Retinal Imaging as a Biomarker of Ageing

Key Takeaways

Retinal imaging biomarkers use fundus photographs, optical coherence tomography, or derived image features to study age-related structure and vascular state. [1]
Deep-learning retinal age models estimate age from retinal images and compare that estimate with chronological age. [1] [3]
Retinal age gaps have been associated with mortality, stroke, cardiovascular disease, and other outcomes in large cohorts, but these are risk associations rather than direct measurements of ageing rate. [1] [3] [6]
Interpretation depends on image quality, eye disease, vascular risk factors, training population, and the algorithm used. [1] [7]

The retina is increasingly studied as an ageing biomarker because it is an accessible neural and microvascular tissue that can be photographed non-invasively. Retinal images can capture vessel patterns, optic-disc features, macular structure, and neural-layer measurements that change with age and disease burden. [1] [2] [8]

Who This Is Useful For

This page is useful for readers trying to understand what retinal age, eyeAge, RetiAGE, or related retinal-imaging biomarkers can and cannot tell us about biological ageing. It is especially relevant for comparing imaging-based clocks with molecular clocks, vascular biomarkers, and functional measures. [1] [4]

What Retinal Imaging Measures

Retinal fundus photography records the back of the eye, including the retinal vessels, optic disc, and macula. Deep-learning studies have shown that fundus photographs contain enough information to predict chronological age, sex, smoking status, blood pressure, and cardiovascular risk factors above chance. [2]

Optical coherence tomography adds cross-sectional structural information, including retinal nerve fibre layer and optic-nerve-head measurements. These measurements are relevant to ageing research because normal ageing can change retinal neural and non-neural tissue thickness, which must be separated from disease-related change. [8]

Retinal Biomarker Types at a Glance

Biomarker Type	Typical Input	What It Captures	Main Limitation
Retinal age gap	Fundus photographs [3]	Difference between algorithm-estimated retinal age and chronological age [3]	Depends on the training cohort, image processing, and model target [1]
RetiAGE-style scores	Fundus photographs [5]	Probability or score reflecting an older-appearing retina [5]	May not be interchangeable with age-regression retinal clocks [1]
Microvascular traits	Segmented fundus images [7]	Vessel calibre, area, tortuosity, and related vascular features [7]	Strongly affected by blood pressure, diabetes, smoking, and measurement method [7]
OCT structural measures	Optical coherence tomography [8]	Retinal nerve fibre layer, macular layers, and optic-nerve-head structure [8]	Age effects can overlap with glaucoma, neurodegeneration, and ocular disease [8]

Why Researchers Are Interested

Retinal imaging is attractive because it is non-invasive, already common in eye care and large cohorts, and can capture neural and vascular information in the same image. A scoping review found that retinal age research has produced several deep-learning models with moderate to high age-prediction accuracy, while also identifying unresolved questions about generalisability and biological interpretation. [1]

The interest is not limited to eye disease. Retinal images have been used to predict systemic risk factors such as blood pressure and smoking status, which suggests that the image contains information about broader vascular and physiological state. [2] Large retinal-vessel studies also show that age, smoking, blood pressure, obesity, diabetes, and inflammatory markers can relate to vessel traits, reinforcing the view that retinal images mix ageing signal with cardiovascular and metabolic context. [7]

What Retinal Age Models Have Found

In UK Biobank fundus images, a retinal age model achieved a mean absolute error of 3.55 years and found that each one-year increase in retinal age gap was associated with a small increase in all-cause mortality risk after adjustment for multiple covariates. [3]

A separate retinal ageing clock, eyeAge, was trained on EyePACS images and tested in EyePACS and UK Biobank. It predicted chronological age with mean absolute errors of 2.86 and 3.30 years in quality-filtered data, and eyeAge acceleration retained an association with all-cause mortality after adjustment for phenotypic age. [4]

RetiAGE used retinal photographs to estimate the probability of being at least 65 years old and then stratified mortality and major morbidity risk in UK Biobank participants. This supports the idea that retinal images can summarize risk-relevant biology, although RetiAGE is not the same construct as a continuous retinal age gap. [5]

Outcome-specific studies have also linked retinal age gap with incident stroke. In one UK Biobank analysis, participants in the highest retinal age gap quintile had higher stroke risk than those in the lowest quintile after multivariable adjustment. [6]

