Moorfields Eye Hospital and UCL develop AI foundation model for ophthalmology

Moorfields Eye Hospital and the UCL Institute of Ophthalmology have developed an artificial intelligence foundation model for ophthalmology called RETFound, which is said to be capable of detecting markers of disease from retinal images and “can identify some of the most debilitating eye diseases across diverse populations”.

The model was trained with a curated dataset of 1.6 million images from Moorfields Eye Hospital through self-supervised learning, which researchers say “aims to alleviate data inefficiency by deriving supervisory signals directly from data, instead of resorting to expert knowledge by means of labels”. UCL adds that the need for expert human labels is a key challenge for developing AI models, as they are “often expensive and time-consuming to acquire”.

The study indicates that RETFound “is able to match the performance of other AI systems whilst using as little as ten percent of human labels in its dataset”, through the self-supervising approach which sees RETFound “mask part of an image, and then learn to predict the missing portions by itself”.

UCL states that RETFound “consistently outperforms existing state-of-the-art AI systems across a range of complex clinical tasks”. Additionally, by training the model on a large dataset representing London’s ethnical diversity, RETFound is said to form a “valuable base for researchers worldwide to build their systems in healthcare applications such as ocular disease diagnosis and systemic disease prediction”.

Professor Pearse Keane, senior author on the study, says that along with the use cases already identified for RETFound, the model “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.”

The study can be accessed in full here.

In other news from Moorfields and UCL, in May we shared their work using AI to help identify retinopathy of prematurity – the leading cause of childhood blindness.

Earlier in the year, Moorfield’s chief clinical information officer and director of digital medicine Peter Thomas joined us for a HTN Now virtual event to discuss building the foundations of scaled AI implementation and the applications of AI in eye care.

In AI diagnosis, we covered a study suggesting that “using AI to interpret images from a handheld ultrasound device is comparable at detecting how well the heart pumps as the gold-standard of diagnosis currently used in the NHS”.