April 13, 2023 | Retinal scans offer a fascinating window into the health of the body’s circulatory system and brain functioning as well as identify eye diseases that aren't visibly noticeable. Now, a new study suggests that this artificial intelligence-based technology is also good for predicting how well people are aging, according to co-author Pankaj Kapahi, Ph.D., professor at the Buck Institute for Research on Aging.
Working collaboratively with Google Research and Google Health, Kapahi demonstrated how a retinal aging clock known as “eyeAge” could more accurately track aging than other aging clocks and validated that its genetic basis includes ALKAL2, previously shown to extend lifespan in Drosophila. As covered in a research article published in eLife (https://doi.org/10.7554/eLife.82364), it is anticipated to be a tool for evaluating the effectiveness of anti-aging interventions and eye aging.
The imaging target of eyeAge is the fundus, the blood vessel-rich tissue in the retina, says Kapahi. A genome-wide association analysis conducted on more than 64,000 patient samples from the UK Biobank identified the genes influencing this accelerated aging phenotype, which builds on previous Buck research uncovering a connection between diet, eye health, and lifespan in fruit flies.
A more accurate positive prediction ratio was achieved for two consecutive visits by individuals compared to two random time-matched ones. This suggests that eyeAge predictions may be informative for tracking aging patterns longitudinally, Kapahi says.
It is now possible to determine the trajectory of aging with 71% accuracy based on changes in the eyes of people being treated with gero-protective therapeutics, Kapahi says. This reflects the proportion of patient visits correctly ordered in the study by a deep learning model trained on “hundreds of thousands” of fundus images of the retina at different ages, which were obtained from the web-based diabetic retinopathy screening system EyePACS.
This is not the kind of assessment that is possible for a human to make, he notes. And researchers still don’t know how machine learning algorithms recognize what they see, but Google researchers who trained and tuned the model have previously used it to identify more than 50 eye diseases as well as to predict cardiovascular risk.
In the future Kapahi expects that eyeAge can be used to measure response to anti-aging interventions, including GLYLO, a supplement marketed by Juvify Health—which he founded—and backed by positive research in mice studies from his laboratory. The glycation-lowering cocktail is designed to reduce the toxic effects of sugar which is known to accelerate aging of various organs including the eye, he says.
Interestingly, epigenetic clocks derived from DNA methylation—one of the most widely used tests for quantifying biological age—have no correlation with eyeAge. This suggests that the aging process “can’t be boiled down to one biomarker,” says Kapahi. Rather, multiple systems are aging and therefore multiple, complementary tests are needed to figure out what’s happening inside the body.
Retinal scans are commonly done, primarily to detect eye conditions such as diabetic retinopathy, glaucoma, macular degeneration, and detached retina. The retina also provides a wealth of information about the vascular system that can be directly and noninvasively visualized and monitored repeatedly over time, Kapahi says, noting that calcifications become visible in blood vessels at the back of the eye with advancing age.
Artificial intelligence-based technology already exists to predict people’s age based on a photograph of their face, he continues. Google is instead using image analysis to pick up on data from retinal images that are not discernable to the naked eye. The advance here is that it is now being done to track aging itself rather than specific, age-associated diseases.
The enabling technology, as is customary with Google, is all open source. How to analyze the resulting data is likely of greater import to the company, Kapahi says, including how machine learning algorithms arrive at their conclusions.
There is no shortage of aging clocks and many of them can be utilized easily enough at home since measures of muscle strength, breathing capacity, and blood pressure all track with age, he adds. Glucose levels, and thus insulin sensitivity, also decline with age.
Properly motivated individuals, trying to move the needle on their health, need some way to monitor their progress. For example, intermittent fasting is believed to slow down age-related visual decline, as Buck researchers showed only last year in a study in Drosophila melanogaster (Nature Communications, DOI: 10.1038/s41467-022-30975-4) where the fruit flies additionally gained some lifespan. A key finding was that fasting enhances the genes important for circadian rhythmicity, improving vision, which—along with eating healthy and reducing noise pollution—is “something that is practical to do in people,” says Kapahi.
The changing demography of society means more people are living into their 80s and 90s and many interventions are being developed to help keep them as healthy as possible. Mobility and eyesight are two of the most common, age-related concerns people are bringing to their doctor, Kapahi says, citing a potential role for eyeAge in the clinic alongside multiple other biomarker tests.
Changes to the eye can be seen as early as the 30s in some patients, especially those who are diabetic or otherwise in poor health, and “they could do something about it,” he continues. They could also be particularly sensitive to diabetic retinopathy—over time, more than half of people with the disease will develop the complication—and eyeAge could establish its trajectory.
The early diagnosis and progression of multiple diseases can happen via the eye, notes Kapahi. As Buck researchers have learned, genes important to cardiovascular risk, Alzheimer’s disease, and even cancer can influence eye function.
Kapahi and his colleagues are currently investigating the genetics behind eye aging and hope to use eyeAge in a clinical trial of the efficacy of GLYLO, which can be purchased on Amazon. The retinal aging clock opens the door to non-invasively measuring the impact of other anti-aging interventions based on lifestyle change. The retinal aging clock could also be used in clinical trials of eye-related drugs requiring regulatory approval.
Google has an automated retinal disease assessment tool that has shown promise in detecting diabetic retinopathy as well as cardiovascular risk factors, at least using existing tabletop cameras in clinics, he adds. The next step is to see if photos taken by smartphones can help detect diseases from external eye photos.