Seoul, Nov 5: Artificial intelligence (AI) training in the identification of warning signs within retinal images holds the potential to predict the development of glaucoma, the leading cause of irreversible blindness worldwide, according to a recent study.
Detecting glaucoma, especially in cases where individuals exhibit early signs of optic nerve damage without the hallmark of elevated intraocular pressure (IOP), presents a significant challenge for healthcare professionals. Researchers from Seoul National University Hospital in South Korea emphasized that without the presence of elevated IOP, it becomes challenging to predict who will eventually develop glaucoma and face the risk of sight loss.
Recent advancements in AI have led to the development of algorithms aimed at better detecting glaucoma progression. However, none of these algorithms have previously incorporated clinical data to predict disease progression among individuals at high risk.
Professor Ki Ho Park, the corresponding author from the university’s Department of Ophthalmology, explained, “Our results suggest that [deep learning] models that have been trained on both ocular images and clinical data have the potential to predict disease progression in [glaucoma suspect] patients. We believe that with additional training and testing on a larger dataset, our [deep learning] models can be made even better, and that with such models, clinicians would be better equipped to predict individual [glaucoma suspect] patients’ respective disease courses.”
To explore the use of AI in bridging this diagnostic gap, the research team examined clinical data from 12,458 eyes showing initial signs of glaucoma. They narrowed their focus to 210 eyes that eventually progressed to glaucoma and 105 that did not, all of which had been under observation every 6-12 months for at least seven years.
The researchers used specific indicators in retinal images obtained during these monitoring periods, along with 15 critical clinical factors, to create a set of predictive combinations. These combinations were then input into three machine learning classifiers, algorithms that automatically organize or categorize data.
The clinical factors considered in the analysis included age, sex, IOP, corneal thickness, retinal nerve layer thickness, blood pressure, and body weight (BMI). All three algorithms demonstrated excellent performance, consistently and accurately predicting the onset and progression of glaucoma with an accuracy rate ranging from 91% to 99%.
The research team noted the potential benefits of predicting disease progression on an individual-patient basis. Such predictions would enable clinicians to offer customized management options to patients, including guidance on follow-up duration, initiation of IOP-lowering treatment, and IOP level targets.
However, the team acknowledged certain limitations in their study, such as the relatively limited dataset used for AI training and the inclusion of only those with normal IOP who had not received glaucoma treatment during the monitoring period.
