Recent Posts
Connect with:
Tuesday / June 18.
HomemieyecareRetinal Age: A Novel Biomarker of Ageing

Retinal Age: A Novel Biomarker of Ageing

The retina is a window into the body, and with the use of deep learning algorithms it can reveal much about a person’s overall health.

Globally, the number of adults aged 60 years and over is projected to more than double by 2050, increasing from 962 million in 2017 to 2.1 billion.1 The shift in distribution of the population towards older ages inevitably leads to a substantial increase in morbidity and mortality.2 The Australian Institute of Health and Welfare reported that 3.7 million Australians were aged 65 and over in 2016, but accounted for one-third of the health burden across the Australian population.3 These numbers are expected to grow even further, imposing an increasing burden on an already strained healthcare system. Healthy ageing is one of the priorities of the World Health Organisation’s work between 2015 and 2030.

The retina shares similar embryological, anatomic and physiological characteristics with vital organs such as the brain and the heart

Figure 1. Different aging trajectories, when compared to equivalent chronological and biological age in years, with their biological age accelerated or decelerated compared to their chronological age. Figure adapted from Yu et al, 2020 Cancer Research.9


Chronological age refers to the number of years a person has been alive. Ageing is only loosely associated with chronological age, and the rate at which this occurs can range greatly between different people.5 Some 70-year-olds live healthy and energetic lives that are similar to those lived by 30-year-olds. Other 70-year-olds may require extensive support for daily basic activities like walking and eating. This suggests that a person’s chronological age may not fully account for the ageing process.

Biological age is a measurement reflecting functional decline, which results from the accumulation of a wide range of molecular and cellular damage over time.6 Functional decline in physical and mental capacity leads to an increased risk of morbidity and mortality.7,8 The concept of biological age is of significance, as it is a better indicator in quantifying the ageing process and assessing disease risk, and is more reliable and accurate than chronological age.

The relationship between chronological age and biological age characterises different ageing rates. When chronological age is equal to biological age, it represents a healthy ageing rate. Delayed ageing is characterised by a decelerated biological age when compared to chronological age, while faster ageing is characterised by an accelerated biological age when compared to chronological age. Figure 1 illustrates these three ageing trajectories.

In the context of the unprecedented growth rate of the world’s ageing population, there is a clear need for a simple and reliable quantification of the biological age to track the state of biophysiological ageing and implement tailored prevention strategies and interventions.


Extensive efforts have been made to develop methods to determine biological age during the past decades, such as molecular biomarkers, imaging biomarkers, and composite biomarkers. The epigenetic clock, based on DNA methylation status at CpG sites across the genome, is one of the most robust molecular ageing biomarkers with a high correlation with chronological age and a small deviation from chronological age.10,11 Studies have shown that the epigenetic clock is associated with a wide spectrum of ageing phenotypes (e.g., cognition and physical function),12 non-communicable age-related diseases (e.g., cancer),13 and mortality.10 Although quantification of the ageing process at the molecular level is promising and clinically relevant, it requires invasive procedures to collect biosamples and special facilities with appropriate laboratory equipment, which limits its clinical utility.

Recently, brain age based on neuroimaging has been developed and further validated as a reliable and valid aging biomarker with the prominent advantage of being non-invasive.14 The introduction of deep learning (DL), the state-of-the-art approach in image processing, has further improved its accuracy.15,16 The clinical values of brain age have been validated in neurodegenerative diseases, such as dementia17 and Parkinson’s disease (PD).18 However, the cost and accessibility are two major barriers in the wide adoption of brain age.19 

Compared to single biomarkers, composite biomarkers that combine multiple clinical measures are more informative by reflecting diverse aspects of ageing.20 Frailty index and Klemera-Doubal method-biological age are two commonly used composite biomarkers, with great potentials in predicting agerelated outcomes.21,22 Nevertheless, assessing a panel of biomarkers across different systems is complicated and time-consuming, and therefore may not be practical for largescale population screening.


