Computers Found to Equal Humans in Diagnosing Age-Related Macular Degeneration

Computers can do as well as clinicians in diagnosing age-related macular degeneration, say investigators who have developed tools based on artificial intelligence (AI) to detect the disorder.

“AI and deep learning can help diagnose age-related macular degeneration with accuracy on par with human retinal specialists,” Philippe Burlina, PhD, who leads machine intelligence at the Applied Physics Laboratory of Johns Hopkins University, told MD Magazine®.

Machine-learning techniques, in which scientists provide data and feedback to train computers to act and learn as humans do, could address challenges to the detection and management of age-related macular degeneration, researchers say.

Such hurdles include the cost of screening patients, providing access to health care, and developing novel treatments for the disorder, a team from the Applied Physics Lab working with colleagues the Johns Hopkins School of Medicine wrote in a study published in JAMA Ophthalmology.

“Earlier referrals to a retinal specialist for at-risk individuals can help improve outcome with regard to loss of vision in some cases,” said Burlina, a co-principal investigator for the project.

Age-related macular degeneration, in which damage to the macula of the retina limits a person’s central vision, is the leading cause of blindness in adults over age 50. Yet the intermediate stage of the disease may go undetected because a patient usually has no symptoms at that point.

The ability to identify individuals at this stage could help clinicians monitor disease progression before substantial vision loss has occurred. The patient also might consider dietary supplements that could reduce the risk of progression, the team wrote.

“It is important to identify age-related macular degeneration early to monitor progression,” Burlina said. “But it is particularly important to catch age-related macular degeneration in the intermediate stage when there might not be loss of visual acuity yet and treatment may be advised in the case of neovascular age-related macular degeneration.”

The investigators set out to see how computer-based deep learning programs might perform in detecting and grading age-related macular degeneration.

They accessed more than 130,000 color images of the fundus, the interior section of the eye that includes the retina, from 4,613 participants in the National Institutes of Health’s Age-Related Eye Disease Study (AREDS).

The team then unleashed computers and a deep convolutional neural network (DCNN) to evaluate the images. The deep learning algorithm repeatedly processed the fundus images with many operations to yield a designation for each image.

“One way to think about DCNNs is that they match the input image with successive convolutional filters to generate low- mid- and high-level representations of the input images,” the authors wrote.

The investigators compared the computer results to those in which a human grader had been involved.

The DCNN method ranged in accuracy from 88.4% to 91.6%, with kappa scores close to or greater than 0.8. “This is comparable with human expert performance levels,” the investigators said.

What did they find? “New DL algorithms can perform a screening function that has clinical relevance with results similar to human performance levels to help find individuals that likely should be referred to an ophthalmologist in the management of age-related macular degeneration,” the team said.

The investigators are expanding their work to evaluate the layers of the retina using a technique called optical coherence tomography (OCT). This method provides high-resolution, cross-sectional images of the retina, retinal nerve fiber layer and optic nerve head.

The noninvasive approach can be used to diagnose other retinal diseases such as diabetic retinopathy. It also has the potential to help characterize vascular and neurodegenerative pathologies, the investigators say.

“Fundus imaging is widely available, but OCT is gaining prevalence in clinical practices,” Burlina said. “It can help in some cases to better characterize age-related macular degeneration.”

The technique is also key to deciding a course of treatment for neovascular age-related macular degeneration using anti-vascular endothelial growth-factor (anti-VEGF) drugs, he said.

In this type of so-called “wet” age-related macular degeneration, blood vessels grow abnormally under the retina. Anti-VEGF treatments can reduce swelling and the growth of new blood vessels, research suggests.