels.One image has 45° field of view and 1958×2196 pixels.Source: §2 · p. 2
2.1 Preprocessing
- Arterial trees are extracted from fundus images through semiautomatic segmentation, manual refinement and artery-vein labeling, then skeletonized into graphs with estimated lumen radii.Vessels within the optic disc are removed, and each connected component is assigned a root nearest the disc center.Source: §2.1 · p. 3
2.2 Simulation of the retinal hemodynamics
- A steady 0D model computes pressure and flow in rigid, non-Newtonian vessels using radius-dependent viscosity and Poiseuille resistance along segments.At bifurcations, the model applies mass conservation and pressure continuity.Source: §2.2 · p. 4
- The simulations prescribe mean normotensive inlet pressure and outlet flows based on Murray’s law, with three preset total-flow scenarios representing mean and extreme healthy-subject values.The scenarios are applied to each subject without patient-specific parameters.Source: §2.2 · p. 4
2.3 Bag of Hemodynamic Features (BoHF)
- The Bag of Hemodynamic Features maps variable-sized sets of segment, bifurcation, and terminal summaries into fixed-length vectors using class-specific k-means codewords and nearest-code counts.The resulting vectors can be used for binary classification, such as with ℓ2-regularized logistic regression.Source: §2.3 · p. 5
3 Results
Simulations produced hemodynamic values consistent with prior human retinal-circulation studies, and BoHF classification distinguished glaucomatous from normal subjects above chance. Per-measurement analyses found group differences in pressure drop and velocity, while patient-level comparisons found no significant differences in age or number of segments.
3 Results
- Simulated mean pressure drop and blood velocity were consistent with previous studies of human retinal circulation.Source: §3 · p. 6
- AUC was 0.70248 and accuracy was 68.18% for BoHF classification of glaucomatous versus normal subjects using SC2 simulations and leave-one-out cross-validation.The study used ℓ2-regularized logistic regression on LES-AV.Source: §3 · p. 6
- At the per-patient level, age and number of segments did not differ significantly between groups, and diagnosis was independent of sex.Table 1 summarizes the per-patient comparisons; none of its variables differed significantly between groups.Source: §3 · p. 6, Table 1 · p. 7
- On a per-measurement basis, mean pressure drop and velocity were larger in the healthy group, while age was larger in the glaucomatous group; retinal radius was lower in the glaucomatous group but not significantly so.Age correlations with pressure drop and velocity were near zero and not significant.Source: §3 · pp. 6–7
4 Discussion
The study presents a first approach to characterize glaucoma-related retinal hemodynamics through computational fluid dynamics, with simulated arterial findings aligned with prior observations. Its pilot classification identified an AUC different from random, and the authors released the code and LES-AV database.
4 Discussion
- The study is presented as the first examination of the relationship between glaucoma and simulated hemodynamic outcomes in retinal arterioles.Source: §4 · p. 7
- The pilot study used SC1 for glaucomatous subjects and SC3 for controls, producing lower simulated velocity in glaucoma and physiologically correct retinal pressure drop.These choices followed prior observations of smaller arterial radius in glaucoma and lower in-vivo CRA velocity in affected patients.Source: §4 · p. 7
- BoHF classification using unbiased SC2 outcomes yielded an AUC different from random, indicating that the tool preserved disease-associated arterial hemodynamic variations.Source: §4 · p. 7
- The authors publicly released the MATLAB/C++/Python code and LES-AV, a dataset described as the first to provide arterio-venous segmentations, diagnostics, and clinical parameters at image level.Source: §4 · p. 7