“We are drowning in people we need to see in hospital eye services, and some people are going blind as a result. We are looking at nearly 10m hospital appointments for ophthalmology a year, with an approximately 33% increase over the past five years. Since 2017 the busiest of specialties in the NHS has been for eyes. AI may help to allay these challenges,” said Professor Pearse Keane, speaking on the Optical Suppliers Association stand at 100% Optical in London this week.
“Our new Moorfields Eye Hospital – due to be complete by 2027/8 at Kings Cross – is designed to take advantage of new care pathways, to help work with community optometry and to upskill people, and to give patients the ability to play a greater role in their own care,” he said.
As Professor of Artificial Medical intelligence at UCL, and Medical Retinal Consultant at Moorfields, he is well placed to highlight the UK’s role as a world leader in this emerging field of healthcare. “I am proud that ophthalmology is at the forefront of AI in medicine, as an exemplar for other healthcare sectors, but we are still at the super early stages.
“We have the largest ophthalmic imaging resource in the world at Moorfields – larger than the combination of the top five US providers combined – plus we have gold-standard governance for engagement and privacy.”
The unrivalled foundation of data provided by the NHS for machine learning to aid AI is very evident. The launch of RETFound – a collaboration between UCL and Moorfields – utilises 1.6m retinal images to assist the early research phase of progressing to a functioning algorithm. As an opensource tool this can be used as a building block for others to develop, he explained.
“As a world leader in medical AI the UK has the advantage of NHS data sets and some of the best universities in the world. This has implications, particularly, for rarer or less common diseases as we could find a pathway.”
Pearse said that since the 2016 press reports of Rapid Access Macular Clinics, there had been much hype but that bringing this care into the community and general ophthalmologists is not yet real.
“These AI systems are good at giving better measurements for intra or sub-retinal fluids to help guide treatments. But how do we integrate this into pathways and what is the business model?
“The most important application in the short term is, perhaps not in direct patient care, but in clinical trial planning. We can help to overcome the challenges of recruitment by sending an algorithm to clinics to identify segment pathology to find patients suitable to be approached to participate. This helps us in writing the protocol for trials, such as the size of drusen or atrophy. This does not require regulatory approval but could still bring immense benefit.”
With more than 500 medical AI systems in the FDA approval system, the appetite for progress in this field is evident, but insurance and payment systems will be a major aspect to compute – “No one has yet figured out the right business model for the use of medical AI: the infrastructure for data aggregation, and expertise in formatting governance are still in the melting pot.”
Professor Keane spoke of the wider implications, particularly the Alzheimer Eye Study, linked to the NHS database and impatient admissions. Using retinal scan data as a window to track patients over 40 for early signs of systemic disease with subtle changes is progressing. Changes in the retina of patients with schizophrenia and looking at neuropathy and changes in the ganglion cell and inner nuclear layer of patients who may go on to develop Parkinsons, perhaps seven years before other symptoms appear, were also highlighted.
“With Parkinsons we have shown on average that there are subtle changes, but this is not a predictive test. Also, for cardiovascular disease – this could have big effects for the eyecare community and could have profound public health benefits, including conditions such as diabetes and high blood pressure.”
Prof Keane spoke about emerging abilities starting in healthcare, with medical questions answered by the potential of Chat GPT and Foundation Models which are accelerating AI development – “A large model, trained on a large amount of data and fine-tuned for downstream tasks: This can be for any type of data as demonstrated by the 2023 Canadian study, led by Fares Antaki, which provided accurate clinical performance of 75% vs human performance of 72%.
“We are already seeing AI surpassing human performance, with so much more to come, and the potential of SORA from Open-AI is bringing text-to-video which has enormous potential. Many people around the world are looking to see how this can become a reality in healthcare.”
