Continuous assessment of age-related macular degeneration (AMD) severity could improve disease monitoring and clinical decision-making beyond traditional discrete staging systems. During Bascom Palmer Eye Institute’s 23rd annual Angiogenesis, Exudation, and Degeneration meeting, held virtually on February 7, 2026, Aaron Y. Lee, MD, MSCI, discussed a novel approach to predicting continuous AMD severity from retinal optical coherence tomography (OCT) images using temporal pairwise supervision.

In this Q&A conversation with the Eye Care Network, Lee explains the methodology, clinical implications, and challenges of integrating such models into routine practice. Lee is the Arthur W. Stickle Distinguished Professor and Chairman with the John F. Hardesty, MD, Department of Ophthalmology and Visual Sciences at Washington University (WashU) in St Louis, Missouri.

Note: Transcript edited for clarity and length.

Can you explain how temporal pairwise supervision improves AMD severity prediction compared with single-timepoint models, and what clinical advantages this offers for longitudinal patient monitoring?

Aaron Y. Lee, MD, MSCI: We are trying to tie in the temporal dynamics of AMD as a disease rather than cross-sectional, single timepoint models. Although models trained by single timepoints can perform very well for diagnostics, they can lack the ability to model disease progression. We are hoping that we can build a severity map of the macula with relation to AMD disease progression and take advantage of the fact that AMD is mainly a monotonically progressive disease.

How do you envision this continuous severity prediction model being integrated into real-world clinical workflows, and what barriers exist for deployment in routine retinal practice?

Lee: We hope that this modeling will be able to provide a continuous severity score that can be integrated into existing electronic medical record systems for longitudinal patient tracking and be able to give clinicians the ability to understand the prognosis of early and intermediate AMD. Currently we use a single discrete stage to describe the severity of AMD at a given clinical encounter, but we hope that this approach will unlock the ability to understand how affected the perifoveal or subfoveal area is to AMD.

There are definitely barriers to deployment of such a model but at WashU we have built an informatics pipeline to allow every image to be read in real-time through an artificial intelligence model without the need for a clinician to click. The results would end up automatically back in the picture archiving and communication system or in the electronic health record.

What data sets and patient populations were used to validate the model, and how well does it generalize across different OCT devices, institutions, and demographic groups?

Lee: We used data from these countries: Australia, Germany, [and] US to train the model and data from France to validate the model. This was only done using the Heidelberg Spectralis device and it is unlikely to generalize well to other devices. The demographics and institutions were fairly broad. The generalization gap between the training and the data from France, which was our held-out test set, was small suggesting that the model was not overfitting.

Do you foresee this continuous severity scoring influencing treatment timing or frequency for AMD therapies, and how might it change how clinicians stratify patient risk?

Lee: Continuous severity scores are more likely closer to reflecting true disease progression than a discrete stages. The AMD severity of the macula does not suddenly go from one stage to another in a day. Instead, it is more likely that the disease is evolving slowly over time, and a continuous severity score would be more reflective of that biological process. More clinical validation will be necessary before what we have developed is ready.

Aaron Y. Lee, MD, MSCI
E: [email protected]
Lee is the Arthur W. Stickle Distinguished Professor and Chairman with the John F. Hardesty, MD, Department of Ophthalmology and Visual Sciences at Washington University in St Louis, Missouri.

Reference

1. Lee AY. AMD continuous severity prediction from retinal OCT using temporal pairwise supervision. Presented at: Angiogenesis, Exudation, and Degeneration 2026; February 7, 2026. https://umiamihealth.org/bascom-palmer-eye-institute/healthcare-professionals/continuing-medical-education/angiogenesis/program