Accurate Eye-Tracking and AI To Detect Neurological Diseases: Interview With Co-Founders of C. Light

Being diagnosed with any number of neurological diseases can be scary, not only due to the progressive nature of many of these conditions, but also because they often cannot be detected until their later stages of progression.

Following the failure of a variety of potential therapies for Alzheimer’s within clinical trials in the past few years, there is an increasing interest in early detection of neurological diseases, with hopes that earlier treatment will be more effective. This has given birth to a wealth of companies interested in identifying biomarkers of early neurological disease progression to enable timely diagnoses and treatment. One such biomarker is eye-motion, with a variety of studies reporting that progressive neurological diseases can heavily impact the way the eye moves.

Current eye-tracking technologies, however, cannot detect subtle disruptions in eye-motion that may be associated with early disease states That was the motivation for the new eye-tracking system that has enabled C. Light Technologies to delineate between different stages of disease severity in multiple sclerosis (MS) patients.

We recently sat down with Dr. Christy Sheehy and Dr. Zachary Helft, co-founders of C. Light Technologies, a Berkeley, California firm, to discuss how their platform works and how they hope to revolutionize the field of neurology.

Mohammad Saleh, Medgadget: Can you tell us about your background and how you each came to be a part of C. Light Technologies?


Dr. Christy Sheehy: My background is primarily in the optical engineering and ophthalmic space. I worked in industry for a while at Corning as a System Test Engineer. They helped me to get my Masters in Optical Engineering, and then I decided that I wanted to transition from more of a manufacturing and testing position to a biomedical area.

I decided to pursue a PhD and I applied to UC Berkley for their Vision Science program. My particular area of study was creating new ophthalmic devices for assessing eye health.

The core technology behind C. Light stems from my dissertation work. It was during my graduate program there that I met my co-founder, Zach.

Dr. Zachary Helft: Christy and I have very different areas of expertise. I’m more of a neurophysiologist.

I was studying visual restoration approaches to retinal blinding diseases. That was my introduction to her technology, because we were using it in our lab.

During the course of my PhD program, I began to really appreciate how much of the brain was truly involved in, not only seeing, but controlling the eyes and how much of the world we experience. Before coming to UC Berkley, I was at Regeneron Pharmaceuticals, where I was making drugs for macular degeneration. I got to see science be applied to help people live better, happier lives on a large scale.

Medgadget: Give our readers an overview on what C. Light Technologies is and how you’re working towards your mission.

Sheehy: We are a neurotechnology and AI company. We’ve created a novel 10-second retinal eye-tracking scan to assess brain health and we’re starting out within the multiple sclerosis field. The big vision is to be able to utilize the eye as a window into neurological health. We always like to say that the back of the eye is also the front of the brain. By looking at the retinal structure and how it moves, we’re able to have granular feedback of disease state on a very fine scale.

Medgadget: How does your technology differ from standard eye-tracking technologies currently on the market? What are you doing differently?

Sheehy: Typical eye-trackers on the market utilize the pupil, which is the front of the eye, as the feature to track. Pupils are generally a 2-4mm a hole through which light enters into the rest of your eye. By going to the retina, the very back surface of the eye, it allows us to have a sensitivity increase. This is because we image the retinal cells – we look at the photoreceptor layer where our rod and cone cells are. By resolving these at around one micron, we have very fine features that we can then use for the tracking. Compared to typical off-the-shelf pupil-trackers, we’re about 120x more sensitive in the features that we’re able to pick up for eye motion.

An image of a retina acquired by C. Light’s platform

Medgadget: Are you tracking cellular structure, cell physiology, or a mixture of those two?

Sheehy: By imaging the cellular structure, it allows us to have extremely fine details that we can monitor as the eyes move. We record a video for 10 seconds, we watch how these cellular structures move over time, and extract the eye motion from those videos.

Medgadget: So the technology behind C. Light consists of both the AI algorithm and a unique measuring device that you’ve developed?

Sheehy: Yes, the device that we use is the core technology of C. Light and was exclusively licensed from UC Berkley. That’s the technology that I developed during my PhD. The patent is for the optical design and the way in which we do retinal imaging. We combine that innovative approach with AI for the full solution.

Medgadget: Do you envision this technology in an ophthalmologist’s or a neurologist’s office? And how does the cost compare to other eye-trackers on the market?