The retina shares similar embryological, anatomical and physiological characteristics with vital organs such as the brain and the heart.23,24 Originating from the diencephalon, the retina is an extension of the central nervous system (CNS) via the optic nerve, which is composed of axons of retinal ganglion cells (RGC).24 RGCs are characterised by the typical features of CNS neurons, with a cell body, dendrites and an axon. In terms of the microvasculature, the retinal microvessels share common features with the vasculature of the brain and the heart. RGCs can be measured using optical coherence tomography (OCT) and retinal microvessels can be visualised and quantified using colour fundus photography. The easily accessible neurons and vessels of the eye are, therefore, a window to study subclinical pathology in the brain and the heart.

Figure 2. Association between retina age gap and mortality risk, allowing for non-linear effects. The reference retinal age gap for this plot (with hazard ratio (HR) fixed as 1.0) was 0 years. Evidence of an overall and non-linear association between retinal age gap and mortality risk was observed (Poverall <0.001; Pnon-linear=0.002). The association between retinal age gaps and mortality is depicted as a J-shaped curve, where positive retinal age gaps were associated with substantially increased risks of mortality. Figure adapted from Zhu et al, 2022 British Journal of Ophthalmology.52

There is growing evidence of the association between retinal changes and neurodegenerative diseases.25 Neurodegeneration in the retina, characterised by the thinning of retinal nerve fibre layer, has been observed in patients with neurodegenerative diseases.26 The retinal vasculature alterations, including venular dilation, arteriolar narrowing, vascular fractal dimension reduction, the loss of vessel density and perfusion density, may be potential biomarkers for early detection of neurodegenerative diseases.27-32 Similarly, recent studies have sought to identify, quantify and link retinal microvascular changes with cardiovascular and cerebrovascular diseases. A growing body of evidence has indicated that retinal vessel calibre (e.g., artery narrowing and venules dilation) and qualitative retinal signs (e.g., microaneurysms and retinal haemorrhage) are associated with cardiovascular and cerebrovascular diseases.33-37 


A growing body of evidence has shown that the retina is a window to the brain and the heart.38-40 Throughout the aging process, the retina is expected to undergo changes in molecular metabolism, cytology and morphology that are similar to changes to the brain and heart.

The neurological and vascular aspects of ageing are particularly important to general health. A reliable and valid retinabased biomarker of ageing may have the potential to identify individuals deviating from a healthy neural and vascular ageing trajectory. Furthermore, the retina is amenable to rapid, non-invasive and costeffective assessments. These features of the retina sparked the interest of our team in exploring the potential of using retinal imaging as an accurate and clinically significant ageing biomarker.

To realise the full potential of retinal imaging in the study of ageing, our interdisciplinary team from Centre for Eye Research Australia, University of Melbourne and University of Monash proposed the concept of retinal age (i.e., biological age based on retinal images). Currently, there is no gold standard to determine biological age. Consistent with previous studies,23,24,41 chronological age was equal to the biological age in relatively healthy individuals. Therefore, the DL model to predict retinal age was trained and validated using over 19,000 retinal images from individuals who were deemed disease free. Results showed that the predicted retinal age, based on the retinal images, correlated closely with the chronological age, with a correlation index of 0.81 (the closer the index is to one, the stronger the positive correlation). Additionally, the predicted retinal age achieved a high accuracy, with a mean average deviation of 3.55 years from the chronological age (also known as mean absolute error [MAE], where a lower value is better). The retinal age achieved excellent accuracy, which outperformed most well-established ageing biomarkers. When comparing the accuracy of other methods that measure biological age using MAE, the epigenetic clock has achieved accuracy of 3.3-5.2 years,42,43 blood measures of 5.5-5.9 years,44,45 transcriptome ageing clock of 6.2-7.8 years,46,47 brain age of 4.3-7.3 years,48,49 and three-dimensional facial age of 2.8-6.4 years.50,51 In addition to better accuracy in the prediction of age, the fast, simple, cost-effective, and user-friendly features makes retinal ageing more accessible, especially in remote and rural areas.


