AI Detects Sight-robbing Condition in Babies

For premature infants, this early detection can save their sight

By Deborah Jeanne Sergeant

Physician Luca Zatreanu, chairman of ophthalmology at Rochester Regional Health: “There are approximately 500 infants annually in the US that can lose vision from retinopathy.” he says. “It’s very treatable, but you have to screen and detect it.”

Premature babies and their families face myriad challenges as they navigate a world for which they’re not yet ready.

One area of development that could use medical intervention is vision. An ophthalmologist typically examines the eyes of premature babies in the neonatal intensive care unit to look for issues such as retinopathy of prematurity (ROP).

“It is important to examine the retinas of premature babies to protect their vision and prevent abnormal retinal blood vessel development, which can lead to reduced vision and blindness,” said physician Olachi J. Mezu-Ndubuisi, an associate professor of pediatrics and ophthalmology at University of Rochester.

Preemies have immature retinas that lack all of the blood vessels full-term babies have. Ideally, these blood vessels develop as they should so they can nourish the photoreceptors that carry visual information to the brain. When these blood vessels do not develop properly, it can cause blindness. Fortunately, if discovered early, this problem can be corrected with laser surgery or injections. But detection is key. Artificial intelligence has provided tools to help providers detect at-risk babies more accurately and efficiently.

“Artificial intelligence in ROP would involve using algorithms or codes to teach a computer to quickly differentiate between abnormal and normal retinas,” Mezu-Ndubuisi explained. “AI could be a very useful screening tool for ROP. Advances in perinatal and neonatal care have led to increased survival of extremely premature infants at earlier gestational ages. Therefore, ROP remains the leading cause of preventable childhood blindness worldwide despite advances in neonatal care.”

AI can help ease the burden of ROP exams on the few available ophthalmologists, especially in rural areas.

Mezu-Ndubuisi added that AI exams are more objective than the human assessment of how abnormal blood vessels look.

“This assessment can vary between different observers,” she added. “Significant differences in diagnosis can lead to delays in treatment of ROP which can pose a threat to vision.

“Artificial intelligence is very promising but needs to be optimized to more reliably use information about the individual baby’s condition or risk factors for ROP to accurately guide diagnosis, treatment and monitoring of infants with ROP.”

Matt Gearinger, pediatric ophthalmologist at Flaum Eye Institute, does rounds at URMC’s NICU to examine babies’ eyes. He said that AI exams can help identify which babies are at risk and need treatment, which “can cut the workload extensively” for rural areas not well served by ophthalmology.

“They don’t have enough people who can do the eye exams,” he said. “Babies born at 30 weeks or earlier or who have low birth weight get screened automatically.”

Low birth weight is 1,500 grams (or about 3.3 pounds) or less.

Physician Luca Zatreanu, chair of ophthalmology at Rochester Regional Health, said that he also examines babies who have experienced an unstable clinical course, like bleeding in the brain, supplemental oxygen in the first week of life and a long NICU stay.

“There are approximately 500 infants annually in the US that can lose vision from retinopathy,” he said. “It’s very treatable, but you have to screen and detect it.”

Currently, AI is capable of identifying with 100% accuracy the most severe cases of ROP and 80% accuracy for those later deemed “more than mild.”

“That’s huge,” Zatreanu said. “It removes humans from the equation from the screening except form providing the images. Someone still needs to take pictures of the babies’ eyes in the NICU and feed them into the platform. It would automatically say ROP is present. You can funnel these babies to the appropriate provider.

“It’s a challenge. We definitely need people to help with screenings with preemies. It’s not easy to find these kinds of doctors so having this as an opportunity in the near future would be huge for us. If we had portable cameras and trained personnel on how to take the photos and upload the photos, that would be a huge help.”