Your next eye exam may already capture something your doctor is not looking for. šļø A retinal scan called optical coherence tomography takes 5 to 10 minutes, and a growing pile of studies links thinner retinal nerve layers to dementia diagnosed a decade or more later. We think that is one of the more interesting findings in early detection, and one of the easiest to oversell. š§
Here is our read. The signal is real in large datasets, the effect is modest for any one person, and we found no FDA-cleared retinal test for predicting dementia. Below, we cover what the scan measures and why the eye tells you anything about the brain, what the long-follow-up studies find, where the evidence gets shaky, and what to do with it today. Quick question before we start: do you know when you last had a dilated eye exam?
Why an eye scan says anything about your brain
The logic starts with anatomy. Because the retina and brain originate from the same neural tube, the eyes are often regarded as extensions of the brain, and the retina is the only optically accessible nervous tissue. Optical coherence tomography (OCT) is a non-invasive, rapid imaging method already widely available in ophthalmology clinics, and a scan takes only 5 to 10 minutes. It images the retina in cross-section, so researchers can measure individual layers, including the retinal nerve fiber layer (RNFL) and the ganglion cell-inner plexiform layer (GC-IPL). š¬
Two mechanisms are proposed, and they are not mutually exclusive. The first is neurodegeneration. In an autopsy study in Acta Neuropathologica, researchers analyzed donated retina and brain tissue from 86 people and found more beta-amyloid in the retinas of people with Alzheimerās or early cognitive decline, while microglia, the cells that clear debris, dropped by about 80% in people with cognitive problems. The second is vascular. A retinal-age score from Singapore researchers associates closely with brain imaging markers of cerebral small vessel disease and brain atrophy. So thinning may reflect lost nerve cells, damaged small vessels, or both. š§¬
Four things researchers measure:
OCT layer thickness: RNFL and GC-IPL, the workhorse measures in the big cohorts
OCT angiography: vessel density in the macula, with mixed results across studies
Retinal photographs plus AI: ordinary fundus photos scored by deep learning, such as a āretinal ageā
Hyperspectral imaging: light at many wavelengths, searching for an amyloid signature, still experimental
What the long-follow-up studies find
Start with the largest clean signal. A 2025 analysis in Alzheimerās & Dementia follows 35,433 UK Biobank participants for a median of 12.49 years, during which 392 people (1.11%) develop dementia. People in the lowest macular RNFL thickness quintile have a 64% higher risk of dementia than those in the highest (HR 1.64, 95% CI 1.17 to 2.30), and each 5-micrometer decrease goes with a 15% higher risk. APOE Īµ4 accounts for only 7.6% of that association, so thinning is not just a stand-in for the best-known risk gene. This is observational human data, and it excludes people with eye disease and baseline dementia. š§
A pooled analysis of four cohorts finds a smaller effect. Lower baseline macular RNFL thickness goes with 10% and 11% higher incidence of all-cause dementia and Alzheimerās disease, while GC-IPL thickness is not significantly associated. An earlier UK Biobank analysis of 30,573 people (mean age about 55) finds greater GC-IPL thickness goes with lower Alzheimerās risk (HR 0.968, 95% CI 0.938 to 0.999) over about nine years, with only 148 Alzheimerās cases. That interval barely clears 1.0. š
Does it show up earlier still? The Dunedin Study looked at midlife. In 938 people assessed at age 45, narrower arterioles and wider venules go with higher scores on four established dementia-risk indexes. Note the outcome: risk scores, not dementia.
Then there is AI on ordinary photos. A 2025 study in Alzheimerās & Dementia scores a deep-learning āretinal age,ā trained to mimic the PhenoAge blood clock we cover in our guide to aging blood tests. In 510 Singapore memory-clinic patients, older retinal age predicts cognitive decline (SHR 1.34) and dementia (SHR 1.43), and in 33,495 UK Biobank participants it predicts incident dementia over 12 years (SHR 1.25). The authors declare no conflicts of interest.
The numbers side by side:
HR 1.64: lowest versus highest RNFL quintile, 35,433 people, about 12.5 years
10% to 11% higher incidence: lower RNFL across four pooled cohorts
SHR 1.25: older retinal age and incident dementia, 33,495 people, 12 years
HR 0.968: GC-IPL thickness and Alzheimerās risk, 30,573 people, 148 cases
Why it is a risk flag and not a diagnosis
Look back at how few people got dementia in the biggest study. With 392 cases among 35,433 people, most participants in the thinnest quintile never developed dementia during follow-up. A 64% higher relative risk on a small baseline is still a small absolute risk. š¬
Specificity is the next problem. A systematic review of OCT findings says the value of retinal biomarkers for early detection remains uncertain because of variability and limited disease specificity. A separate review of OCT and brain MRI adds that retinal findings can occur in both dementia and non-neurocognitive conditions, which makes them hard to apply across a broad population.
