Chapter 21
Digital Twins of the Retina: A
Computational Framework for Personalised Microvascular Health Assessment
Wahbi K. El-Bouri1,2* and Rmi J.
Hernandez1,2
1Department of Cardiovascular and Metabolic Medicine, Institute of
Life Course and Medical Sciences, University of Liverpool, UK
2Liverpool Centre for Cardiovascular Science, University of Liverpool
& Liverpool Heart and Chest Hospital, UK
Abstract
Digital twins of the retina have the potential to
personalise treatment options for individuals. This study presents a
computational framework for generating digital twins of retinal vasculatures,
simulating haemodynamics and transit times, on an individual basis. Starting
with optical coherence tomography angiography (OCTA) and fundus photographs, we
developed a pipeline to segment the retinal vasculature, virtually recreate it
and run blood flow simulations tailored to each individual.
Digital twins of six eyes were created, and blood flow
was solved through the networks, demonstrating variability between individuals.
Simulations were validated against literature data for blood flow and blood
velocity distributions. Diseases such as diabetic retinopathy and retinal vein
occlusion were simulated for each case, demonstrating variable risk between
individuals.
Our framework enables computation of microvascular haemodynamic
metrics that are challenging to measure directly. In particular, mean transit
times and transit time heterogeneityimportant markers of microvascular health
in other organswere derived from the digital twins. This proof-of-concept
demonstrates the potential of computational frameworks as tools to explore
associations between microvascular ageing, disease, and retinal health. Future
integration with real-time imaging devices could allow live-streamed feedback
and personalised risk prediction, strengthening the translational impact of
this approach.
Keywords: Retinal Digital Twins, Haemodynamic Simulations, Diabetic
Retinopathy, Retinal Biomarkers, Personalised Healthcare
Introduction
The retina is one of the few places in the human body where the
microvasculature can be visualised non-invasively. Retinal health reflects not
only ocular function but also systemic vascular conditions, including diabetes,
hypertension, and neurodegenerative disorders. Early detection and monitoring of microvascular changes can
provide valuable insights for preventing vision loss and assessing overall
cardiovascular risk.
However, current clinical assessments rely largely on qualitative
grading or global averages, overlooking inter-individual variability in
microvascular topology and function.
Digital twins (virtual replicas of biological systems) are gaining
traction across medicine. In cardiology, orthopaedics, and oncology, digital
twins have been used to simulate organ function, predict disease progression,
and optimise therapies,,. Yet their
application in ophthalmology is still in its infancy,,. We hypothesised that creating digital twins of the retina could
capture individual-specific vascular structures and haemodynamic patterns,
providing a foundation for personalised diagnostics and treatments.
This chapter
presents an image-based computational framework for constructing retinal
digital twins from optical coherence tomography angiography (OCTA) and fundus
images. By modelling blood flow and red blood cell movements, we aimed to (i)
demonstrate variability across individuals, (ii) validate simulations against
existing literature, (iii) explore disease conditions such as diabetic
retinopathy, and (iv) investigate microvascular health markers, particularly
mean transit time and heterogeneity.
Methods
Generating the Retinal Digital
Twins
Six individuals were chosen: 4 from the open-source DRIVE dataset and 2 from an in-house dataset. These images were a combination of colour fundus photographs and
optical coherence tomography (OCT), augmented with en-face OCTA. The vessel
networks were segmented from these images, with Voronoi tessellations
representing the three-layered capillary bed connected.
For each
individual, 20 capillary beds were generated due to the inability to visualise
them in the imaging; thus, each individual had 20 twins with the same retinal
morphology, except for the central capillary beds. The segmentation process is
visualised in Figure 1.
Figure 1: Summary
of the framework to create digital twins, including the plasma skimming and
blood flow model.
Simulating Blood Flow
The model
assumes Poiseuille flow:
where 
is the volumetric blood flow, 
and 
the vessel radius and length,
and 
the pressure drop between the
ends of the vessel. The use of an effective viscosity term 
as a function of vessel radius
and discharge haematocrit (ℎ) accounts for the non-Newtonian behaviour of blood and the
Fhrus-Lindqvist effect. A plasma skimming model, to simulate the uneven
distribution of red blood cells at bifurcations in the network, is also
introduced. The blood flow algorithm is shown in Figure 1.
