Sight loss will be preventable with the help of artificial intelligence, after the development of an AI-based model in the field of ophthalmology, RETFound, in a medical first.
It can help improve the diagnosis of some of the most debilitating eye diseases, including diabetic retinopathy and glaucoma, but also predict systemic diseases such as Parkinson’s, stroke and heart failure. Identifying general health problems through the eyes is an emerging science called “oculomics”.
How sight loss can be prevented with the help of AI
The RETFound AI system can both detect sight-threatening eye diseases and predict general health, including myocardial infarction, stroke and Parkinson’s disease.
The RETFound AI core model was developed by researchers at Moorfields Eye Hospital and University College London (UCL) Institute of Ophthalmology, using millions of eye scans used in the Public Health System (NHS) in the United Kingdom of Great Britain and of Northern Ireland.
RETFound is freely available for use by any institution worldwide as part of global efforts to detect and treat blindness with the help of artificial intelligence. It works well in various populations and in patients with rare diseases.
Innovative approach
One of the biggest challenges in developing artificial intelligence models is the need to label images with the help of experts, a process that is often expensive and time-consuming.
RETFound can match the performance of other AI systems while using only 10% of the human tags in its dataset. This efficiency improvement was achieved by using an innovative self-supervision approach, where RETFound masks parts of an image and then learns to estimate the missing parts by itself.
“By training RETFound on datasets that represent the ethnic diversity of London’s population, we have developed a valuable foundation for researchers around the world to build their systems into healthcare applications such as eye disease diagnosis and systemic disease prediction” , said the study’s lead author, PhD student Yukun Zhou, from UCL’s Center for Computational Medical Imaging, Medical Engineering and Human Medicine, and Moorfields Eye Hospital.
The RETFound base model was trained on a set of 1.6 million images from Moorfields Eye Hospital.
