USC Researchers to Leverage Ophthalmology AI for Improved Patient Care

Researchers at the University of Southern California (USC) Roski Eye Institute are exploring how the use of artificial intelligence (AI) in ophthalmology may help automate clinical tasks, allocate medical resources more efficiently, and improve care quality.

Ongoing healthcare workforce shortages are likely to continue in the United States over the next several years, and aging populations are expected to face significant provider shortages in the future. The press release indicates that this will be a particular challenge in ophthalmology, as these shortages can lead to patient care access challenges.

Successful use of AI in this area could help prevent and address these issues, spurring USC researchers to investigate these tools.

The research focuses on automating glaucoma detection and differentiating between serious and benign eye conditions using AI analysis of fundus photos.

The glaucoma detection initiative aims to use AI and fundus images to help automate the detection and referral of glaucoma patients from the Los Angeles County (LAC) Department of Health Services (DHS). The project seeks to improve access for LAC DHS’s underserved populations.

“Glaucoma is endemic among patients of LAC DHS, where it commonly leads to permanent, severe vision loss. The wait for a glaucoma evaluation typically takes more than six months, even when patients have severe disease,” explained Benjamin Xu, MD, PhD, an assistant professor of clinical ophthalmology at the Keck School of Medicine of USC, who leads the research, in the press release. “This is a significant problem because valuable time is lost when these patients could have received sight-saving treatment.”

The initiative builds on Xu’s research to develop AI to help analyze optical coherence tomography (OCT) images and identify patients at high risk for glaucoma.

Xu indicated that OCT can provide high-quality images for glaucoma detection, but OCT analysis requires both time and specialized expertise, which clinicians may lack. Xu’s team hopes to develop AI tools that can simplify this process.

The researchers were recently awarded a two-year grant from LAC DHS and the Southern California Clinical and Translation Science Institute (SC CTSI), which will support the development of AI solutions for deployment in LAC DHS glaucoma screening clinics.

The second USC initiative to leverage AI in ophthalmology is looking at using algorithms to help distinguish between papilledema and pseudopapilledema.

Cleveland Clinic reports that papilledema is a serious condition that occurs when the optic discs in the eyes swell. Papilledema can be linked to neurological issues, such as brain abnormalities or tumors, and requires immediate treatment.

Conversely, pseudopapilledema is a benign condition, but telling the difference between the two can present a challenge for clinicians.

“Identifying the exact diagnosis typically requires longitudinal patient follow-ups and testing, including MRI scans and lumbar punctures,” said Melinda Chang, MD, an assistant professor of clinical ophthalmology specializing in pediatric neuro-ophthalmology at Keck. “We aim to have an answer earlier, and that’s where AI steps in.”

Chang’s team is collaborating with researchers from Boston Children’s Hospital, Stanford, Children's Hospital Los Angeles, and the University of California, Los Angeles (UCLA) in a multi-center study to provide fundus images for an AI tool designed to differentiate papilledema and pseudopapilledema.

The press release indicates that the model has outperformed humans in this task so far, with a current accuracy ranging between 70 and 80 percent and a sensitivity of up to 90 percent.

Moving forward, the researchers aim to expand the study to more institutions to gain additional data. The team also seeks to investigate the integration of imaging techniques like OCT to improve the model.

Across both initiatives, the researchers emphasized that ethical concerns and health equity considerations are key to the success of their research.

“At present, our AI algorithm is developed using data from Los Angeles County Department of Health Services, where the majority of patients are Latino,” said Xu. “While our algorithm may work effectively for Latino individuals, its performance may degrade when applied to individuals of other races and ethnicities. There should always be concern regarding potential biases that are latent due to unique attributes of data sources used to develop AI algorithms.”

“That is why we have established partnerships with other medical sites nationwide and aspire to collaborate with even more to ensure a more diverse sample,” Chang indicated.

These initiatives underscore a growing interest in using AI to improve ophthalmology.

In 2019, researchers at Duke University's Pratt School of Engineering developed a machine learning (ML) algorithm that enhances the resolution of OCT.

The research team noted that OCT typically has better depth resolution than lateral resolution, resulting in blurred images of irregularly shaped structures. To address this, the researchers designed an ML tool capable of creating maps of the way light bends as it passes through a model.

The application of this approach to OCT has the potential to benefit oncology, cardiology, and ophthalmology.