Skip to Content

AI In Practice

AI in Practice Art.
How will artificial intelligence intersect with the practice of optometry, and what do humans bring to the table?
Written by Zac Unger, Illustration by Brian Stauffer.

A lot has changed in the world of optometry since the end of World War II, but at least one thing has remained consistent: since 1947, there has been at least one Dr. Sarver providing high-quality patient care within two miles of the Berkeley campus. Dr. Morton Sarver (OD, ’47) opened his practice on College Avenue after graduating from the UC Berkeley School of Optometry and was joined by his Berkeley-trained sons Don and Larry in 1971 and 1980, respectively. The first Dr. Sarver taught at Berkeley and was a groundbreaking researcher in the field of contact lens design and corneal physiology; Larry has also taught at Berkeley, as well as working at Stanford’s Laser Eye Center. But research and resumes aside, Larry Sarver boils the practice’s eight decades of success down to one thing: “It’s pretty straightforward. We just like to take care of people.”

A few miles up the street from Sarver’s practice, back on the Berkeley campus, another sterling researcher at the School is approaching eye care from an entirely different, more modern angle. Dr. Jorge Cuadros (OD, ’80), a world-renowned expert on big data and informatics, is conducting cutting-edge research on how artificial intelligence and deep learning systems can revolutionize the practice of optometry. “Whereas a human can look at a few dozen features in an image, AI is finding patterns in a million features,” says Cuadros. “And AI can process different relationships between features to extract biomarkers that are not visible to humans.”

As with every conversation about artificial intelligence in any field, an uneasy tension hovers at the margins when it comes to eye care: Is big data poised to kill the small family practice? Are the Sarvers and Cuadroses of the world in competition, or can we forge a future where these two very different kinds of doctors can collaborate to improve patient care in big ways? The next new technology is always exciting…and also usually a little bit overhyped. And while AI has great promise in its own right, it also may prove valuable in forcing us to ask hard questions about what humans bring to the table, and may even make eye doctors better at the human side of patient care, doing the things a computer won’t ever be able to.

Optometrists and ophthalmologists have been incorporating technology into their practices for decades. In fact, as far back as the 1950s, eye care was one of the first applications for telemedicine—rudimentary retinal cameras were used to perform fundus photography, with the resulting images sent to specialists. The advent of high-speed Internet in the early 2000s supercharged the process of transmitting digital images, yet nobody suggested that optometrists were on the verge of becoming obsolete.

With artificial intelligence dominating conversations across so many disciplines, it’s important first to define what’s being discussed when it comes to the practice of optometry. At present, AI is largely utilized in diagnostics for disease, with diabetic retinopathy being the most well-established application. Dr. Cuadros has been at the forefront of this process ever since UC Berkeley hosted a competition in 2015, where six hundred teams competed to build an artificial intelligence system that could accurately diagnose diabetic retinopathy from a dataset of one hundred thousand images. “Within three months, the algorithms were performing better than humans,” Cuadros recalls.

But this didn’t mean the end of flesh-and-blood clinicians. Instead, the technology allowed the profession to expand its reach, diagnosing retinopathy remotely and automatically in underserved populations like California’s Central Valley and further afield in the developing world. In 2018, the FDA approved the first AI-enabled medical device of any kind—a software program paired with a retinal camera that can be used by a primary care physician to diagnose and grade the severity of diabetic retinopathy.

“I think we need to expand our ideas about how eye care should be practiced beyond the four walls of an examining room to know how AI can really fit in,” Cuadros says. He explains the concept of a “cognitive funnel” to describe a potential division of labor between the digital and the human. “Computers are very good at tedious, narrow, focused tasks, and humans are better at abstraction, taking a larger view of what’s around to help make a better decision.” Computers, however, are only as good as the information inside them; artificial intelligence isn’t actually intelligent—it’s just exceptionally well-informed. Consider a disease like choroidal melanoma, Cuadros says, referring to a malignant intraocular tumor. “It’s not even that rare, but there just aren’t enough images of it yet to train an AI. You need high-quality data.”