The amyloid-imaging work is earlier still. In a Swedish memory-clinic study of 57 patients (35 amyloid-positive, 22 amyloid-negative), the best hyperspectral model reaches a mean AUC of 0.77, accuracy of 0.66, sensitivity of 0.73 and specificity of 0.55. A specificity of 0.55 means roughly 45% of amyloid-negative patients get flagged anyway. Several authors are affiliated with Mantis Photonics AB, a company, an industry link we want you to see. RetiSpecās larger Bio-Hermes-002 study is an ongoing multi-site effort planning about 1,200 participants, with 387 enrolled at the time of reporting and full performance data still forthcoming, and its authors are RetiSpec employees. š§¬
Two more cautions. In the Dunedin cohort of 865 people, lower childhood IQ associates with thinner RNFL at age 45, alongside a greater decline in processing speed, so some of the thinning may be a lifelong trait rather than a sign of decline. And the University of Floridaās June 2026 study uses an AI model trained on more than 40,000 patients to predict blood pressure, smoking, alcohol use and insomnia from retinal photos. Those are risk factors, not dementia itself.
Why we would not act on a single scan:
Low absolute risk: most people with thin layers never develop dementia in the studied window
Overlap with other conditions: thinning is not unique to Alzheimerās
Modest hyperspectral accuracy: small samples, industry-linked authors
Possible lifelong trait: part of the signal may predate any decline
If a scan told you your dementia risk was a quarter above average, what would you change on Monday?
Where regulators and blood tests fit
Regulators have moved on blood tests, not eye scans. The Alzheimerās Association reports that the FDAās August 24, 2026 clearance of Rocheās Elecsys pTau217 is the fourth blood-based biomarker test for Alzheimerās the agency has cleared. The sequence: Fujirebioās Lumipulse in May 2025, Rocheās Elecsys pTau181 in October 2025, C2Nās PrecivityAD2 in August 2026, then Elecsys pTau217. The pTau217 test, developed with Eli Lilly, is cleared for people 55 and older with signs, symptoms or complaints of cognitive decline, and it is not a stand-alone diagnostic, and an intermediate result needs further testing. š
So blood tests answer do you have amyloid pathology, given symptoms, while retinal tests try to answer who should we worry about earlier. Those are different jobs. As far as we can find, retinal amyloid tools are still research-stage. A National Library of Medicine overview says Optina Diagnosticsā retinal test has FDA Breakthrough Device designation but is available for research, not commercial use, and RetiSpecās ocular test is also early-stage and not commercially available. Every company, trial, and funding event connected to this story is tracked and searchable in LongevityHub Pro. š¬
Where each tool stands:
Amyloid blood tests: FDA-cleared, for symptomatic adults
Retinal OCT: routine for eye disease, not cleared to predict dementia (as far as we can find)
Hyperspectral amyloid imaging: research use, with Breakthrough Device designation for one company
Retinal-photo AI: validated in cohort studies, not a clinical product
What to do with this now
Nothing here requires you to buy a new test. It does suggest a few sensible moves. š§
First, keep your routine eye exams. The UF team notes that even regular eye exams for prescription glasses can capture retinal photos. If you already have OCT or retinal images from an eye doctor, ask for a copy. We think a baseline you can compare against later is worth more than a single reading, though no guideline tells you to do this. Researchers have also floated the idea that continuous OCT monitoring holds promise for predicting cognitive trajectories.
Second, treat retinal risk as a prompt to fix what you can. The UF model picks up blood pressure, smoking, alcohol and insomnia, all modifiable. Our piece on brain habits covers a 2025 GeroScience meta-analysis of 39 cohort studies linking insomnia to a 36% higher dementia risk, and our list of medical screenings that can add years covers checks worth scheduling.
Third, if you are 55 or older and notice memory symptoms, ask about the blood tests. Retinal screening does not replace that conversation.
What to do this month:
Book a dilated eye exam if you are overdue, and ask whether OCT is part of it
Request your images so you have a baseline
Check blood pressure and sleep, the retinal-linked risk factors you can change
Raise memory symptoms early with a clinician if you are 55 or older
Which of those four would you start with, and what is stopping you?