The retina – a light-sensitive layer of tissue at the back of the eye – provides a non-invasive ‘window’ to the nervous system and the capillary network, made of tiny blood vessels and vital to the function of organs such as the brain, heart and kidneys.
The model, called RETFound, is one of the first AI foundation models in healthcare and the first in ophthalmology and has been developed to improve the diagnosis of some of the most debilitating eye diseases and predict other conditions beyond the eye. It has been enabled by INSIGHT, the Health Data Research Hub for Eye Health.
A foundation model is a very large, complex AI system, trained on huge amounts of unlabelled data, which can be fine-tuned for a diverse range of tasks. Published in Nature, researchers trained RETFound using a dataset of 1.6 million retinal scan images from Moorfields Eye Hospital.
RETFound was shown to consistently outperform existing state-of-the-art AI systems across a range of complex clinical tasks, although the performance was limited. More importantly, it addresses a significant shortcoming of many current AI systems by working well in diverse populations, and in patients with rare diseases.
Senior author, Professor Pearse Keane, Director of INSIGHT, Professor of Artificial Medical Intelligence at University College London and Consultant Ophthalmologist at Moorfields Eye Hospital, said: “This is another big step towards using AI to reinvent the eye examination for the 21st century, both in the UK and globally. We show several exemplar conditions where RETFound can be used, but it has the potential to be developed further for hundreds of other sight-threatening eye diseases that we haven’t yet explored.
“If the UK can combine high quality clinical data from the NHS, with top computer science expertise from its universities, it has the true potential to be a world leader in AI-enabled healthcare. We believe that our work provides a template for how this can be done.”
One of the key challenges when developing AI models is the need for experts to add labels to large amounts of data, an important part of training as it shows the AI what it needs to look for in the images. However, doing this is often expensive and very time-consuming. RETFound was able to match the performance of other AI systems whilst using as little as 10% of the labels in its dataset. This improvement in label efficiency is achieved by using an innovative self-supervising approach in which RETFound masks parts of an image, and then learns to predict the missing portions by itself.
First author of the study, PhD student Yukun Zhou at UCL Centre for Medical Image Computing, UCL Medical Physics & Biomedical Engineering, and Moorfields Eye Hospital NHS Foundation Trust, said: “By training RETFound with datasets representing the ethnical diversity of London, we have developed a valuable base for researchers worldwide to build their systems in healthcare applications such as ocular disease diagnosis and systemic disease prediction.”
The research team are making the system open-source, freely available for use on GitHub by any researchers worldwide, to speed up global efforts to detect and treat blindness using AI.
Read the full paper
