A machine learning model developed by researchers at the Wilmer Eye Institute promises to improve the detection of glaucoma progression through innovative applications of optical coherence tomography (OCT) data. Glaucoma, known to be one of the leading causes of irreversible blindness globally, has raised significant concerns as it is expected to affect over 110 million people by 2040. The early detection and monitoring of its progression are of utmost importance to manage the disease effectively and prevent substantial vision loss.
Monitoring glaucoma progression has typically relied on visual field (VF) testing coupled with OCT imaging. Each modality offers distinct advantages; OCT is particularly sensitive to detecting early structural changes, whereas VF testing is more informative during later stages when structural measurements may plateau. This research sought to explore the potential of using OCT data to estimate mean deviation (MD)—a key metric used alongside VF tests—in order to improve assessments of glaucoma progression.
The study involved the creation of a machine learning model trained on extensive datasets—70,575 paired OCT and VF studies. Researchers divided this data between two groups: one to develop the model and another consisting of 4,044 eyes with longitudinal data for validation. This model was constructed to predict how VF-MD varies over time, thereby aiding clinicians in monitoring patients with glaucoma more efficiently.
According to the findings, the ML model’s OCT-MD estimates achieved a mean absolute error (MAE) of 1.62 dB. Notably, this is lower than previous models; yet, the study concluded the performance was still inferior to direct VF measurements for detecting progression. The authors noted, “Our model’s OCT-MD estimates had an MAE of 1.62 dB, which is lower than those published previously.” Adjustments would be needed, as the authors suggested, “Overall, our data suggest future models converting OCT data to VF-MD must achieve prediction errors (MAE ≤ 1 dB) to be clinically valuable.”
The research pursued the idea of potentially replacing some VF data with OCT-MD estimates to lessen the testing burden on patients. By using OCT-derived estimates, it was hypothesized Health professionals would gain valuable insights faster, but findings revealed combining OCT-MD with VF-MD did not improve detection ability compared to utilizing VF-MD data alone. “Combining OCT-MD with VF-MD did not improve detection ability compared to using VF-MD alone,” the authors noted, marking the necessity for additional research and model refinement.
This comprehensive study outlines the challenges inherent to translating structural OCT data to functionally relevant VF data, indicating future directions for research will need to focus on increasing the accuracy of OCT-generated MD estimates. Strategies may involve incorporating additional structural features or different data inputs, such as macular data combined with optic nerve head scans, which may yield greater accuracy and clinical utility.
Understanding the limitations and potential applications for OCT-derived metrics is fundamental for improving glaucoma management, and this study serves as an important step toward enhancing the effectiveness of monitoring strategies utilized by healthcare professionals worldwide.