Dr. Cuadros Medical technologists who work with artificial intelligence make a distinction between “assistive” and “autonomous” processes. Autonomous AI—like the program that independently diagnoses diabetic retinopathy—has made the biggest splash in our consciousness and inspired sci-fi fears about all-knowing robots replacing clinicians. But the more likely use case, at least in the near future, is as a supremely smart and capable assistant. Soon, says Cuadros, AI will be able to “help with triage, to tell an optometrist, ‘Hey, this patient is more urgent,’ or maybe just make their diagnostic skills more consistent without completely replacing them.”

That AI expansion is also going to rapidly encompass the field of oculomics, or the ability to use ophthalmic biomarkers to detect systemic health problems or disease states in other parts of the body. In a 2021–22 study using over 38,000 patients, Cuadros collaborated with data scientists to train a deep learning system to use simple eye photography to predict a patient’s laboratory results for liver, kidney, thyroid, and blood sugar levels. The photographs were basic and noninvasive, without any need to pull the eyelid down to expose the lower conjunctiva. The artificial intelligence successfully revealed severe kidney abnormalities quickly and noninvasively from a simple photograph.

Which is exciting from a research perspective, but what does it actually mean for patients? A lot of people skip their annual visit with their primary physician because they feel fine, but they keep their appointments with their optometrists because they notice the effects of outdated eyeglass prescriptions every time they try to read something on their phone. “For people like that, going to an optometrist and finding out they have kidney disease could be a big thing,” says Cuadros, suggesting a potentially large expansion of the value of a simple annual eye exam. “And in our naïve mind we think that should be sufficient. But it’s not. It’s going to take a lot more to motivate someone to really make a change in their lives.”

Having mountains of information available can be a double-edged sword for both practitioners and patients. If an AI assistant suggests to an eye doctor that a patient is at risk for kidney disease, the easiest thing to do would be to refer that patient to a specialist. “And we think, ‘Oh well, that should be sufficient,’” says Cuadros. But for the patient with significant disease, a simple referral without follow-up isn’t necessarily enough to motivate them to see the right doctors or modify their lifestyle.

The opposite alternative is also worrisome, says Cuadros. “Over-referrals are a hidden and important problem,” he says, “because if we send them down a road and it turns out they’re fine, we’ve wasted their time and resources, and the next time a doctor tells them they have a problem, they may not trust it.” A computer isn’t going to be able to combine the data with the intangibles the way a doctor can, nor will it be able to design a care plan that fits the specific ways each unique individual interacts with the medical establishment. “And,” Cuadros warns, “if a doctor relies on autonomous AI, it’s not just the AI that’s going to get sued.”

Dr. Angela Shahbazian (OD, ’16), an assistant clinical professor at Berkeley’s Herbert Wertheim School of Optometry & Vision Science who specializes in primary and community health care, has been thinking a lot about how AI might change the practice of eye care and how we can begin to prepare optometry students for the shift. “Things are changing dramatically from the time a student enters to when they graduate,” she says. “So we definitely need to make them aware of what exists and what the potentials are.”

Shahbazian sees promise for assistive AI as a backstop to help identify rare diseases and also to increase access for people who can’t afford the time or expense to physically visit a high-tech eye center like Berkeley’s. Recently, Dr. Shahbazian saw a patient with a rare chromosomal condition called hereditary spastic paraplegia, a condition that leads to gradual loss of use of the leg muscles. “But it’s also often associated with retinal dystrophy,” she says, which can lead to vision loss. “He never had a reason to see an eye doctor, so he didn’t know he had this type of the disease.” Had his primary care doctor had a simple retinal camera connected to an AI, or even an AI scanning for constellations of symptoms in a patient’s charts, the patient could have made life choices a decade ago to help him prepare for vision loss.

