Alzheimer’s Hiding In The Eye?

An AI model analyzing a simple eye scan can identify Alzheimer’s disease with over 90% accuracy — and researchers at Mayo Clinic say this could change how we catch the disease before it steals a single memory.

Story Snapshot

AI trained on retinal photos correctly identified Alzheimer’s patients more than 90% of the time in Mayo Clinic and European population studies.
Physical signs of Alzheimer’s — including amyloid and tau protein deposits — have been found directly in the retinas of patients and animal models.
A University of Florida study used AI on over 40,000 eye scans to detect key Alzheimer’s risk factors like high blood pressure and smoking history.
Researchers are still validating results against established tests like amyloid PET scans, and no cure exists yet for presymptomatic Alzheimer’s.

Your Eye Doctor May See Alzheimer’s Before Your Neurologist Does

The retina is brain tissue. It grows from the same embryonic cells, shares the same blood vessel structure, and responds to the same disease processes. That is not a metaphor. It is anatomy. Mayo Clinic neurologist Dr. Oana Dumitrascu and Arizona State University computer scientist Dr. Yalin Wang built an AI model on that biology — and the results stopped researchers in their tracks. The model analyzed standard retinal photos and separated Alzheimer’s patients from healthy controls with more than 90% accuracy.

What makes this striking is what the AI is reading. Changes in medium and small blood vessels in the retina serve as key markers tied to Alzheimer’s disease. But those vascular shifts are only part of the picture. Scientists have physically identified amyloid beta protein deposits and abnormal tau proteins inside the retinas of Alzheimer’s patients and animal models — the same hallmarks found in the brain. The eye is not just a mirror of the brain. In some ways, it may be an earlier warning system.

The AI Is Reading Risks You Cannot See With the Naked Eye

A research group at the University of Florida pushed this further. They trained an AI model on retinal photographs from more than 40,000 patients in the UK Biobank — one of the largest health databases in the world. The model detected signals tied to blood pressure, smoking history, and alcohol use directly from eye scans. Those three factors are known Alzheimer’s risk drivers. Spotting them through a routine eye exam, years before symptoms appear, is the kind of early warning the medical system has never had before.

Duke University is running a clinical trial using a technology called Optical Coherence Tomography Angiography to study retinal blood vessel structure in patients with mild cognitive impairment and Alzheimer’s. Mayo Clinic has its own dedicated study underway, led by Dr. John Chen, focused on using retinal imaging to identify people at risk and track how the disease progresses. These are not fringe experiments. They are institution-grade research programs with serious infrastructure behind them.

Why This Is Not a Done Deal Yet

Here is where intellectual honesty matters. The Mayo Clinic team openly states that their findings still need to be validated against gold-standard tests — amyloid PET scans and cerebrospinal fluid analysis. That work is ongoing. A 2023 systematic review found weak statistical evidence for some widely studied retinal markers, with small study sizes and inconsistent methods making it hard to draw firm conclusions. A review published in the Journal of Neurology, Neurosurgery and Psychiatry went further, stating that retinal imaging cannot yet be considered a confirmed source of Alzheimer’s biomarkers.

There are also practical limits. On standard retinal photos, only vascular changes are directly visible. The AI infers the presence of amyloid, tau, and inflammation — it does not see them outright. Telling Alzheimer’s apart from mixed dementias remains a hard problem for the model. And even if the AI is right, there is currently no approved treatment for presymptomatic Alzheimer’s, which raises a real ethical question: what do you do with a diagnosis you cannot act on? These are not reasons to dismiss the research. They are reasons to track it carefully.

The Bigger Picture Behind the Eye Scan

Current diagnostic tools — cognitive testing, MRI, cerebrospinal fluid analysis, PET scans — are expensive, invasive, and often used too late. Dr. Dumitrascu notes that by the time patients reach those tests, the disease has usually been progressing for years. A retinal scan costs a fraction of a PET scan and takes minutes. If the validation studies confirm what early results suggest, the case for integrating retinal screening into routine eye care becomes very hard to argue against. The researchers’ stated goal is exactly that — making this part of standard care. Getting there requires large, diverse datasets and multi-center studies that have not yet been completed. But the direction is clear, and the momentum is real.