Sheehy: We’re looking at this as a part of a neurologist’s clinic. We’ve been working with an optical manufacturing firm to assess the commercialization at scale and roughly how much a device like this would cost. The price range is typically between $5,000 and $7,500. Compared to other retinal imaging devices and cameras, that’s quite low. Typically, those devices run between $20,000 to $60,000.

Medgadget: Tell us more about this technology driving your company’s success. Do you require an extensive training dataset?

Sheehy: We’ve done some work comparing controls to the MS population, as well as comparing the MS population to itself. That is – patients with a low disease burden compared to more severe disease. Our algorithm is driven by a feature-based approach and is built on the study of about 115 subjects. We can look at particular features and elicit supplementary information to be interpreted by the neurologist. As we expand and acquire more data, we can transition to a convolutional neural net or different types of approaches to explore the larger dataset. However, even with this relatively small number of datapoints from 115 patients, we’re able to see around an 86% accuracy of predicting neurological disability when we combine our method with a 25-foot walk and a fatigue survey. With eye-tracking alone, we’re at about 82% accuracy. We’re extremely excited to have such a high accuracy with a relatively small patient sample. Usually, AI algorithms need training sets consisting of hundreds of thousands of scans.

Medgadget: How can you do all of this in just 10 seconds?

Sheehy: We extract the eye motions from the video about 500-1000 times per second, depending on which type of algorithm we’re running. This allows us to acquire a wealth of data in a very short time from any given individual. All of that data is then fed into our machine learning algorithm, giving us hundreds of thousands of data points that we can use from a short ten-second scan. That’s what drives our ability to be fast – it’s our ability to capture a lot of information in a very short period of time.

Medgadget: How does a clinician or patient interact with your platform? Do you provide guidance to whoever is interpreting the data?

Sheehy: The overall approach of using feature-based neural networks allows the data to be much more interpretable to those looking at our data. The goal is to allow a physician to compare a particular patient’s eye motion compared to a control population, as well as compared retrospectively to themselves. We are in our early stages with our prototype device, which has been deployed at two research institutions. So the specifics of the readout that a physician will receive is yet to be finalized. We will be conducting physician interviews to tease out exactly the type information that would be most useful to a clinician in different scenarios.

Medgadget: How has this solution impacted patients’ lives so far?

Sheehy: Our first MS study was recently published. We’ve shown that eye-motion detection with our device can distinguish between different levels of disability.

Medgadget: Beyond multiple sclerosis, I understand that C. Light Technologies has plans to tackle a variety of different conditions?

Helft: We have an Alzheimer’s study currently open. We’ve only collected data from less than 10 patients thus far, so we’re unable to draw any meaningful conclusions just yet. We know, though, that eye-motion is affected in late-stage AD with a large-magnitude. Our theory is that large eye-motion abnormalities start off as small ones early in the disease progression. We also have an ongoing study looking at concussions with 50 patients and 50 age-matched controls. Future plans include looking at Parkinson’s, epilepsy, and ALS. So many different neurological disorders manifest as in some type of eye-motion abnormality because so many different areas of the brain coordinate together to generate motion of the eye. Our job will be to determine how early we can detect those changes using our incredibly accurate eye-tracking technology. We also need to determine the level of specificity that we can delineate between diseases using this technology – which is why we’re focusing on MS patients who are already diagnosed. We can bring extreme value to these patients by determining their disease progression and providing feedback on whether their current medication is effective.

Medgadget: Down the line, do you see this more as a predictive or diagnostic tool? Where do you see this technology in 10 years?

Helft: We want this to be one of the earliest warning indicators, so that we can get patients into a neurologist’s office even before any severe symptoms begin. Right now, neurologists only see sick patients. We want to create an opportunity for neurologists to join the likes of cardiologists – they’re able to detect that something is headed in a bad direction and start to work on preventing the disease.

Sheehy: We definitely hope to be a platform technology for neurological health. We’d like to have enough data for MS, Parkinson’s, Alzheimer’s, and concussion to delineate between them. Our big vision is to provide a preventative solution for the neurology field. We want to enable individuals to enjoy their life and their loved ones longer and provide patients and families with powerful information to inform patient care and decrease their stress as they go through the diagnosis.

Source