UK (Commonwealth Union) – Researchers from the University College London (UCL) and Moorfields Eye Hospital have successfully crafted an artificial intelligence (AI) system that possesses remarkable capabilities. This AI system not only demonstrates proficiency in identifying eye diseases that can lead to vision loss but also exhibits the potential to predict broader health conditions such as heart attacks, strokes, and Parkinson’s disease.
Known as RETFound, this marks one of the pioneering AI foundation models in the realm of healthcare, specifically within ophthalmology. It was meticulously developed using an extensive dataset of millions of eye scans sourced from the NHS. The research team is now making this cutting-edge system openly available, turning it into an invaluable resource for institutions worldwide. By doing so, they aim to establish it as a fundamental tool in the global pursuit of using AI to detect and manage conditions that cause blindness. This groundbreaking achievement was recently published in the journal Nature.
The realm of AI is witnessing a rapid evolution, exemplified by the emergence of foundational models like ChatGPT. These foundation models represent exceedingly complex AI systems, meticulously trained on vast amounts of unlabeled data. They can subsequently be fine-tuned for a wide spectrum of diverse tasks. RETFound distinguishes itself by consistently surpassing the performance of existing state-of-the-art AI systems across various intricate clinical tasks. More importantly, it addresses a critical limitation often found in current AI systems by exhibiting efficacy across diverse populations and even among patients with rare diseases.
Senior author Professor Pearse Keane, affiliated with the UCL Institute of Ophthalmology and Moorfields Eye Hospital, expressed the significance of this milestone. He indicated that it is another significant stride in harnessing AI to revolutionize the 21st-century eye examination, both in the UK and on a global scale. While they showcase several exemplary conditions where RETFound can be employed, its potential extends to hundreds of other sight-threatening eye diseases that have yet to be 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.”
AI foundation models have garnered significant attention, being heralded as “a game-changing technology” in a report released by the UK government earlier this year. This spotlight on foundational AI models intensified with the introduction of ChatGPT in November 2022, a model grounded in extensive text data and designed as a versatile language tool. In a pioneering venture, RETFound has embarked on a parallel path, utilizing millions of retinal scans to create a model capable of adaptable applications, marking a world-first in the realm of AI and ophthalmology.
One of the primary obstacles in the development of AI models revolves around the necessity for expert human annotations, a process often characterized by high costs and time-consuming efforts. As detailed in the research paper, RETFound manages to achieve performance levels comparable to other AI systems while relying on as little as 10% of human annotations within its dataset. This enhancement in annotation efficiency is realized through an innovative self-supervised technique, where RETFound strategically conceals segments of an image and autonomously learns to predict the missing components.
Researchers further pointed out the potential applications of RETFound extend far beyond the realm of eye scans. This AI model has the capability to significantly enhance the diagnosis of debilitating eye conditions, such as diabetic retinopathy and glaucoma. Moreover, it possesses the ability to forecast systemic ailments like Parkinson’s disease, stroke, and heart failure. The nascent field of ‘Oculomics,’ a term coined in 2020 by one of the paper’s co-authors, Professor Alastair Denniston, explores the intriguing concept of identifying general health issues through the eyes.
