How Retinal Imaging May Help Identify Alzheimer's Disease

Brianne N. Hobbs, OD


March 04, 2021

Retinal imaging can identify structural changes that accompany neurodegenerative conditions, such as Alzheimer's disease and Parkinson's disease.[1,2,3,4,5] Rendering this information clinically useful, however, is not so straightforward. Instead, it requires generating a model capable of translating retinal imaging data into an accurate means of detecting symptomatic disease.

In an exploratory study, investigators at Duke University sought to do just that, designing and testing a proof-of-concept machine modeling approach to the diagnosis of Alzheimer's disease by harnessing optical coherence tomography (OCT).

OCT multimodal imaging can capture and quantify an enormous amount of information about the retina. In many types of ocular disease, the link between specific OCT findings and the likelihood of the diagnosis is well-established because retinal imaging is commonly performed. There is a large, diverse pool of data to draw from; essentially, we know what layers to look at and which measurements matter. But far less is known when the use of retinal imaging is extended to nonocular conditions, so investigators first must determine which layers are the most sensitive indicators of disease.

It turns out that the ganglion cell-inner plexiform layer (GC-IPL) is the most predictive for Alzheimer's disease. This is not an unexpected finding, given that the ganglion cell axons comprise the optic nerve that can be affected by glaucoma, another neurodegenerative disease specific to the eye. Understanding the importance of the GC-IPL complex was a major advance in being able to generate a machine model that could predict Alzheimer's disease.

To evaluate different predictive models, researchers scanned the eyes of 36 patients with Alzheimer's disease and 123 patients who were cognitively normal. They found that the model with the highest predictive value included the GC-IPL map, quantitative OCT and OCT angiography data (determined from a previous study), and patient data.

Impressively, when this model was then applied to an independent test set, the model had a predictive value of 0.841. What is perhaps equally impressive is that the model using only images without any quantitative or patient data performed nearly as well as the model including these elements.

If an image-only model using artificial intelligence (AI) could be used in the diagnosis of Alzheimer's disease, it would greatly streamline the diagnostic process and improve patient experience.

AI has previously been used in creating a machine learning model for predicting Alzheimer's disease using MRI and PET scans, but this retinal imaging–based model is a more appealing means of identifying neurodegenerative disease. Rather than enduring expensive neuroimaging scans, a symptomatic patient could sit for a retinal scan that takes only a few seconds, is comfortable, and is completely noninvasive. The results of these scans are available immediately, so wait time would be minimized.

Some might question the value of being able to diagnose an incurable neurodegenerative disease with greater efficiency, given that it doesn't affect the outcome of the disease — a valid concern. Yet identifying the relationship between retinal thinning and Alzheimer's disease may ultimately prove beneficial in the larger picture, either by evaluating the efficacy of a treatment or providing new insights about the pathophysiology of the disease.

The study has several obvious limitations that must be taken into consideration. The sample size was small and lacked diversity, although this also means that a larger and more diverse pool of participants would probably enhance the performance of model by providing more data with which to train the AI. In addition, the patients in the study were largely free from ocular disease, even though Alzheimer's disease often clinically coincides with age-related macular degeneration, glaucoma, and visually significant cataracts.

Despite these limitations, the findings of this study hold considerable promise. They indicate that by taking advantage of ophthalmology's powerful technology, we may one day have the means to help our neurologist colleagues diagnose and potentially treat one of the most debilitating diseases in medicine.

Brianne N. Hobbs, OD, is on staff at the Charlotte Community Based Outpatient Clinic and Charlotte Health Care Center within the Veterans Health Administration in North Carolina. Previously, she served as associate director of exam innovation at the National Board of Examiners in Optometry, where she was engaged with the creation of a new clinical skills exam for optometry. She has spent most of her career in academia and has also worked in a hospital-based setting.

Follow Medscape on Facebook, Twitter, Instagram, and YouTube



Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.