3 Ways Artificial Intelligence Will Affect Your Eye Care Practice

Artificial Intelligence

Artificial Intelligence. It sounds so futuristic, doesn’t it? Like it takes place in some far-off lab in the Silicon Valleys of the world—not on Main Street. But that’s a huge misconception. Artificial intelligence (AI) already plays a role in many of our daily activities, and it may (soon play an even bigger role in how providers do their jobs—especially in the eye care specialty.

If you’re a little apprehensive about what this means for you and your eye care practice, you’re not alone. (AI is writing news articles now, so I’m right there with you.) So let’s address that elephant in the room: You’re not going to be replaced by a robot doctor. Ophthalmologists shouldn’t worry that AI will make their work less relevant or replace them, urges Robert Chang, MD, assistant professor of ophthalmology at Stanford University Medical Center, who briefed the press on AI’s recent advances in eye care at AAO’s 2018 annual meeting.

AI’s purpose “is not to eliminate the ophthalmologist, but to enhance and increase access for patients,” Dr. Chang explained. Eye care practices shouldn’t fear the technology, but rather, embrace it as a way to make their practices more efficient and effective. As the repetitive tasks you perform are eliminated, you’ll be able to focus more on doing the things that AI can’t do—like providing personal care to your patients and building relationships.

Artificial Intelligence: An Easy Primer

What is artificial intelligence? Pop culture often personifies AI, portraying it as a robot that appears suspiciously human. That’s far from the reality, and more dramatic, too. AI is simply an algorithm—basically a set of rules that a computer follows to solve a problem or perform tasks that would normally require human intelligence. And it’s not futuristic at all. In fact, you probably use some form of AI every day. For example:

  • Voice recognition (Hey there, Alexa).
  • Personalization/curation of your social media feeds.
  • Options presented to you on Netflix, Spotify, or YouTube.
  • Online ads that seem to (creepily) read your mind.
  • Navigation maps like Waze or Google Maps
  • Customer service chatbots

Another buzzword surrounding AI is machine learning. This term is sometimes used interchangeably with AI, but there’s a difference. Machine learning is really a way to power artificial intelligence. It involves the creation of algorithms that can learn on their own, without human involvement. The algorithm is “trained” using extremely large data sets. Eventually, it learns to apply its observations to new data sets that it’s never “seen” before. The algorithm can then make decisions or predictions on its own.

A very relevant example is the current use of AI in diagnosing diabetic retinopathy. An algorithm is fed large data sets consisting of fundus or OCT photos where diabetic retinopathy is present. Eventually, the algorithm learns what disease characteristics to look for, and can diagnose the disease in new photos.

Will Artificial Intellegence Be Your Partner in Screening?

Over the next few years, artificial intelligence (AI) will revolutionize eye care—and screening for diabetic retinopathy and diabetic macular edema is the first area to benefit from AI advances. The hope is that AI will help the industry overcome the challenges of limited resources by paring down the number of patients who actually need to see an ophthalmologist.

Take DR screenings, for example. Currently, an estimated 40 percent of patients with diabetes don’t get annual screenings for DR. And, a substantial percentage of people with diabetes who are sent to ophthalmology offices for DR screening don’t have the condition. Given the shortage of ophthalmologists and retina specialists, it’s somewhat inefficient to waste chair time on such all-clear visits. If primary care doctors, optometrists, or endocrinologists could screen for DR in their offices, they could screen a higher percentage of patients with diabetes and the patients they did send to ophthalmology practices would be far more likely to actually have DR.

Last year, the Food and Drug Administration approved technology that can do just that—the IDx-DR, which relies on fundus photos. Because the technology employs “deep learning” in collaboration with retina specialists, it potentially brings the retina specialist’s diagnostic expertise directly into primary care settings, Dr. Chang explained.

Will AI Increase Accessibility?

Along with telemedicine, AI screenings are often framed as a solution for rural areas. Rural areas are ripe for rolling out AI-assisted DR screening, observed William Pavia, PhD, executive director at Oklahoma State University’s Center for Health Systems Innovation and speaker at the Machine Learning and AI Summit at HIMSS 2019. There is an acute shortage of specialists in rural areas, which reduces “turf issues” one might find in urban areas, Pavia said. And there is a great unmet need: 90 percent of patients with diabetes in rural areas aren’t getting DR screenings, he said. His goal is to make OSU “the Silicon Valley of rural and Native American health innovation and predictive analytics.”

AI can speed up research. For example, in a current study at Massachusetts Eye and ear, researchers are looking for associations between photographs of patients whose AMD worsened, to try to draw out commonalities. AI could help process many more images must faster than a human.

Will Artificial Intelligence Live Up to the Hype?

Optimistic ophthalmologists anticipate that AI will allow them to see more patients as the population ages, screen more people to detect eye disease sooner, ensure the right people get care at the right time, and offer a solution to the current shortage of ophthalmologists. However, there are practical concerns that limit the technology’s potential.

Many physicians are concerned that AI will eventually cause job loss, especially in areas that are heavy on image reading, like radiology. But those jobs won’t truly be lost, experts say. They’ll be replaced by other, more highly skilled jobs. “The key skill is adaptability,” Dr. Chang added. “As repetitive low skill jobs are replaced, many new jobs will require high skill.”

Another potential hurdle is the cost of the equipment required—and not just for the equipment. Providers will need to consider how an AI solution will fit into their IT infrastructure, as well as their workflow. But there will be pay-for-performance measures that incentivize physicians in primary care settings to purchase and use the technology, observes Vinay A. Shah, MD, clinical professor at Dean McGee Eye Institute at the University of Oklahoma, who presented at AAO alongside Chang. Plus, the equipment will be less expensive—and more productive—than an additional human staff member.

Here are a few other caveats:

  • An AI algorithm only looks for what you tell it to look for. Therefore, it might miss things that a human would notice.
  • If a variable changes, the algorithm will have to re-learn. For example, the IDx AI system uses images taken with a Topcon retinal camera. Change the camera, and the algorithm may no longer recognize the images.
  • AI algorithms have been described as a “black box,” because it’s sometimes impossible to “see inside.” Researchers sometimes can’t explain why one algorithm works when another doesn’t.

Other Applications for Artificial Intelligence

Diabetic retinopathy and DME aren’t the only conditions that AI will screen for. Here are some other applications:

  • Retinopathy of Prematurity (ROP)
  • Macular Degeneration
  • Glaucoma (although the body of research is thin compared to retinal research)
  • Identifying refractive error from fundus photos (to speed up the physician’s refraction)

Recent Ophthalmology & AI Highlights

December 2016: Research published in JAMA demonstrates that Google’s deep learning algorithm, when trained on a large set of fundus images, can detect diabetic retinopathy with more than 90 percent accuracy.

April 2018: IDx becomes the first FDA-approved, autonomous AI system for detecting DR in primary care settings. The technology was developed by Michael Abramoff, MD, PhD, who first became concerned about poor access to DR screenings as an ophthalmology resident 20 years ago.

August 2018: A study published in Nature demonstrates that AI read OCT scans and identify 50 different eye diseases as accurately as ophthalmologists. More importantly, AI can explain to clinicians how it reached its conclusions, thus shedding light on AI’s “black box” problem.

April 2019: A study by Google’s AI research group and published in AAO’s Ophthalmology journal shows that physicians and algorithms working together diagnose DR more accurately than either alone.

 

Leave a Reply

Your email address will not be published. Required fields are marked *