Why is Alzheimer's so hard to diagnose?
Alzheimer's disease slowly damages the brain, causing memory problems and later dementia, meaning a loss of independence due to declining thinking skills. According to the authors, there is no simple screening test: diagnosis usually involves expensive and sometimes invasive tests that are rarely available outside specialist centers (a spinal tap or a PET brain scan, for example).
The retina, at the back of the eye, is an extension of the nervous system. It can be photographed painlessly, without needles, using a fundus camera found in many eye clinics and diabetes eye screening programs. That is why researchers looked for signs of Alzheimer's in these photos.
How was the AI built and tested?
The researchers pooled data from 11 studies in different countries that had enrolled people with Alzheimer's dementia and people without the disease. For each person, the deep learning AI (a program that learns from examples) looked at four photos: for each eye, one centered on the optic nerve and one on the macula, the part of the retina used for sharp vision.
The model was trained and checked on 6 studies, then tested on 5 others it had never seen. Three of those test studies had PET scans showing whether the brain had a heavy load of amyloid, a protein that builds up in Alzheimer's disease.
What were the results?
In internal validation, the AI caught 93.2% of people with the disease (sensitivity) and had an area under the curve of 0.93, a score where 0.5 means chance and 1 means a perfect test. In the outside test groups, accuracy ranged from 79.6% to 92.1%, with an area under the curve of 0.73 to 0.91.
In the groups with PET scans, it could also tell apart people with and without significant amyloid, with accuracy from 80.6% to 89.3%. Notably, it did better in people who also had an eye disease (89.6%) than in those without one (71.7%).
Key figures
- 12,949retinal photos, from 648 patients with Alzheimer's disease and 3,240 people without it
- 93.2%sensitivity in internal validation (share of people with the disease correctly flagged)
- 0.93area under the curve in internal validation
- 79.6% to 92.1%accuracy across the external test groups
Bringing up memory concerns, with Mima
This AI is still research: no eye photo is used today to diagnose Alzheimer's. If you or someone close to you notices worrying memory lapses, a primary care doctor is the one who can refer you for a memory evaluation.
Mima helps you note what you notice day by day (memory lapses, sleep, mood) and turns it into a clear picture for the appointment. You can prepare your question: "Should I have a memory checkup?"
What could this mean for patients?
If the results hold up, a quick retinal photo could one day help flag people who should be referred for a full memory evaluation. The authors see potential for screening in community settings, away from specialist hospitals.
Today, no eye exam is used to diagnose Alzheimer's. If you or a loved one notice memory problems, the right first step is still to talk to a primary care doctor.
What are the limits of this study?
This was a retrospective case-control study: it compared people already diagnosed with dementia to people without the disease. It does not show that the AI can catch Alzheimer's early, before symptoms, or in the general population.
Performance varied a lot between groups and dropped clearly in people without eye disease, which raises questions about what the AI is really picking up. Prospective studies that follow people over time are needed before any routine use.
Frequently asked questions
Can an eye exam detect Alzheimer's?
Not in everyday practice yet. This study shows an AI can tell apart people with Alzheimer's dementia using retinal photos, but it is not a validated test.
What is a retinal photo?
It is a picture of the back of the eye taken with a fundus camera, sometimes after drops that widen the pupil. It is quick and painless.
Can this AI find Alzheimer's before symptoms start?
The study does not show that: the people in it already had dementia. Studies that follow people over time are still needed.
What should I do if I'm worried about my memory?
Talk to your primary care doctor, who can check for treatable causes and refer you to a memory specialist if needed.
The original study
Cheung CY, Ran AR, Wang S, et al. “A deep learning model for detection of Alzheimer's disease based on retinal photographs: a retrospective, multicentre case-control study” The Lancet Digital Health, September 30, 2022. DOI : 10.1016/S2589-7500(22)00169-8
Read the full study in The Lancet Digital HealthThis article summarizes a scientific study and is not medical advice.