What the Signal May Represent

Retinal age is probably not one biological pathway. Saliency and review evidence suggest that retinal vessels, the optic disc, and macular regions can contribute to age prediction, so a retinal age score may blend vascular ageing, neural structure, ocular anatomy, and disease-related features. [1] [5]

This mixed signal is scientifically useful but interpretively difficult. A high retinal age gap could reflect systemic vascular risk, ocular disease, image artefact, model bias, or an ageing-related pattern that is not yet mechanistically separated. [1] [7] [8]

Evidence Quality and Interpretation

Confidence is strongest that deep-learning models can predict chronological age from retinal images with useful accuracy in large datasets, and that some retinal age gaps are associated with mortality and vascular outcomes at the cohort level. [1] [3] [4]

Confidence is more moderate that retinal age is a general biological-age measure. The available models differ in architecture, training population, outcome target, and validation setting, and the scoping review noted that standardised development and broader external testing remain important gaps. [1]

Confidence is weaker for using one retinal age score as a standalone individual clinical result. Image quality, ocular pathology, vascular risk factors, population transferability, and model interpretability all affect meaning. [1] [7] [8]

What This Does Not Mean

It does not mean the retina measures whole-body biological age by itself. [1]
It does not mean retinal age, eyeAge, and RetiAGE are interchangeable outputs. [1] [4] [5]
It does not mean a high retinal age gap proves faster ageing is the direct cause of future disease. [3] [6]
It does not mean ocular disease, blood pressure, diabetes, or smoking can be ignored when interpreting retinal biomarkers. [7] [8]

Practical Interpretation Examples

If a model estimates an older retinal age: that means the image resembles older or higher-risk patterns in that model, not that the retina has directly measured a hidden true age. [1] [3]
If retinal age predicts stroke or mortality in a cohort: that supports prognostic association, but it does not prove that the retinal feature is causal. [3] [6]
If two retinal algorithms disagree: they may be using different training labels, image features, populations, and definitions of ageing deviation. [1] [4] [5]

Summary

Retinal imaging is a promising ageing-biomarker domain because it captures accessible neural, microvascular, and structural information in images that are already widely collected. The strongest evidence supports retinal age as a cohort-level imaging marker associated with chronological age and several health outcomes. The main limitation is interpretation: retinal age is a model-derived signal shaped by image quality, ocular and systemic disease, population context, and algorithm design. [1] [3] [4]

References

Grimbly, M. J., et al. (2024). Estimating biological age from retinal imaging: a scoping review. BMJ Open Ophthalmology. https://bmjophth.bmj.com/content/9/1/e001794
Poplin, R., et al. (2018). Prediction of cardiovascular risk factors from retinal fundus photographs via deep learning. Nature Biomedical Engineering. https://www.nature.com/articles/s41551-018-0195-0
Zhu, Z., et al. (2023). Retinal age gap as a predictive biomarker for mortality risk. British Journal of Ophthalmology. https://bjo.bmj.com/content/107/4/547
Ahadi, S., et al. (2023). Longitudinal fundus imaging and its genome-wide association analysis provide evidence for a human retinal aging clock. eLife. https://elifesciences.org/articles/82364
Nusinovici, S., et al. (2022). Retinal photograph-based deep learning predicts biological age, and stratifies morbidity and mortality risk. Age and Ageing. https://doi.org/10.1093/ageing/afac065
Zhu, Z., et al. (2022). Retinal age gap as a predictive biomarker of stroke risk. BMC Medicine. https://bmcmedicine.biomedcentral.com/articles/10.1186/s12916-022-02620-w
O'Neil, A., et al. (2025). Factors Associated With Retinal Vessel Traits in the Canadian Longitudinal Study on Aging. Investigative Ophthalmology & Visual Science. https://pmc.ncbi.nlm.nih.gov/articles/PMC11895846/
Patel, N. B., et al. (2014). Age-Associated Changes in the Retinal Nerve Fiber Layer and Optic Nerve Head. Investigative Ophthalmology & Visual Science. https://pmc.ncbi.nlm.nih.gov/articles/PMC4137486/

Educational Disclaimer

This content is provided for educational purposes only and does not constitute medical advice.