Beyond good performance in age prediction, a clinically relevant ageing biomarker needs to be closely related to the risk of age-related morbidity and mortality. To further explore the clinical value of the retinal age, we estimated the retinal age gap – the difference between the predicted retinal age using retinal images and the chronological age – in a largescale cohort of more than 35,000 middleaged and older adults. A positive retinal age gap indicated an ‘older’ appearing retina compared to the chronological age, while a negative one indicated a ‘younger’ appearing retina compared to the chronological age. Following 11 years of follow-up, participants were linked to the death registry to capture vital status and, if available, date of death. We evaluated the risk of mortality for those who had ‘older’ or ‘younger’ appearing retinas. Our findings showed that for each one-year increase in retinal age gap there was an associated 2% increase in mortality risk. In addition, we found a J-shaped relationship between retinal age gap and the risk of mortality, where the ‘older’ appearing retina was associated with substantially increased risk of mortality (Figure 2).


More Diverse Data 

The current DL model was based on a UK population, which is mainly Caucasian. To improve the generalisability, retinal images from different ethnic populations, from different facilities, and with different field of views, are needed to refine and optimise this model.

Black Box in AI 

Even though we performed attention maps to track the contributions of every pixel from fundus images and found that the areas around retinal vessels were highlighted in the prediction of age, the black box in the DL model may raise concerns in real-world applications.53 Recent efforts have been made to uncover the invisible decision-making process to make AI techniques more transparent and interpretable,54,55 thus providing the opportunity for a future study to specify the mechanism underlying ageing.

Clinical Validation 

Clinical validation of retinal age gap in the prediction of mortality has been confirmed by our multidisciplinary research group. Further clinical validation in specific agerelated diseases, such as cardiovascular diseases and neurodegenerative diseases, will add further value in the application of the retinal age gap in the future.

Influencing Factors 

Investigations on protective and risk factors influencing retinal age, especially modifiable factors, raise the possibility of delaying ageing at an individual level. For example, environmental factors, lifestyle changes, and pharmacological interventions can change the trajectory of ageing.

Clinical Application 

The simple, fast, non-invasive and costeffective features of retinal imaging enable retinal age to be an accessible screening tool for the identification of people at high risk of accelerated ageing. The recent development of smartphone-based retinal cameras has further corroborated potentials for large-scale population-based screening. For those fast agers with ‘older’ appearing retinas, individualised and tailored interventions could be proposed, such as lifestyle changes (e.g., smoking cessation, physical activity, healthy sleep, and dietary changes), and intensive management of chronic diseases (e.g., blood glucose control, and weight control).


We are the first group to propose the concept of retinal age as a novel biomarker of ageing. The accuracy and clinical utility to predict the risk of mortality for retinal age has been verified. There is more to explore in this field, such as refinement and validation of the current DL model, identification of factors influencing retinal age, uncovering the black box of the DL model, and further investigation into the clinical validation of specific age-related diseases.

In the context of population ageing globally, the concept of retinal age may help to achieve accurate risk stratification and assist personalised prevention and interventions, ultimately leading to tremendous public health benefits.

Dr Lisa Zhuoting Zhu is a research fellow in the Ophthalmic Epidemiology Department at the Centre for Eye Research Australia. Her work has focused on how images of the eye can be used to view a person’s overall health. She has presented at national and international conferences, published over 45 manuscripts and attracted more than 220 citations. Her research has been supported by a number of grants, including a NHMRC Investigator Grant, and been awarded multiple patents. 

Dr Alice Ruiye Chen is a PhD student in the Ophthalmic Epidemiology Department at the Centre for Eye Research Australia. Her PhD project focuses on the concept of retinal age. She is a certified medical practitioner in China with a background in the field of medical imaging. She is a recipient of the Melbourne Research Scholarship from the University of Melbourne. 

