Summary: A new study using artificial intelligence to analyze routine retinal photographs has revealed an inexpensive, non-invasive way to predict major Alzheimer’s disease risk factors years or even decades before clinical symptoms emerge. By training deep learning models on retinal images from more than 40,000 participants in a UK-based databank, researchers identified specific retinal regions—particularly retinal blood vessels and the optic nerve head—that correlate with biological and lifestyle factors linked to Alzheimer’s vulnerability.
The AI models reliably predicted attributes such as biological sex, blood pressure, smoking status, alcohol use, and insomnia. Because the retina is an extension of the central nervous system, widely available retinal photographs act as an “integrated biological sensor,” reflecting cumulative neurovascular changes and offering a practical window for early intervention long before irreversible brain damage occurs.
Key Facts
- The ocular window: Retinal structure—especially the retinal vasculature and optic nerve head—contains measurable clues about neurovascular health and Alzheimer’s risk.
- Large-scale validation: Deep learning models were trained and validated on retinal images from over 40,000 individuals in a major UK databank.
- Objective risk assessment: The approach detects lifestyle and biological risk factors (high blood pressure, smoking, alcohol use, sleeplessness) more objectively than self-reported records.
- Early detection opportunity: Because Alzheimer’s pathology develops over decades, retinal screening can identify at-risk individuals well before late-stage, irreversible changes occur.
- Accessible and cost-effective: Retinal photography is already common in routine eye exams and chronic disease screenings, making this approach scalable and far less expensive than MRI or PET imaging.
Source: University of Florida
Often called “the window to the soul,” the eyes may also reveal important information about brain health.
In a study of tens of thousands of retinal photographs, researchers demonstrated that common color fundus images can predict many established risk factors for Alzheimer’s disease. The findings suggest retinal imaging combined with AI could serve as a practical screening tool to flag individuals who may benefit from early preventive measures.
“Alzheimer’s disease develops over many years, yet most diagnostic tools detect late-stage pathology, when interventions are less effective,” said Ruogu Fang, Ph.D., professor of biomedical engineering at the University of Florida and lead author of the study. “By examining novel biomarkers such as retinal morphology, we can identify people at elevated risk and offer targeted tests and lifestyle recommendations that may reduce their long-term risk.”
Fang and collaborators, including Adam Woods, Ph.D., and Yunchao Yang, Ph.D., published their results in the Journal of Alzheimer’s Disease (June 16). Because retinal photographs are routinely taken for conditions like diabetes, glaucoma, cataracts, and during standard eye exams, this screening approach could be implemented at low cost and with minimal disruption to current clinical workflows.
Using deep learning on more than 40,000 retinal images from a UK-based repository, the research team localized retinal regions that most strongly contributed to predicting Alzheimer’s-related risk factors—especially the optic nerve head and retinal vasculature. These saliency maps showed that the model was focusing on biologically sensible areas rather than arbitrary image features.
Seowung Leem, a doctoral student at UF and first author, explained: “AI enables detection of subtle retinal changes across thousands of images that would be difficult to spot by eye. These patterns can serve as reproducible indicators of future disease risk.”
The model successfully predicted both categorical traits (sex, smoking, sleeplessness, socioeconomic status, alcohol use, depression) and continuous measures (age, education years, body mass index, systolic and diastolic blood pressure, HbA1c). Importantly, retinal-derived scores differed in people who later developed Alzheimer’s compared with matched controls, suggesting overlap between retinal signs of risk factors and preclinical AD changes.
Unlike medical charts that often rely on incomplete or biased patient reports, retinal imaging provides an objective record of accumulated biological damage. As Fang noted, “Retinal morphology functions less like a questionnaire and more like a biological sensor reflecting cumulative neurovascular burden.” That objective signal could guide clinicians to recommend aggressive cardiovascular risk control, improved sleep interventions, cognitive training programs, or early-stage medical options when they are most likely to be beneficial.
Funding: The work was supported in part by the National Science Foundation.
Key Questions Answered:
A: The retina is developmentally and anatomically an extension of the central nervous system. Its blood vessels and nerve fibers mirror the brain’s microvasculature and neural pathways, so systemic inflammation, vascular disease, and neurodegenerative changes often manifest as detectable changes in retinal structure.
A: Patient records frequently rely on self-reported lifestyle details that are incomplete or biased. The AI reads objective structural consequences of those lifestyles on ocular tissue accumulated over time, turning a simple photo into a measurable indicator of neurovascular vulnerability.
A: Early identification enables proactive prevention: improved cardiovascular management, targeted cognitive training, sleep therapies to enhance brain toxin clearance, lifestyle changes (smoking cessation, reduced alcohol use, exercise, diet), and consideration of early therapeutic strategies when appropriate.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- The journal article was reviewed in full.
- Additional context was added by editorial staff.
About this AI and Alzheimer’s disease research news
Author: Eric Hamilton
Source: University of Florida
Contact: Eric Hamilton – University of Florida
Image: Image credited to Neuroscience News
Original Research: Closed access. “Prediction of Alzheimer’s disease risk factors from retinal images via deep learning: Development and validation of biologically relevant morphological associations in the UK Biobank” by Ruogu Fang et al., Journal of Alzheimer’s Disease. DOI:10.1177/13872877261457650
Abstract
Prediction of Alzheimer’s disease risk factors from retinal images via deep learning: Development and validation of biologically relevant morphological associations in the UK Biobank
Background
Systemic, metabolic, and lifestyle factors are linked to Alzheimer’s disease through epidemiology and biomarker research. It remains unclear whether standard color fundus photography contains structural retinal signatures that correspond to these Alzheimer’s-related risk domains.
Objective
To determine whether deep learning models can predict a set of 12 Alzheimer’s-related risk factors from color fundus photographs and to identify the retinal structures that support those predictions, thereby assessing whether retinal images reflect pathways to Alzheimer’s vulnerability.
Methods
Using 62,876 color fundus photographs from 44,501 participants in the UK Biobank, deep learning models were trained to predict 12 factors linked to Alzheimer’s incidence or pathology: six categorical (sex, smoking, sleeplessness, socioeconomic status, alcohol use, depression) and six continuous (age, years of education, body mass index, systolic and diastolic blood pressure, HbA1c). Model performance, saliency maps, and saliency-derived scores (CAM-Score) were evaluated and compared to morphometric retinal measurements. Scores were also compared between individuals who later developed Alzheimer’s (on average 8.55 years before diagnosis) and matched controls.
Results
Model performance varied by factor, with AUROC values for categorical predictions ranging from 0.5654 to 0.9480 and R2 for continuous predictions ranging from −0.0291 to 0.7620, generally outperforming morphometry-based models. Saliency analyses consistently highlighted biologically relevant regions, notably the optic nerve head and retinal vasculature, and corresponded with measured morphometric variations. Several saliency-derived scores differed significantly between incident Alzheimer’s cases and matched controls, suggesting overlap between retinal correlates of risk factors and early disease-related changes.
Conclusions
Color fundus photography encodes retinal signatures associated with Alzheimer’s risk factors. While not diagnostic on its own, deep learning–derived retinal representations may reveal biologically meaningful structural changes that reflect vulnerability to Alzheimer’s disease and could inform cost-effective early screening and preventive strategies.