Your eyes could hold the key to unlocking early detection of Alzheimer’s and dementia, with a groundbreaking AI study. Called Eye-AD, the deep learning framework analyzes high-resolution retinal images, identifying small changes in vascular layers linked to dementia that are often too subtle for human detection. The approach offers a rapid, non-invasive screening for cognitive decline, helping doctors slow progression and improve patient outcomes.
Alzheimer’s Disease (AD) affects over 50M people worldwide, with cases expected to rise as the population ages. Early detection and treatment are critical to improving a patient’s quality of life, enabling clinical interventions to slow disease progression, and giving families more time to plan for long-term care and support.
Referred to as a ‘window to the brain” the retina shares embryonic origins with the brain. Studies have shown that changes in the retina’s microvasculature—tiny blood vessels—are often linked to cognitive decline. However, these early-stage changes are hard to detect and traditional methods like MRI and spinal fluid analysis are more costly and invasive for making an official diagnosis.
The researchers developed Eye-AD, a model that combines a convolutional neural network (CNN) to extract features from the retinal images with a graph neural network (GNN) to analyze relationships within and between retinal layers for disease detection.
The model uses Optical Coherence Tomography Angiography (OCTA) images—a noninvasive imaging technology—to visualize blood flow and vascular details in the back layers of the eye. By analyzing the OCTA images, Eye-AD identifies clinical biomarkers to predict early-onset Alzheimer’s Disease (EOAD) and mild cognitive impairment (MCI).
The model was trained on 5,751 OCTA images from 1,671 patients using PyTorch on a workstation with four NVIDIA GeForce RTX 3090 GPUs, accelerating training time and enabling efficient processing of the high-resolution images.
Eye-AD outperformed other models and excelled in detecting EOAD, with an AUC (a measure of model accuracy) of 0.9355 on internal datasets, and 0.9007 on external datasets. Its performance was slightly lower for MCI detection, with an AUC of 0.8630 internally and 0.8037 externally.
The researchers also found that the deep vascular complex—a layer of blood vessels in the retina—showed significant changes linked to cognitive decline, and served as a key clinical biomarker for Eye-AD in accurately predicting early disease.
According to the study, the model represents a significant advancement in early and efficient dementia detection, with the possibility for widespread use in cognitive health assessments. Future efforts will focus on validating Eye-AD across diverse populations and integrating it with other diagnostic tools to help doctors in dementia screening in clinical practice.
The source code is available on GitHub.
Read the research Early detection of dementia through retinal imaging and trustworthy AI.
Photo courtesy of Freepik