Results
Digital Twins of Health Eyes
Six digital twins of healthy eyes were developed, as shown in Figure 2. Total
retinal blood flow through the eyes was 43.28 12.19 L/min, similar to what
has been measured in vivo in the retina (30 50 L/min). Transit times of red blood cells were also computed, with mean
transit times (MTT) across the 6 twins ranging from 4.8 s to 9.9 s, which
compare well with the measured MTT of 4.9 1.9 s. As such, we
validate the blood flow through the digital twins.
Figure 2: Example
of 6 generated digital twins. The graph of the vasculature is shown next to the
generated vascular graph. Arteries (in red) and veins (in orange) are digitised
from the segmentation. Capillaries (in blue) are randomly generated with
Voronoi tessellation.
Diabetic Retinopathy and in silico Biomarkers
Diabetic
retinopathy affects the blood vessels of the retina. One aspect of this disease
is capillary rarefaction. As such, we implemented a diabetic retinopathy model
to simulate the advancement of diabetic symptoms through random capillary
removal. We simulate an intermediate case by removing 2% of vessels and an
advanced case by removing 5% of vessels. An example of blood pressure maps for
one individual at different stages of the disease is shown in Figure 3.
Discussion
We also sought
to discover whether mean transit time (MTT) and capillary transit time
heterogeneity (CTH) could be used as in silico biomarkers of different
stages of disease. As shown in Figure 4, there is a clear shift in each individual towards longer MTT and
CTH as the disease progresses, suggesting its potential for early detection of
diabetic retinopathy.
Personalised Retinal Healthcare
Our findings
highlight the potential of retinal digital twins to inform personalised
approaches in ophthalmology. By capturing patient-specific vascular structures
and haemodynamics, digital twins go beyond population-based averages and open
avenues for risk stratification tailored to individuals. In particular,
physics-informed simulations of blood flow, and potentially oxygen transport,
can provide further insight into disease impact for individuals and can help
tailor personalised treatment.
Figure 3: Blood
pressure changes with diabetes in one digital twin. From left to right,
capillary rarefaction increases from the baseline conditions to the advanced
stages. Pie charts show the percentage of lost capillary area between healthy
and intermediate diabetic stages (outer ring) and between the healthy and
advanced diabetic stages (inner ring).
Figure 4: Plot of
the probability density functions of transit times with the advancement of
diabetic symptoms. Each line corresponds to a single digital twin. The
vertical dashed line shows the means of MTT and CTH among the six individuals
plotted.
Clinical Implications
MTT and CTH are
established markers of pathology in cerebral and cardiovascular research. We have demonstrated the potential of these markers to detect
disease in silico, potentially extending their use in ophthalmology as
indicators of retinal microvascular health. If validated in larger cohorts,
these metrics could provide early warning signals for diseases such as diabetic
retinopathy or glaucoma, enabling pre-emptive interventions.
Translational Potential
One of the
strengths of this framework is its compatibility with existing imaging
modalities. OCTA and fundus photography are already widely available in
clinical settings. Integrating our pipeline with these devices could enable
near real-time generation of digital twins. This would provide clinicians with
immediate haemodynamic insights, including MTT and CTH biomarkers, alongside
routine imaging.
Real-time Retinal Digital Twins
The proper
definition of an engineering digital twin includes a real-time stream of
information from the real world to the virtual one, which the digital twin can
update. In medicine, due to the fundamental constraints of real-time monitoring
of patients, it has not been feasible to provide that live stream of
information to the digital twin. However, the retina, of all the organs, is
arguably one of the easiest to image, allowing for repeat scans to be taken
with enough frequency to allow for a longitudinal digital twin to be built up.
We are currently working towards this through our study at the St Pauls Eye
Centre, Liverpool, where we are collecting repeat scans of patients over 3
years. This will eventually allow us to build a longitudinal digital twin,
which, if integrated into imaging platforms, would provide instant feedback,
enabling ophthalmologists to explore disease trajectories virtually.
Limitations
This
proof-of-concept is limited by a small sample size and simplified boundary
conditions. Disease models, such as that for diabetic retinopathy, remain
stylised and do not account for biochemical processes such as angiogenesis or
vessel remodelling. Moreover, direct in vivo flow validation was unavailable
for individuals in our dataset (validation was done at the population level).
Larger studies with multimodal validation are required.
Conclusions
We presented a computational framework to generate digital twins of
the retinal vasculature from OCTA and fundus images. The model reproduced
literature-consistent haemodynamics, revealed inter-individual variability, and
simulated disease scenarios with personalised outcomes. Importantly, it enabled
the derivation of perfusion markers such as MTT and CTH, which may serve as new
retinal biomarkers of microvascular health.
