Summary: Researchers have developed an innovative machine learning model that uses retinal imaging to distinguish normal cognition from mild cognitive impairment (MCI). This approach offers a non-invasive, cost-effective opportunity to detect early signs of cognitive decline that can precede Alzheimer’s disease.
The model evaluates multimodal retinal images acquired with optical coherence tomography (OCT) and OCT angiography (OCTA), identifying image features and quantitative measurements that correlate with cognitive status. By combining image-derived metrics with basic patient information, the system reliably identifies individuals with MCI, representing an important advance toward earlier, scalable screening for neurodegenerative risk.
Key Facts:
- The machine learning model achieved 79% sensitivity and 83% specificity for distinguishing individuals with mild cognitive impairment from those with normal cognitive function using OCT/OCTA retinal images and clinical data.
- This study is the first to apply retinal OCT and OCTA imaging specifically to differentiate mild cognitive impairment from normal cognition.
- The Duke Health group previously designed a related model that used retinal imaging to identify patients with established Alzheimer’s disease, and this new work extends that approach to earlier stages of cognitive decline.
Source: Duke University
Machine learning model at Duke Health differentiates normal cognition from mild cognitive impairment using retinal imaging
A team at Duke Health developed a convolutional neural network that analyzes multimodal retinal images—OCT and OCTA—alongside demographic and clinical variables to detect mild cognitive impairment. The algorithm pinpoints structural and microvascular features in the retina that are associated with early cognitive changes, offering a practical screening tool that is non-invasive and relatively inexpensive.
Published in the journal Ophthalmology Science, the study highlights the retina as an accessible window into brain health. OCT and OCTA capture high-resolution cross-sectional and vascular images of the neurosensory retina, enabling the detection of subtle structural and microvascular alterations that correlate with neurodegenerative processes.
“This is particularly exciting because earlier models struggled to separate mild cognitive impairment from normal cognition,” said senior author Sharon Fekrat, M.D., professor in Duke’s departments of Ophthalmology and Neurology and associate professor in the Department of Surgery. “Our findings move us closer to identifying cognitive impairment earlier, before progression to Alzheimer’s dementia.”
The current model builds on prior Duke research that successfully detected established Alzheimer’s disease using retinal imaging. By applying advanced machine learning to OCT/OCTA images and incorporating patient-level data—age, sex, visual acuity, education level, and quantitative OCT/OCTA metrics—the team trained an algorithm to recognize patterns indicative of MCI.
Co-first author C. Ellis Wisely, M.D., notes the clinical importance: “A non-invasive, affordable method to reliably identify patients with mild cognitive impairment is increasingly valuable, especially as new therapies for Alzheimer’s disease are developed.” Co-lead author Alexander Richardson, a student in Duke’s Eye Multimodal Imaging in Neurodegenerative Disease lab, added, “Machine learning applied to retinal imaging can enable large-scale neurological screening by leveraging a readily available, cost-effective test.”
The research team includes Sharon Fekrat, C. Ellis Wisely, Alexander Richardson, Ricardo Henao, Cason B. Robbins, Justin P. Ma, Dong Wang, Kim G. Johnson, Andy J. Liu, and Dilraj S. Grewal. The study received partial support from the Alzheimer’s Drug Discovery Foundation.
About this machine learning and dementia research news
Author: Sarah Avery
Source: Duke University
Contact: Sarah Avery – Duke University
Image: Image credited to Neuroscience News
Original Research: Open access. “A convolutional neural network using multimodal retinal imaging for differentiation of mild cognitive impairment from normal cognition” by Sharon Fekrat et al., published in Ophthalmology Science.
Abstract
A convolutional neural network using multimodal retinal imaging for differentiation of mild cognitive impairment from normal cognition
Mild cognitive impairment (MCI) is often viewed as an intermediate clinical state between normal cognitive aging and Alzheimer’s disease (AD). In MCI, cognitive deficits are present but daily functioning remains largely preserved. MCI can be classified into amnestic and non-amnestic subtypes; amnestic MCI primarily affects memory and is more likely to progress to Alzheimer’s disease than non-amnestic forms.
Some estimates suggest that individuals with amnestic MCI may convert to Alzheimer’s disease at rates near 20% per year, underscoring the urgency of early detection. Identifying MCI reliably—especially in those at greatest risk of progression—remains a clinical challenge. Widely available, non-invasive biomarkers are therefore needed to support diagnosis and guide early intervention as therapeutic options advance.
This study demonstrates that machine learning applied to multimodal retinal imaging—coupled with demographic and clinical measures—can serve as a promising biomarker for MCI. By detecting retinal structural and microvascular signatures associated with early cognitive decline, retinal imaging could become an accessible screening strategy to identify patients for further neurological evaluation and potential entry into early treatment pathways.