Professor Mingguang He is the Professor of Ophthalmic Epidemiology at the University of Melbourne. He undertook his medical training in China and holds a Master of Public Health degree from Johns Hopkins University in Baltimore, and a PhD in ophthalmology at Moorfields Eye Hospital in London. 

Professor He is a global expert in vision-related clinical and epidemiologic research. He has led some important epidemiological studies and clinical trials, including the first population-based study on myopia in China, the first population-based study on glaucoma in China, a clinical trial to prove the efficacy of increased outdoor time on myopia prevention published in JAMA (2015) and a prophylactic clinical trial on angle closure glaucoma published in Lancet (2019). His publications have attracted more than 13,000 citations. 


  1. United Nations, Department of Economic and Social Affairs, Population Division (2017). World Population Ageing 2017 – Highlights (ST/ESA/SER.A/397). 
  2. Cheng X, Yang Y, Schwebel DC, et al. Population ageing and mortality during 1990-2017: A global decomposition analysis. PLoS Med 2020; 17(6): e1003138. 
  3. Australia’s health 2018; https://www.aihw.gov.au/ reports/australias-health/australias-health-2018/contents/ table-of-contents. 
  4. Rudnicka E, Napierała P, Podfigurna A, Męczekalski B, Smolarczyk R, Grymowicz M. The World Health Organization (WHO) approach to healthy ageing. Maturitas 2020; 139: 6-11. 
  5. Lowsky DJ, Olshansky SJ, Bhattacharya J, Goldman DP. Heterogeneity in healthy aging. J Gerontol A Biol Sci Med Sci 2014; 69(6): 640-9. 
  6. López-Otín C, Blasco MA, Partridge L, Serrano M, Kroemer G. The hallmarks of aging. Cell 2013; 153(6): 1194-217. 
  7. Calderón-Larrañaga A, Vetrano DL, Ferrucci L, et al. Multimorbidity and functional impairment-bidirectional interplay, synergistic effects and common pathways. J Intern Med 2019; 285(3): 255-71. 
  8. Lee Y. The predictive value of self assessed general, physical, and mental health on functional decline and mortality in older adults. J Epidemiol Community Health 2000; 54(2): 123-9. 
  9. Yu M, Hazelton WD, Luebeck GE, Grady WM. Epigenetic Aging: More Than Just a Clock When It Comes to Cancer. Cancer Res 2020; 80(3): 367-74. 
  10. Marioni RE, Harris SE, Shah S, et al. The epigenetic clock and telomere length are independently associated with chronological age and mortality. Int J Epidemiol 2018; 45(2): 424-32. 
  11. Horvath S, Raj K. DNA methylation-based biomarkers and the epigenetic clock theory of ageing. Nat Rev Genet 2018; 19(6): 371-84. 
  12. Marioni RE, Shah S, McRae AF, et al. The epigenetic clock is correlated with physical and cognitive fitness in the Lothian Birth Cohort 1936. Int J Epidemiol 2015; 44(4): 1388-96. 
  13. Horvath S. DNA methylation age of human tissues and cell types. Genome Biol 2013; 14(10): R115. 
  14. Cole JH, Franke K. Predicting Age Using Neuroimaging: Innovative Brain Ageing Biomarkers. Trends Neurosci 2017; 40(12): 681-90. 
  15. Kuo CY, Tai TM, Lee PL, et al. Improving Individual Brain Age Prediction Using an Ensemble Deep Learning Framework. Front Psychiatry 2021; 12: 626677. 
