AAO journals show OCT AI prediction
- A March 2026 Ophthalmology Science paper reported that deep-learning models built from optical coherence tomography scans could estimate glaucoma visual field loss, comparing R2 U-Net, Dense U-Net, and UNet++ architectures. - The study used retinal nerve fiber layer thickness maps from OCT images and found all three U-Net variants could predict visual field outcomes, extending OCT analysis from structure toward function. - The work lands as glaucoma researchers push harder on forecasting progression from imaging and longitudinal data instead of relying only on repeated field tests. (aao.org)
Glaucoma slowly damages the optic nerve, and the vision it takes away cannot be restored. Doctors usually track it with two tools: optical coherence tomography scans that show eye structure, and visual field tests that measure what a patient can still see. (ophthalmologyglaucoma.org) (reviewofophthalmology.com) The problem is that visual field testing is noisy and patient-dependent. A patient has to respond spot by spot during the exam, and results can vary from fatigue, attention, or test familiarity. (ophthalmologyglaucoma.org) (ajo.com) That is why researchers keep trying to infer function from structure. If a scan can show where retinal nerve tissue is thinning, a model may be able to estimate where vision is most likely to be weak now or worsen later. (nature.com) (ajo.com) A new Ophthalmology Science study, published March 22, 2026, tested that idea with three U-Net family deep-learning models: R2 U-Net, Dense U-Net, and Nested U-Net, also called UNet++. The models used retinal nerve fiber layer thickness maps from optical coherence tomography images to predict visual field outcomes in glaucoma. (sciencedirect.com) (scilit.com) U-Net is a type of image model first built for medical segmentation, which means separating meaningful tissue patterns from background pixels. In glaucoma work, that architecture can be adapted so the scan is not just outlined, but translated into an estimate of functional loss. (sciencedirect.com) (ajo.com) The American Academy of Ophthalmology highlighted adjacent work on April 15, 2026, in a weekly journal roundup, though that item focused on a different artificial-intelligence model. In that study, a graph isomorphism network trained on 24-2 standard automated perimetry data from 676 patients and 1,009 eyes reached an age-adjusted area under the curve of 0.982 for detecting glaucomatous visual field defects. (aao.org) That April highlight was about reading visual field patterns directly, not predicting them from optical coherence tomography. But it points in the same direction: glaucoma algorithms are moving from simple classification toward more detailed mapping of where damage sits and how it may evolve. (aao.org) (ophthalmologyglaucoma.org) Other recent studies have pushed the same frontier with longitudinal data, which means repeated scans and tests over time instead of a single clinic visit. An American Journal of Ophthalmology report said baseline and serial structural measures could predict visual field progression with clinically relevant accuracy. (ajo.com 1) (ajo.com 2) Researchers are also trying multimodal systems that combine optical coherence tomography, visual fields, and clinical records. A medRxiv preprint from Mass Eye and Ear described a retrospective cohort of 10,864 patients for forecasting future glaucoma progression, but it has not yet been peer reviewed. (medrxiv.org) None of this means clinicians can replace standard visual field testing tomorrow. The published studies repeatedly note limits including single-center datasets, retrospective design, and the need for external validation before these models can guide routine care. (aao.org) (ajo.com) What the new papers show is narrower and more concrete: glaucoma imaging models are getting better at turning a structural scan into a functional forecast. In a disease defined by slow, irreversible loss, that is the question doctors have been trying to answer earlier. (sciencedirect.com) (ophthalmologyglaucoma.org)
