by Kelly DeVos, Arizona State University

Credit: Unsplash/CC0 Public Domain

Myopia, also known as nearsightedness, is on the rise, especially among children. Experts predict that by the year 2050, myopia will affect approximately 50% of the world's population. Researchers believe that an increase in what's called "near work"—when we interact with close objects like phones and screens—is partially to blame.

For many people, the struggle to see faraway objects is a problem easily managed with glasses or contacts, but for others this develops into a far more serious condition called myopic maculopathy.

A team of researchers in the School of Computing and Augmented Intelligence at Arizona State University, is developing new diagnostic tools that use the power of artificial intelligence, or AI, to more effectively screen for this disease.

Myopic maculopathy occurs when the part of the eye that helps us see straight ahead in sharp detail is stretched and damaged. Over time, the eye's shape becomes elongated—more like a football and less like a sphere. When this happens, vision is distorted.

This serious condition is the leading cause of severe vision loss or blindness. In 2015, myopic maculopathy resulted in visual impairment in 10 million people. Unless things change, more than 55 million people are predicted to have vision loss and approximately 18 million people worldwide will be blind due to the disease by 2050.

Because myopic maculopathy is irreversible, experts want to intervene early. Catching the condition as soon as possible can improve health outcomes, a particularly urgent goal when children are concerned. Ophthalmologists can prescribe special contact lenses or eye drops that slow the progression of the disease.

Yalin Wang, a Fulton Schools professor of computer science and engineering, says innovations in technology can provide important solutions.

"AI is ushering in a revolution that leverages global knowledge to improve diagnosis accuracy, especially in its earliest stage of the disease," he says. "These advancements will reduce medical costs and improve the quality of life for entire societies."

A challenge to see things in a new way

In response to this need, the Medical Image Computing and Computer Assisted Intervention, or MICCAI, Society issued a challenge in 2023. The professional organization that seeks to drive innovation in biomedical research asked experts to improve computer-aided screening systems for retinal images.

Currently, myopic maculopathy is diagnosed using optical coherence tomography scans that use reflected light to create pictures of the back of the eye. These scans are then often manually inspected by an ophthalmologist, a time-consuming process that can require specialized experience.

Wang and his team in the Geometry Systems Laboratory answered the call. The researchers were one of the winners of the MICCAI challenge.

For the first part of the work, Wang and his team—which includes computer engineering doctoral student Wenhui Zhu as well as neurologist and Fulton Schools adjunct faculty member Dr. Oana Dumitrascu—addressed the classification of myopic maculopathy. The disease has five classifications that describe its severity. Determining the correct level helps ophthalmologists to provide more tailored, effective solutions to the patient.

The Fulton Schools researchers created new AI algorithms called NN-MobileNet. These sets of instructions that computer programs follow to do their work are designed to help software more effectively scan retinal images and predict the correct classification of myopic maculopathy. The research was published in Myopic Maculopathy Analysis.

Next, the team turned their attention to efforts in the scientific community to use a type of AI called deep neural networks to predict the spherical equivalent in retinal scans. The spherical equivalent is an estimate of the eye's refractive error that doctors need when prescribing glasses or contacts. In deep neural networks, researchers task computers with analyzing huge sets of data and apply AI-powered algorithms to draw helpful conclusions.

With a more accurate measure of the spherical equivalent, doctors can make more accurate treatment recommendations. So, Wang and the team again developed new algorithms that focused on data quality and relevance. Their new model of retinal image analysis achieved exceptional results while minimizing the amount of computing power needed. The results of this research were also published in Myopic Maculopathy Analysis.

Finally, Wang collaborated with other winning teams from the MICCAI challenge on a third research paper, published in JAMA Ophthalmology in September, that presented their collected results. The researchers from universities around the world made their challenge findings available to stimulate additional advancements and discoveries in the early and effective diagnosis of myopic maculopathy and improving health care outcomes for people across the globe.

A better vision for global health

Wang explains that one motivating force behind his work is to solve health disparities.

"People living in rural areas find it difficult to access sophisticated imaging devices and see physicians," he says. "Once AI-powered technology becomes available, it will significantly improve the quality of life in worldwide populations, including those who live in developing countries."

Ross Maciejewski, director of the School of Computing and Augmented Intelligence, says Wang's project is an important example of the excellent work being done by faculty members in the medical space.

"With both myopia and myopic maculopathy increasing, solutions are needed to prevent vision loss and help health care professionals provide the best treatment options for their patients," Maciejewski says. "Yalin Wang's innovative research is a principled use of artificial intelligence to address this dire medical issue."

More information: Wenhui Zhu et al, Beyond MobileNet: An Improved MobileNet for Retinal Diseases, Myopic Maculopathy Analysis (2024). DOI: 10.1007/978-3-031-54857-4_5

Huayu Li et al, Prediction of Spherical Equivalent with Vanilla ResNet, Myopic Maculopathy Analysis (2024). DOI: 10.1007/978-3-031-54857-4_6

Bo Qian et al, A Competition for the Diagnosis of Myopic Maculopathy by Artificial Intelligence Algorithms, JAMA Ophthalmology (2024). DOI: 10.1001/jamaophthalmol.2024.3707

Journal information: JAMA Ophthalmology 

Provided by Arizona State University