Artificial intelligence (AI) technology can accurately and independently detect 100% of severe cases of a blindness-causing condition that affects prematurely born babies, according to new research out of the United States.

Researchers from Oregon Health and Science University (OHSU) and collaborators, in an article published in JAMA Ophthalmology,1 said the technology has the potential to expand worldwide screening – and ultimately sight-saving treatment – for retinopathy of prematurity (ROP).

About two million babies annually are born early enough to develop ROP, although in most cases the disease is mild and resolves without treatment. Severe cases cause about 50,000 babies globally to go blind every year. More cases go untreated in low- and middle-income countries, where there are fewer ophthalmologists to examine and treat premature babies.

The study’s corresponding author, Dr J. Peter Campbell, Associate Professor of Ophthalmology at OHSU, said although ROP could not be fully prevented, “we can almost always prevent blindness from ROP”.

“While there are doctors who are skilled in ROP treatment in many parts of the world, there simply aren’t enough to screen all of the babies who are at risk,” Dr Campbell continued. “This paper demonstrates that AI can effectively replace the physician for bedside screening and refer the most urgent cases to a physician for treatment.”

REAL WORLD DETECTION

When he was still at OHSU, Dr Michael Chiang, now director of the National Eye Institute of the National Institutes of Health, and colleagues first developed the i-ROP Deep Learning system, which uses an AI algorithm to identify blood vessel anomalies in retinal images. Currently, specially trained ophthalmologists must manually review these images to diagnose ROP.

The team’s earlier research showed that its AI technology could accurately diagnose ROP and can also be effectively used remotely through telemedicine appointments instead of traditional, in-person eye exams.

This new study marks the first time that autonomous AI screening for ROP has been shown to work in a real-world population, meaning the technology correctly flags the condition on its own, without ophthalmologist support and without preselecting images to improve data quality.

While many AI algorithms work in controlled experiments, they often fail to work in the real world due to differences between training data and real-world use, the study authors said.

For this study, the AI system analysed nearly 12,000 images of more than 4,000 babies’ retinas. The photos were taken by nurses at neonatal intensive care units in United States and Indian

hospitals. Ophthalmologists had previously reviewed the images as part of telemedicine programs in both countries and found that about 1.2% of the babies had severe forms of ROP, while about 5.8% had more-than-mild cases. The AI system correctly identified all of the severe cases and accurately detected 80% of the cases with more-than-mild ROP.

i-ROP DEEP LEARNING SYSTEM

The i-ROP Deep Learning system received “breakthrough” status by the United States Food and Drug Administration in 2020 to accelerate its development.

In addition to his role at OHSU, Dr Campbell is also chief executive officer of Siloam Vision, which licensed the technology and is currently leading clinical trials to evaluate the AI system’s safety and effectiveness. Siloam Vision has also partnered with the non-profit Orbis International to implement the technology in low- and middle-income countries.

If the technology is ultimately approved by regulatory agencies, ROP would become the second eye disease that can be independently detected by AI. Diabetic retinopathy can already be detected with three autonomous AI devices that are approved for use in the US.

Reference 1. Coyner, A.S., Murickan, T., Campbell, J.P. et al., Multinational external validation of autonomous retinopathy of prematurity screening, JAMA Ophthalmology, March 7, 2024, DOI:10.1001/ jamaophthalmol.2024.0045.