  16. Cole JH, Poudel RPK, Tsagkrasoulis D, et al. Predicting brain age with deep learning from raw imaging data results in a reliable and heritable biomarker. Neuroimage 2017; 163: 115-24. 
  17. Beheshti I, Maikusa N, Matsuda H. The association between “Brain-Age Score” (BAS) and traditional neuropsychological screening tools in Alzheimer’s disease. Brain Behav 2018; 8(8): e01020. 
  18. Beheshti I, Mishra S, Sone D, Khanna P, Matsuda H. T1- weighted MRI-driven Brain Age Estimation in Alzheimer’s Disease and Parkinson’s Disease. Aging Dis 2020; 11(3): 618-28. 
  19. Lystad RP, Pollard H. Functional neuroimaging: a brief overview and feasibility for use in chiropractic research. J Can Chiropr Assoc 2009; 53(1): 59-72. 
  20. Shamir L. Composite Aging Markers Can Be Used for Quantitative Profiling of Aging. Gerontology 2015; 62(1): 66-8. 
  21. Howlett SE, Rockwood MR, Mitnitski A, Rockwood Standard laboratory tests to identify older adults at increased risk of death. BMC Med 2014; 12: 171.
  22. Klemera P, Doubal S. A new approach to the concept and computation of biological age. Mech Ageing Dev 2006; 127(3): 240-8. 
  23. Flammer J, Konieczka K, Bruno RM, Virdis A, Flammer AJ, Taddei S. The eye and the heart. Eur Heart J 2013; 34(17): 1270-8. 
  24. London A, Benhar I, Schwartz M. The retina as a window to the brain-from eye research to CNS disorders. Nat Rev Neurol 2013; 9(1): 44-53. 
  25. Ptito M, Bleau M, Bouskila J. The Retina: A Window into the Brain. Cells 2021; 10(12). 
  26. La Morgia C, Ross-Cisneros FN, Sadun AA, Carelli V. Retinal Ganglion Cells and Circadian Rhythms in Alzheimer’s Disease, Parkinson’s Disease, and Beyond. Front Neurol 2017; 8: 162. 
  27. Gatto NM, Varma R, Torres M, et al. Retinal microvascular abnormalities and cognitive function in Latino adults in Los Angeles. Ophthalmic Epidemiol 2012; 19(3): 127-36. 
  28. Liew G, Mitchell P, Wong TY, et al. Retinal microvascular signs and cognitive impairment. J Am Geriatr Soc 2009; 57(10): 1892-6. 29. Cheung CY, Ong S, Ikram MK, et al. Retinal vascular fractal dimension is associated with cognitive dysfunction. J Stroke Cerebrovasc Dis 2014; 23(1): 43-50. 
  29. Czakó C, Kovács T, Ungvari Z, et al. Retinal biomarkers for Alzheimer’s disease and vascular cognitive impairment and dementia (VCID): implication for early diagnosis and prognosis. Geroscience 2020; 42(6): 1499-525. 
  30. Kromer R, Buhmann C, Hidding U, et al. Evaluation of Retinal Vessel Morphology in Patients with Parkinson’s Disease Using Optical Coherence Tomography. PLoS One 2016; 11(8): e0161136. 
  31. Cabrera DeBuc D, Somfai GM, Koller A. Retinal microvascular network alterations: potential biomarkers of cerebrovascular and neural diseases. Am J Physiol Heart Circ Physiol 2017; 312(2): H201-h12. 
  32. Gopinath B, Chiha J, Plant AJ, et al. Associations between retinal microvascular structure and the severity and extent of coronary artery disease. Atherosclerosis 2014; 236(1): 25-30. 
  33. Chandra A, Seidelmann SB, Claggett BL, et al. The association of retinal vessel calibres with heart failure and long-term alterations in cardiac structure and function: the Atherosclerosis Risk in Communities (ARIC) Study. Eur J Heart Fail 2019; 21(10): 1207-15. 
