AI Revolution

People think of their eyes as windows onto the world, but the physicians and professionals at the Byers Eye Institute at Stanford know they offer a window into our health. It’s that quality that allows ophthalmology to be at the forefront of artificial intelligence (AI) advances. The AI programs developed at the Byers Eye Institute are already improving access to health care and helping doctors diagnose and predict patient outcomes. 

One of the earliest and largest AI-enabled initiatives spearheaded by faculty at the Byers Eye Institute is the Stanford Teleophthalmology Autonomous Testing and Universal Screening (STATUS) program, which helps patients with diabetes get an annual eye exam using AI-enabled cameras without scheduling a separate appointment with an ophthalmologist. 

“In ophthalmology, many clinical decisions are based on some sort of image, where other fields are more apt to use lab results,” said David Myung, MD, PhD, associate professor of ophthalmology and founding director of the STATUS program. “An amazing amount of research and development has gone toward using AI to read ocular images, which led to the first-ever FDA approval for AI-based disease detection.” 

Today, 10 AI-enabled STATUS cameras are spread across Bay Area primary care clinics affiliated with Stanford. More than 3,000 patients, about 100 each month, have had a photo of their eye screened by the AI system. The results show promise in combating disease and saving vision, says Theodore Leng, MD, MS, associate professor of ophthalmology and a retina specialist.

“Being able to detect disease earlier and intervene when we can actually turn things around and preserve vision is what is really important,” Leng said. 

As AI technology progresses, eye images could also yield insights into patients’ overall health and help detect diseases like Parkinson’s or Alzheimer’s diseases, and hypertension, Myung added.

Improving care for diabetes

The STATUS program grew out of an institution-wide quality improvement initiative that aimed to improve eye care for people with diabetes, which affects 400 million people worldwide. Diabetic retinopathy is the leading cause of blindness in working-age adults. Even so, many don’t make it to their critically important annual eye exam.  

In 2018, Myung launched an effort to offer remote exams in primary care clinics to close the screening gap for diabetic patients, and he thought ahead: The FDA had just approved the first AI system that could be used to analyze images for diabetic eye disease, so he made sure the cameras had that technology.

In addition to Leng, Byers Eye Institute retina specialists Diana Do, MD; Prithvi Mruthyunjaya, MD, MHS; Vinit Mahajan, MD, PhD; Christopher Or, MD; and Darius Moshfeghi, MD, have all been critical to the growth and success of the program, Myung says. 

The STATUS program added a key innovation of AI-MD assistive interaction: when needed, the system pushes the images to a team led by Leng at the Stanford Ophthalmic Reading Center (STARC) for a closer look. Those who need to see an ophthalmologist get a notification within a day.

Early results show that people are more likely to schedule follow-up visits, possibly because patients find out about their eyes much sooner than they could in the past, Myung said. 

Vision testing at home

Another research project is opening the possibility of vision testing at home. Chris Piech, PhD, assistant professor of computer science, received a diagnosis of uveitis, or intraocular inflammation, when he was 8 years old, and has spent his life battling eye complications. 

After undergoing cataract surgery at Byers Eye Institute by Charles Lin, MD, clinical associate professor of ophthalmology, Piech asked Lin to collaborate on a project to improve the traditional eye chart, called the Snellen vision test. Robert Chang, MD, associate professor of ophthalmology, joined the collaboration, as did Ali Malik, a PhD computer science student in Piech’s lab, and Laura Scott, Piech’s wife. 

Their collaboration formed the basis for the Stanford Acuity Test, an online vision test driven by AI, which can now be accessed at www.myeyes.ai. The test doesn’t replace an eye exam but could help lay the foundation for home vision testing.

Transitioning diagnostics from the clinic into the home may offer major advantages for detecting disease or the progression of diseases. 

For example, new technologies could allow some peripheral vision testing for glaucoma or for other 

neuro-ophthalmic diseases to be done remotely, reducing burden on patients and clinics, and improving diagnostic accuracy by testing more often. 

“This is the forefront of next-generation healthcare,” Jeffrey Goldberg, MD, PhD, the Blumenkranz Smead professor and chair of ophthalmology, said.

Big data and algorithms

But any scientist can attest that artificial intelligence systems are only as good as the data that goes into them. At the Mary M. and Sash A. Spencer Center for Vision Research at the Byers Eye Institute, researchers are using new approaches to ensure AI systems are trained on high-quality data. 

Leng and Daniel Rubin, MD, MS, professor of biomedical data science, radiology, and medicine, and, by courtesy, of ophthalmology, have developed algorithms that can predict which patients with dry age-related macular degeneration (AMD) will likely progress to wet AMD, an explosive, damaging form of the disease.

Similarly, Sophia Wang, MD, assistant professor of ophthalmology, has trained an algorithm to improve how doctor’s notes and other electronic health information are analyzed. 

She and her team at the Byers Eye Institute are using AI to help predict which glaucoma patients need surgery and which patients would do well with less invasive treatment options.

Wang is collaborating with Chang, Tina Hernandez-Boussard, MD, PhD, MPH, associate professor in medicine (biomedical informatics), and 
Suzann Pershing, MD, associate professor of ophthalmology and of health research and policy. 

A study that Wang authored and published this year in the peer-reviewed journal, Frontiers in Medicine, shows this new technology can tell with higher accuracy than most humans which patients need what treatment. But Wang’s not stopping there; she and her colleagues are continuing to tweak her model to become even more accurate. 

“I think of AI as comparable to other tools,” Myung said. “We used to use a hand screwdriver for everything. Then someone invented an electronic screwdriver. So far, AI is a force multiplier.”