WEHI creates the most detailed retinal maps ever produced

11 Feb 2025

AI analysis of OCT data in UK Biobank library reveals links between retinal thickness and disease.

A project described as one of the largest eye studies in the world could pave the way for routine eyecare imaging to assume new importance in disease screening.

Researchers at Australia's WEHI (Walter and Eliza Hall Institute) medical research center applied AI analysis to over 50,000 OCT images contained in UK Biobank, a biomedical database and research resource first established in the United Kingdom in 2002.

The goal was to use these maps as a means to identify new genetic factors that influence retinal thickness (RT), and see if retinal thinning was linked with a range of diseases. The results were published in Nature Communications.

Since the retina is part of the central nervous system along with the brain and spinal cord, diseases linked to degeneration or disruption of this critical system can manifest themselves in the retina, potentially an easily accessible location to look for them.

Previous studies in this area have included the use of fluorescence imaging as a means to assess build-up of amyloid plaques in the retina, and monitor the progress of treatments for dementia.

The new WEHI study focused on RT, already used as a parameter of interest when monitoring the behavior of immune cells as they become inflamed in response to disease.

Routine OCT examinations in ophthalmology clinics measure this thickness as a matter of course, with a total of some 85,000 data sets being available in UK Biobank.

Although some previous studies of the RT dataset had sought genome-wide associations with diabetic retinopathy, WEHI applied a deep learning-based image segmentation method to produce a high-resolution RT dataset sourced from individuals without explicit retina illness, and investigated relationships between RT and a number of medical and physiological factors.

OCT shows progression of multiple sclerosis

The project applied a deep convolutional neural network to generate RT data over 128 by 256 pixel grids, each pixel capturing an area of retina roughly 47 by 12 microns. This allowed a total retina area of 6000 by 6000 microns to be assessed.

A number of correlations between RT and other patient data were then made, including to age, sex, and retinal development data, with the enhanced resolution available via the new technique said to reveal novel RT associations not previously studied.

"Through the application of AI, we generated the highest-resolution spatial dataset of RT ever produced," commented the team in its paper.

"Our analyses reveal RT to be intricately related to a plethora of factors spanning the genome, metabolites, blood traits, and diseases, with the parafoveal area most enriched for associations. We found reduced RT, or retinal thinning, to be associated with poorer health and increased burden of disease."

In some of the most important findings, a reduction of RT was shown to be highly associated with multiple sclerosis, noted the project.

This result provides strong, independent confirmation of multiple reports of the utility of OCT as the source of biomarkers for MS and MS progression, said WEHI. It indicates that future studies utilizing RT as a biomarker should focus on the nasal perifoveal region of the macula, as this region contains the greatest signal.

WEHI has made its results publicly accessible through Retinomics, an interactive web portal where location plots show which health factors have now been linked with specific areas of the macula.

"Technologies like AI fuel discovery, and when fused with brilliant minds, there is an extraordinary ability to transform big population data into far-reaching insights," said WEHI's Melanie Bahlo. "There has never been a time in history where this powerful combination has come together to advance human health."