Artificial Intelligence Passes Macular Degeneration Test

Laird Harrison

June 29, 2020

A new artificial intelligence system — iPredict.Health from iHealthScreen — can accurately screen patients for age-related macular degeneration (AMD), results from a prospective clinical trial suggest.

"We are delighted to say that the tool is now ready for clinical use and potential remote telemedical deployment," said Sharmina Alauddin, MBBS, from New York Eye and Ear Infirmary of Mount Sinai Hospital in New York City.

"If we are able to diagnose AMD by screening in any clinical setting, we could start treating early and reduce blindness," said Alauddin, who presented the study findings at the virtual Association for Research in Vision and Ophthalmology 2020 Annual Meeting.

Most people with AMD don't become aware that they are affected until their vision starts to become compromised, and there are not enough retina specialists to screen everyone at risk for the disease. But with artificial intelligence that can screen fundus images, primary care practitioners and optometrists might be able to make a preliminary diagnosis and refer appropriate patients.

Multiple teams around the world are working to develop such systems. Most use a large database of images already classified as showing the absence or presence of AMD, and the disease stage, to look for patterns that distinguish the different types of images. The Age-Related Eye Disease Study (AREDS) dataset — containing 150,000 images — is one of the most widely used.

For their study, Alauddin and her colleagues used a deep-learning system trained to simulate the neural networks of the human brain to recognize patterns. They previously tested the system with the AREDS dataset and determined that it could identify AMD with 95.3% accuracy and could classify disease stage with 86.0% accuracy, which is comparable to rates achieved by human retina specialists.

To see whether the system could identify the disease in a real-life population, the researchers used a fundus camera to image both eyes of 150 unselected nondilated patients older than 50 years at New York Eye and Ear. All had high-quality images available and none had a confounding condition, such as diabetic retinopathy, macular edema, and previous retinal surgery.

The iHealth system classified 66 patients — on the basis of the worst eye — as referable for AMD because of intermediate or late AMD and 84 as nonreferable because of normal macula or early AMD.

When the system results were compared with readings from two ophthalmologists, accuracy was 88.67%; sensitivity, indicating true positives, was 86.57%; and specificity, indicating true negatives, was 90.36%.

"Very Promising"

That accuracy is impressive, said Emma Pead, PhD, from the University of Edinburgh in the United Kingdom, who has been working on a similar system.

"I was not aware of anyone actually using [artificial intelligence] in newly acquired images," she told Medscape Medical News. "It's very promising. The only way to see if these systems work is to test them out in a clinic."

Because of these encouraging results, Mount Sinai Hospital plans to place fundus cameras in primary care clinics so that patients can be screened for AMD, said Theodore Smith, MD, PhD, one of the study researchers. "Based on those reads, we're going to provide care to patients," he said.

The researchers have also used the system to predict which patients will progress to late AMD, and whether the disease will take the wet or dry form in these patients. They plan to follow the patients and test the accuracy of these predictions.

Alauddin, Smith, and Pead have disclosed no relevant financial relationships.

Association for Research in Vision and Ophthalmology (ARVO) 2020 Annual Meeting.

Follow Medscape on Facebook, Twitter, Instagram, and YouTube


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.