  34. Wong TY, Hubbard LD, Klein R, et al. Retinal microvascular abnormalities and blood pressure in older people: the Cardiovascular Health Study. Br J Ophthalmol 2002; 86(9): 1007-13. 
  35. Wong TY, Klein R, Couper DJ, et al. Retinal microvascular abnormalities and incident stroke: the Atherosclerosis Risk in Communities Study. Lancet 2001; 358(9288): 1134-40. 
  36. Klein R, Klein BE, Jensen SC, Moss SE, Meuer SM. Retinal emboli and stroke: the Beaver Dam Eye Study. Arch Ophthalmol 1999; 117(8): 1063-8. 
  37. Liew G, Wang JJ. [Retinal vascular signs: a window to the heart?]. Rev Esp Cardiol 2011; 64(6): 515-21. 
  38. Chiquita S, Rodrigues-Neves AC, Baptista FI, et al. The Retina as a Window or Mirror of the Brain Changes Detected in Alzheimer’s Disease: Critical Aspects to Unravel. Mol Neurobiol 2019; 56(8): 5416-35. 
  39. Schlecht A, Vallon M, Wagner N, Ergün S, Braunger BM. TGFβ-Neurotrophin Interactions in Heart, Retina, and Brain. Biomolecules 2021; 11(9). 
  40. Wong CW, Wong TY, Cheng CY, Sabanayagam C. Kidney and eye diseases: common risk factors, etiological mechanisms, and pathways. Kidney Int 2014; 85(6): 1290-302. 
  41. Hannum G, Guinney J, Zhao L, et al. Genome-wide methylation profiles reveal quantitative views of human aging rates. Mol Cell 2013; 49(2): 359-67. 
  42. Weidner CI, Lin Q, Koch CM, et al. Aging of blood can be tracked by DNA methylation changes at just three CpG sites. Genome Biol 2014; 15(2): R24. 
  43. Putin E, Mamoshina P, Aliper A, et al. Deep biomarkers of human aging: Application of deep neural networks to biomarker development. Aging (Albany NY) 2016; 8(5): 1021-33. 
  44. Mamoshina P, Kochetov K, Putin E, et al. Population Specific Biomarkers of Human Aging: A Big Data Study Using South Korean, Canadian, and Eastern European Patient Populations. J Gerontol A Biol Sci Med Sci 2018; 73(11): 1482-90. 
  45. Peters MJ, Joehanes R, Pilling LC, et al. The transcriptional landscape of age in human peripheral blood. Nat Commun 2015; 6: 8570. 
  46. Fleischer JG, Schulte R, Tsai HH, et al. Predicting age from the transcriptome of human dermal fibroblasts. Genome Biol 2018; 19(1): 221. 
  47. Liem F, Varoquaux G, Kynast J, et al. Predicting brainage from multimodal imaging data captures cognitive impairment. Neuroimage 2017; 148: 179-88. 
  48. Cole JH, Ritchie SJ, Bastin ME, et al. Brain age predicts mortality. Mol Psychiatry 2018; 23(5): 1385-92. 
  49. Xia X, Chen X, Wu G, et al. Three-dimensional facialimage analysis to predict heterogeneity of the human ageing rate and the impact of lifestyle. Nat Metab 2020; 2(9): 946-57. 
  50. Chen W, Qian W, Wu G, et al. Three-dimensional human facial morphologies as robust aging markers. Cell Res 2015; 25(5): 574-87. 
  51. Zhu Z, Shi D, Guankai P, et al. Retinal age gap as a predictive biomarker for mortality risk. Br J Ophthalmol 2022. 
  52. The Lancet Respiratory M. Opening the black box of machine learning. Lancet Respir Med 2018; 6(11): 801. 
  53. Zhang Z, Beck MW, Winkler DA, Huang B, Sibanda W, Goyal H. Opening the black box of neural networks: methods for interpreting neural network models in clinical applications. Ann Transl Med 2018; 6(11): 216. 
  54. Poon AIF, Sung JJY. Opening the black box of AIMedicine. J Gastroenterol Hepatol 2021; 36(3): 581-4.