A research team at Asan Medical Center (AMC) has developed an artificial intelligence model that can identify glaucoma patients who have a high risk of going blind.

Glaucoma, caused by increased intraocular pressure with age and damage to the optic nerve, is incurable. The only treatment is to slow the progression as much as possible.

"It is crucial to detect at an early stage and prevent glaucoma from getting as severe as possible," the hospital said. "However, even if detected and treated early, glaucoma can still progress, making it hard for doctors to select high-risk patients accurately."

Until now, ophthalmologists had no choice but to predict the progression of glaucoma by conducting periodic visual field examinations at intervals of about six months based on clinical experience.

To resolve this problem, the team, led by Professors Sung Kyung-rim and Shin Joong-won of the Department of Ophthalmology, developed the model after training the AI with 96,542 visual field test results through artificial neural network technology. Professor Son Gil-hwan at Gangneung Asan Hospital also participated in the study.

To further increase the accuracy, the team used a convolutional neural network, which stacked the results of the patient's three consecutive visual field tests to teach the AI model.

While the AI model showed a sensitivity of about 42 percent, the team managed to confirm a high specificity of about 95 percent and confirmed high-risk glaucoma patients with an overall accuracy of about 86 percent.

"The visual field test is a test performed to diagnose glaucoma, but due to the nature of the test, it is inconvenient for patients as they have to undergo regular examinations for a long time to diagnose whether or not glaucoma has progressed," Professor Sung said. "Our AI ​​model diagnoses high-risk glaucoma, which can cause blindness, at an early stage with just three visual field tests at intervals of about six months."

The is excellent as doctors can establish the optimal treatment direction, such as additional drug treatment or surgery, more effectively, Sung added.

Professor Shin also said, "We plan to improve further the accuracy of the high-risk glaucoma diagnosis model using artificial intelligence to minimize the possibility of vision loss due to glaucoma."

The results of the research were published in the latest issue of the American Journal of Ophthalmology.