Digital twins
represent a promising tool for personalised ophthalmology, bridging imaging
data with computational modelling. Future work will focus on real-time device
integration, scaling to larger cohorts, and linking retinal twins to systemic
models of ageing and disease
Acknowledgements
RJH is funded by
an EPSRC Doctoral Network in AI & Future Digital Health studentship.
References
Asghar, Uzma Saddia, and Chung, Caroline.
November 2025. Application of digital twins for personalized oncology. Nature
Reviews. Cancer 25 (11): 82325. doi:10.1038/s41568-025-00850-7
Bjrnhall, G., O. Mepea, G. O. Sperber, C.
Lindn, E. Mnestam, and A. Alm. December 2002. Analysis of mean retinal
transit time from fluorescein angiography in human eyes: Normal values and
reproducibility. Acta Ophthalmologica Scandinavica 80
(6):65255. https://doi. org/10.1034/j.1600-0420.2002.800617.x
Brown, Emmeline E, Andrew A Guy, Natalie A
Holroyd, Paul W Sweeney, Lucie Gourmet, Hannah Coleman, Claire Walsh, et al.
2024. Physics-Informed Deep Generative Learning for Quantitative Assessment of
the Retina. Nature Communications 15 (1). https://doi.org/10.
1038/s41467-024-50911-y
Chua, Jacqueline, Bingyao Tan, Damon Wong,
Gerhard Garhfer, Xin Wei Liew, Alina Popa-Cherecheanu, Calvin Woon Loong Chin,
Dan Milea, Christopher Li-Hsian Chen, and Leopold Schmetterer. November 2024.
Optical coherence tomography angiography of the retina and choroid in systemic
diseases. Progress in Retinal and Eye Research 103:101292. https:// doi.org/10.1016/j.preteyeres.2024.101292 doi:10.1016/j.preteyeres.2024.101292
Corral-Acero, Jorge, Francesca Margara, Maciej Marciniak,
Cristobal Rodero, Filip Loncaric, Yingjing Feng, Andrew Gilbert, et al. 21
December, 2020. The Digital Twin to enable the vision of precision
cardiology. European Heart Journal 41 (48): 455664. doi:10.1093/eurheartj/ ehaa159
Dean, Michael C, Jacob F Oeding, Pedro Diniz,
Romain Seil, and Kristian Samuelsson, and the ESSKA Artificial Intelligence
Working Group. 10 November, 2024. Leveraging digital twins for improved
orthopaedic evaluation and treatment. Journal of Experimental
Orthopaedics 11 (4): e70084. doi:10.1002/jeo2.70084
Doblhoff-Dier, Veronika, Leopold Schmetterer, Walthard
Vilser, Gerhard Garhfer, Martin Grschl, Rainer A Leitgeb, and Ren M
Werkmeister. 28 January, 2014. Measurement of the total retinal blood flow
using dual beam Fourier-domain Doppler optical coherence tomography with
orthogonal detection planes. Biomedical Optics Express 5 (2):
63042. doi:10.1364/ BOE.5.000630
Hernandez, Rmi J., Wahbi K. El-Bouri, Savita
Madhusudhan, and Yalin Zheng. 2024. AI and the Eye Integrating Deep Learning
and in Silico Simulations to Optimise Diagnosis and Treatment
of Wet Macular Degeneration, February. https://doi.org/10.1101/2024.02.13.23299445
Hernandez, Rmi J, Savita Madhusudhan, Yalin
Zheng, and Wahbi K El-Bouri. 1 April, 2024. Linking Vascular Structure and
Function: Image-Based Virtual Populations of the Retina. Investigative
Ophthalmology & Visual Science 65 (4): 4040. doi:10.1167/iovs.65.4.40
Hernandez, Rmi J, Paul A Roberts, and Wahbi K
El-Bouri. 2023. Advancing Treatment of Retinal Disease through in Silico
Trials. Progress in Biomedical Engineering (Bristol, England) 5
(2): 22002. doi:10.1088/2516-1091/acc8a9
stergaard, Leif. 2020. Blood Flow, Capillary
Transit Times, and Tissue Oxygenation: The Centennial of Capillary
Recruitment. Journal of Applied Physiology 129 (6): 141321. https://doi.org/10.
1152/japplphysiol.00537.2020
Staal, Joes, Michael D. Abrmoff, Meindert
Niemeijer, Max A. Viergever, and Bram van Ginneken. April 2004. Ridge-based
vessel segmentation in color images of the retina. IEEE Transactions
on Medical Imaging 23 (4): 50109. doi:10.1109/TMI.2004.825627