Related Experiment Video
Updated: Aug 1, 2025
06:16
Author Spotlight: Unraveling the Pathogenesis of Age-Related Macular Degeneration and Discovering Potential Therapies
Published on: July 28, 2023
2.6K
Exploring Healthy Retinal Aging with Deep Learning
Martin J Menten1,2, Robbie Holland1, Oliver Leingang3
1BioMedIA, Imperial College London, London, United Kingdom.
Ophthalmology Science
|April 28, 2023
Summary
Deep learning models visualize individual retinal aging. Counterfactual generative adversarial networks (GANs) reveal subject-specific changes in retinal layers, aiding biomarker discovery for healthy and pathologic aging.
Area of Science:
- Ophthalmology
- Artificial Intelligence
- Medical Imaging
Background:
- Healthy aging causes changes in retinal layer structure.
- Understanding individual aging trajectories is crucial for identifying potential biomarkers.
- Previous studies analyzed population-wide changes in retinal layers.
Purpose of the Study:
- To investigate the individual course of retinal changes during healthy aging using deep learning.
- To develop a method for visualizing hypothetical scenarios of retinal aging.
- To analyze subject-specific alterations in retinal layers with age and sex.
Main Methods:
- Retrospective analysis of 85,709 retinal OCT images from the UK Biobank.
- Development of a counterfactual generative adversarial network (GAN) to synthesize high-resolution OCT images and longitudinal time series.
- Generation of counterfactual images by altering age/sex while keeping subject identity fixed.
Main Results:
- The counterfactual GAN successfully visualized individual retinal aging trajectories.
- Quantified average age-related changes per decade: RNFL -0.1 μm, GCIPL -0.5 μm, INL-RPE -0.2 μm, RPE +0.1 μm.
- Demonstrated the ability to explore individual variations in retinal layer thickness changes (increase, decrease, or stagnation).
Conclusions:
- Counterfactual GANs are effective tools for researching retinal aging.
- This approach generates high-fidelity OCT images and longitudinal data for hypothesis generation.
- Enables exploration of potential imaging biomarkers for healthy and pathological aging, guiding future clinical trials.
