A photo of the back of the eye, the kind many eye clinics take during a routine exam, may hold clues linked to Alzheimer’s disease risk years before a diagnosis, according to a new study covered on October 311. Researchers used artificial intelligence (AI) to read the photos and estimate 12 factors already known to be tied to Alzheimer’s, and found that the photos of people who later developed the disease looked different, on average 8.55 years before they were diagnosed1. The authors are clear that this is not a way to diagnose Alzheimer’s1.
What happened
The study used data from the UK Biobank, one of the largest health databases in the world1. The team analyzed nearly 63,000 retinal photographs (pictures of the retina, the light-sensing layer at the back of the eye) from more than 44,000 people1.
The researchers chose 12 factors already known to be associated with Alzheimer’s1. Six were yes-or-no or category-type traits: sex, smoking status, sleep quality, economic status, alcohol use and depression1. The other six were measurable numbers, including age, years of education, body mass index, blood pressure and HbA1c, a blood test that reflects average blood sugar1.
They then asked whether deep learning, a type of AI that learns patterns from large numbers of examples, could find features in the eye’s structure that match these risk factors1. A second question was whether the retina, which develops from the same early tissue as the brain, might reflect the paths that lead toward Alzheimer’s1.
The evidence so far
According to the coverage, the AI models performed well across all 12 risk factors1. To see what the AI was paying attention to, the researchers used a method that highlights the parts of an image a model weighs most heavily1. Two areas came up again and again: the optic nerve, the cable that connects the eye to the brain, and the blood vessels in the eye1.
The patterns the AI picked up lined up with real, measurable structural differences in the retina, which the coverage says suggests the models were finding genuine anatomy rather than noise1.
The finding drawing the most attention came from a smaller group: the people in the database who went on to develop Alzheimer’s1. Their retinal patterns differed significantly from those of people who did not develop the disease, and the average gap between the eye photo and the Alzheimer’s diagnosis was 8.55 years1. The researchers suggest the retinal patterns may mirror early brain changes linked to Alzheimer’s1.
What it does not show yet
This is an observational study. It looked back at photos and health records that already existed and searched for patterns. That kind of study can show that certain eye features tend to go along with risk factors and with a later diagnosis. It cannot show that an eye photo can reliably tell one person whether they will get Alzheimer’s.
The study stops short of claiming the eye can diagnose the disease, and the authors say the findings are not diagnostic1. Several useful details are also missing from the coverage available to Biohack: it describes the AI’s performance only as “well,” without accuracy figures, and it does not say how many of the 44,000-plus people went on to develop Alzheimer’s1. Those numbers matter for judging how strong the 8.55-year signal is.
Another open question is what the AI is really detecting. Many of the 12 factors, such as age, blood pressure and blood sugar, affect blood vessels throughout the body. An eye photo that reflects those factors may be picking up general health rather than something specific to Alzheimer’s. The study was also done in one database of UK volunteers; whether the same patterns hold in other groups of people, and with different eye cameras, would need separate testing.
Who is behind it
The coverage links to a paper published through SAGE journals but does not name the authors, their institutions or who paid for the work1. Biohack did not have access to the full paper, so it cannot report funding or possible conflicts of interest. The data came from the UK Biobank, a shared research resource1.
The coverage itself appeared on mindbodygreen, a wellness site that sells supplements, including an eye health product advertised on the same page1.
How it fits the field
The idea that the eye can offer a view of the brain has been building for years. In a 2023 study published in the journal Acta Neuropathologica, researchers examined retina and brain tissue from 86 people after death2. They found higher levels of a form of amyloid beta, a protein that builds up in the brains of people with Alzheimer’s, in the retinas of people who had mild cognitive impairment (early memory and thinking problems) or Alzheimer’s, compared with people whose thinking was normal2. Those retinal changes tracked with the person’s cognitive scores and with the stage of the disease2. The senior author, Maya Koronyo-Hamaoui, said:
These findings may eventually lead to the development of imaging techniques that allow us to diagnose Alzheimer’s disease earlier and more accurately and monitor its progression noninvasively by looking through the eye.
That quote is from the 2023 study coverage2, not from the new UK Biobank work.
At Duke University, researchers built a machine learning model that used a different, more detailed type of eye scan called optical coherence tomography (OCT), together with patient details such as age, sex and education, to tell apart people with normal thinking from people with mild cognitive impairment3. That model correctly flagged 79% of people with mild cognitive impairment and correctly cleared 83% of people with normal cognition3. Duke notes that standard tools for brain diseases, such as brain MRI, PET scans, spinal fluid tests and genetic testing, are expensive and not always easy to get, which is part of the push toward eye imaging3. The Duke team also flagged a practical problem: eye scans can be spoiled by image flaws, more often in people with cognitive impairment, which is why they worked on automatic quality checks3.
The new study differs from these in two ways. It used ordinary color photos rather than specialized scans, and it estimated risk factors in a large general population rather than sorting people who already had memory problems13.
It joins other efforts to spot Alzheimer’s risk long before symptoms. Biohack recently covered a blood test that flagged Alzheimer’s-linked memory decline up to 10 years ahead.
What to watch next
- The full numbers. Accuracy figures for each of the 12 risk factors and the number of people who developed Alzheimer’s will show how much weight the 8.55-year finding can bear.
- Testing in other groups. Whether the same patterns appear in people outside the UK Biobank, and on photos from different cameras.
- Specific to Alzheimer’s or general health. Whether the eye signal adds anything beyond what age, blood pressure and blood sugar already tell doctors.
- Links to brain measures. Studies that compare eye photos with brain scans or blood tests for Alzheimer’s proteins in the same people would help show whether the retina truly mirrors early brain changes, as the researchers suggest1.
For now, the study points to the eye as a possible low-effort window into brain health that is still being mapped, not as a test anyone can take1.
