NeuroVoices: Sharon Cohen, MD, FRCPC, on the Diagnostic Potential of Retinal Imaging for Alzheimer Disease

Sharon Cohen, MD, FRCPC

RetiSpec is a medical imaging company that uses advanced artificial intelligence paired with a hyperspectral retinal imaging device to detect Alzheimer disease (AD) biomarkers in the eye. The noninvasive tool captures retinal images using a broad wavelength of light to locate developmental and biological similarities to the brain.

It has been studied in a few different settings, most notably in a cross-sectional validation study led by Sharon Cohen, MD, FRCPC. Here, adults between ages 50 to 90 years old who were at risk for or who had preclinical AD or mild cognitive impairment (MCI) underwent amyloid-beta (Aß) assessment via cerebrospinal fluid and/or PET within 12 months of retinal imaging. Conventional and spectral images for each of the retinal targets were located in recordings from each eye. Data were then classified with a Multiple Instance Learning algorithm based on mapped big space methods to detect spatial-spectral patterns in hyperspectral-retinal images.

The findings, presented at the 14th Clinical Trials in Alzheimer Disease (CTAD), November 9-12, 2021, indicated that hyperspectral imaging is effective in predicting brain Aß, with patients reporting an easy and comfortable experience while undergoing RetiSpec. Cohen, neurologist and assistant professor at the University of Toronto, sat down with NeurologyLive® to provide additional context on the data, the advantages of this approach, and when clinicians can anticipate retinal imaging becoming more commonplace in the diagnosis of AD.

NeurologyLive^®: Can you discuss the results presented at CTAD 2021?

Sharon Cohen, MD, FRCPC: Toronto Memory Program, which is the site I lead, was one of two centers who contributed patients to this validation study and the validation of RetiSpec as a predictor of amyloid in the brain. Shiva Medical Center in Israel was the second site that contributed some of the patients for the validation study results presented at CTAD 20221. We had 108 patients who underwent a RetiSpec scan, or picture of the eye. These were individuals who were either at risk for developing Alzheimer disease but were cognitively normal, or who already had a diagnosis of preclinical Alzheimer disease, were cognitively normal, but had amyloid changes in the brain. The other group was slightly further along in the Alzheimer continuum. This included individuals with prodromal Alzheimer, so the pre-dementia stage called mild cognitive impairment.

Diagnosis matters. Early diagnosis, not just in the late stage of the disease or at autopsy, matters. With RetiSpec, we wanted to see whether taking a picture of a retina could yield information that had high accuracy and therefore could avoid the problems of cost, access, and radiation exposure in the case of PET amyloid imaging, or invasiveness, as in lumbar puncture to get spinal fluid. The 108 subjects we reported on were either amyloid positive by the gold standard of PET or CSF, or amyloid negative. In a blinded manner, the RetiSpec team performed the eye scans. They took pictures of each of the patient’s eyes with a special hyperspectral camera, analyzed the results, and then compared them with the gold standard tests. This is where we got the area under the curve, or the ability of RetiSpec to accurately predict brain amyloid accumulation. The AUC was 0.80, which was really exciting. That translated to a sensitivity of 86% and specificity of 80%, making it very hopeful, at least with this initial validation data, that we have a tool that can be used at point of care in a memory clinic or in an optometrist office that noninvasive, relatively inexpensive, and can predict brain amyloid accumulation with accuracy.

Hyperspectral retinal imaging seems vert useful; why aren’t we taking full advantage of this technology yet?

Science takes its time and validation of biomarkers need to reach a certain threshold before we roll this out for clinical use. I do think the technology is fascinating. It’s not a completely new idea that we might be able to see changes in the brain by looking at the retina, but previous results with other retinal imaging technologies have not yielded the same optimism. With hyperspectral imaging and RetiSpec’s unique proprietary algorithm of analyzing, we’re able to look at the retina and see what happens when you expose it to a very broad range of wavelengths of light, much brighter than the human eye can see. We’re not actually looking at a picture of amyloid in the retina, we’re looking at a signature, if you will, of the biophysical properties.

Through this machine learning algorithm that accompanies the camera, the analysis is completely objective. It’s not like a radiologist thinking “is there amyloid or not?” and having to make a decision. With this algorithm it gets better the more cases it assesses. It’s very fascinating, and I think as we approach the level of validation that will allow it to be commercialized, it will be very useful whether as a standalone diagnostic or as a screening measure, to narrow the funnel of who needs to go on more definitive biomarker confirmation. The rollout is yet to be seen, but the fact that it’s under development and already has some good results is promising.

What other types of research need to be conducted to further expand on the utility of this imaging?

We welcome people who want to move science forward, whether an individual feels at risk by virtue of memory symptoms or a family history, whether that’s the presence of risk gene APOE. We currently have an ongoing validation study. People are very keen to participate. It gives them access to information that they might not otherwise get and a chance from an altruistic end point to move the field forward. In the US, there’s a study called the BIO-HERMES study that’s led by the Global Alzheimer’s Platform Foundation. RetiSpec is one of several biomarkers being validated in the BIO-HERMES study. There are several sites within the US where one could participate.

The group that we struggle the most to attract are those who feel they’re at low risk, and so are not as motivated to engage. But for normal healthy controls, what does the retina look like? This group has a lot to offer. We need to improve our messaging to say, please come join us. If not for you, for next generations, for your neighbor, for people that do need this early diagnosis. With disease-modifying therapies starting to become more exciting, it will be important that we don’t miss an early diagnosis, because by the time Alzheimer disease becomes symptomatic, millions or brain cells have already been lost. Going the earliest you can and diagnosing allows us the opportunity to protect the brain. It’s like any other area in medicine. You wouldn’t say, “oh well, there’s only a little nest of cancer cells, we’ll just watch it and see if anything happens.” You would jump in and say, here’s the opportunity to treat. With RetiSpec, if we can help people understand the importance of early detection, this will advance the field in terms of diagnostics and allow therapeutics to be rolled out successfully.

Transcript was edited for clarity. For more installments of NeuroVoices, click here.

REFERENCE
1. Cohen S, Plante A, Hazan Al, et al. RetiSpec’s hyperspectral imaging system: results of a validation study in preclinical AD and MCI. Presented at CTAD 2021; November 9-12. Poster LP16