May 4 (UPI) -- Artificial intelligence was more accurate than medical experts at diagnosing a devastating cause of blindness in premature babies in a recent study, researchers report.
Researchers at the Oregon Health & Science University and Massachusetts General Hospital studied how an algorithm achieved 91 percent accuracy at diagnosing retinopathy of prematurity in images of infant's eyes. Their findings were published this week in the Journal of American Medical Association Ophthalmology.
Retinopathy of prematurity, or ROP, is a condition in babies born before 31 weeks of gestation and is caused by abnormal blood vessel growth near the retina, the light-sensitive portion in the back of an eye. ROP is the leading cause of blindness in children.
Up to 16,000 U.S. babies are diagnosed with ROP, but roughly 600 become legally blind each year from it, according to the National Eye Institute.
"There's a huge shortage of ophthalmologists who are trained and willing to diagnose ROP," Dr. Michael Chiang, a professor of ophthalmology and medical informatics & clinical epidemiology in the OHSU School of Medicine and a pediatric ophthalmologist at the school's clinic, said in a press release. "This creates enormous gaps in care, even in the United States, and sadly leads too many children around the world to go undiagnosed."
To diagnose the condition, physicians usually use a magnifying device that shines light into a baby's dilated eye. The method, however, can be inaccurate as it is a subjective examination.
In the study, a team of eight physicians with ROP expertise correctly analyzed 82 percent of the same images -- 9 percentage points less than the AI method. Data were collected from July 2011 to December 2016.
"This algorithm distills the knowledge of ophthalmologists who are skilled at identifying ROP and puts it into a mathematical model so clinicians who may not have that same wealth of experience can still help babies receive a timely, accurate diagnosis," said Dr. Jayashree Kalpathy-Cramer, of the Center for Biomedical Imaging at Massachusetts General Hospital and an associate professor of radiology at Harvard Medical School.
Last month, the Food and Drug Administration approved an AI device that detects diabetic retinopathy by mimicking how humans perceive the world through vision, including identifying objects. In a clinical study, IDx-DR correctly identified the presence of more than mild diabetic retinopathy 87.4 percent of the time, and those without less mild diabetic retinopathy 89.5 percent of the time.
For the new study on ROP, Massachusetts General Hospital researchers combined two existing AI models to create the algorithm. OHSU researchers developed standards to train it.
The AI computers were trained to examine 5,511 pictures from infant visits to an ophthalmologist. Then, they were trained to tell the differences between healthy and diseased vessels. Researchers then compared the algorithm's accuracy with the evaluation of trained experts who viewed the same images.
Now, the researchers are working with a collaborator in India to see if the algorithm can diagnose ROP in Indian babies as they did for the primarily Caucasian babies studied by OHSU and MGH. Additionally, they are exploring whether the algorithm can diagnose the condition other parts of the retina besides vessels.
"These results may change the way ROP is diagnosed in the future and are broadly relevant to other medical fields that rely primarily on subjective image-based diagnostic features," the researchers wrote in the study.