Both Dr. Cuadros and Dr. Shahbazian are quick to point out that a diagnosis is not the same thing as care. “AI can increase access so everybody can get a diagnosis,” Shahbazian says, “but the care still requires human interaction and empathy and understanding.” She experienced that first-hand when dealing with a family member with glaucoma, which made it even harder to navigate the world given her additional diagnosis of Parkinson’s disease. Not every doctor took the time to consider the additional difficulties posed by the two unrelated diseases, but one in particular took care to extend that human touch and talk to Shahbazian’s loved one about the lived reality of her daily life. “AI could probably have that conversation too,” says Shahbazian, “but what people find valuable is when they feel like they’re actually being taken care of. A diagnosis is where care begins. If you get a diagnosis and that’s the end of it, you’re not being taken care of, you’re just being told something.”

While the progression of serious disease is the focus of many high-level researchers, most people interact with the world of eye care at private practice clinics like Dr. Sarver’s. A computer and the accompanying AI might be able to refine a prescription down to hundredths or thousandths of diopters, so your prescription might not be 1.75, but 1.7492, he says, “but our visual brains are not actually going to be able to detect that.”

The real value of a hands-on clinic like his is the years of relationship and knowledge that come with a long-term doctor/patient relationship. Most optometrists like Sarver will tell you that there’s a bit of art sprinkled into the science of figuring out what a patient needs—a dance that a computer would have a hard time mastering. “Take a patient who is having trouble adapting to their prescription,” Sarver says. “You do the testing and you take a good first stab at it. But then you have to adjust by really listening to the patient, figuring out what they’re sensitive to, what their typical day looks like.” The human touch becomes even more important as patients experience more serious issues. “I do a lot of cataract co-management and not every patient proceeds with cataract surgery at the same pace,” Sarver says. “Some wait to get really bad. Some do it right away, some need a lot of handholding. And you can’t replace that with a computer.”

Sooner or later, technology is coming to change the fields of optometry and ophthalmology, just the way it always has. “Doctors who were trained thirty years ago were really worried that the next generation wouldn’t be able to use the direct ophthalmoscope,” Shahbazian says, because of their increasing reliance on fundus cameras and other scanning technologies. “But as we develop new tools, we’re able to do more. AI is probably going to make us better at some things and worse at some things,” she says. “If clinicians feel like they’re going to become obsolete, they’re just being uncreative.”

Shahbazian does worry that the training optometrists receive isn’t adapting as quickly as the technology. Corporations are already developing diagnostic products that incorporate AI, she says, and she worries that their bottom line will drive how physicians practice, rather than the reverse. “Medical device companies are not really interested in improving health,” she says. “They just want their device to get used a lot so they can make a lot of money. The practice of medicine should really be getting to the bottom of what causes disease, so that it doesn’t happen in the first place.” In addition, like so many medical trials of years past, AI runs the risk of drawing conclusions based on the assumption that the “average” patient is a middle-aged white man. “If we’re going to use AI in healthcare,” she says, “we really need to make sure it’s accurate and fair.”

“The possibilities for AI to improve eye care are truly limitless, but we’re just beginning to imagine the scope.”

The possibilities for AI to improve eye care are truly limitless, but we’re just beginning to imagine the scope. A 2023 study used a deep learning system to very accurately predict the five-year risk of severe myopia in children. Knowing what might be down the road for a young patient actually increases the role for a physician, rather than diminishing it. A doctor armed with an AI prediction can intervene early, suggesting optometric and behavioral changes that will have truly beneficial effects both for the patient and for society at large. Even at the clinic level, some doctors are using AI voice assistants to do large portions of their charting, freeing them up to spend more quality time directly interacting with patients. “I’ve already worn out one elbow with carpal tunnel syndrome,” says Dr. Sarver, “and I’m working on the other. So that sounds pretty interesting!”

Dr. Cuadros has spent decades developing and refining deep learning and information systems to improve eye care. Even with this subject as his life’s work, he doesn’t foresee a future in which computers will replace the intuitive, empathetic skills that are inherent to humanity. Eye care, he says, is still a very human endeavor. But he does echo a clear word of warning for the next generation of eye doctors: “AI is not going to replace clinicians,” he says, “but clinicians who use AI may replace those who don’t.”

Related Information

About Dr. Cuadros About Dr. Shahbazian

About the Photos

1. “AI In Practice” Illustration by Brian Stauffer.
2. Jorge Cuadros, OD, PhD.