How the application of machine learning and artificial intelligence (AI) can assist in the speed and quality of diagnosis
Dr Pearse A Keane
NIHR Clinician Scientist and Honorary Consultant Ophthalmologist NIHR Biomedical Research Centre
Moorfields Eye Hospital NHS Foundation Trust and
UCL Institute of Ophthalmology
A collaborative Research venture is now progressing between Moorfield Hospital’s Dr Pearse Keane and Google’s DeepMind team led by the co-founder Mustafa Suleyman, aimed at developing ways in which Artificial Intelligence (AI) can be applied to ophthalmology and in particular to the type of imaging of the eye called Optical Coherence Tomography (OCT – an established medical imaging technique that uses light to capture 3-dimensional images).
Undoubtedly, the scanning of patients’ eyes using OCT (three-dimensional scans of the retina which is much better at revealing eye disease than traditional retina photography) heralded one of the biggest developments in modern ophthalmology – but unfortunately, the increasing number of false positive referrals received by hospitals like moorfields will soon be exacerbated by the imminent roll out of OCT devices amongst opticians who may not have the sufficient training to interpret the scans. These scans are quick, easy and safe to acquire, but too many patients are being referred for the wrong reasons, leading to a clogging up of the clinics and the resulting inability of clinicians to treat genuine cases of sight loss (e.g diabetic retinopathy or wet amd) within an appropriate time scale. This swamping of services could not be happening at a worse time. Ophthalmology is already the second-busiest speciality in the NHS, with more than 9 million outpatient appointments per year.
Therefore, how can the application of machine learning and AI assist in improving the quality and speed of diagnosis thus allowing for earlier and more effective treatment whilst reducing patient numbers and prioritizing those that require urgent treatment first? The answer is that generally, AI is delivering huge improvements on e.g speech recognition, very good translation, very good image labelling and image recognition. There are much improved machine learning models, access to very large-scale computers and there is increasingly enough training data to help build effective models.
So with that developmental momentum, the millions of OCT scans held by Moorfields has presented DeepMind with the ideal dataset for DeepMind to apply its research. In a way, DeepMind is replaying all of the scans to the machine learning system in the same way that an expert consultant ophthalmologist might sit in front of their computer and watch scans and case studies over and over again – this is what the DeepMind calls “experience replay”. The system being applied to the Moorfields data is “imagining” an abstract form of the disease it looks for such as Diabetic Retinopathy (DR) , seeing it in its “mind’s eye”. This is similar to the technology used for example to look at photographs on Google Photos or Image Search or Facebook or to recognise faces in the photos
Despite The fact that the application of AI in spotting eye diseases is currently very much a research project, there is growing optimism that it will soon be able to “grade” eye scans more effectively and certainly much more quickly and more cheaply than a human. Mass adoption of OCT which is supported by AI within opticians may well be only 3 years away – people will be able to walk into a high-street optician, have an OCT scan and have it graded by an AI system.
However, although machine learning will become an invaluable tool in early diagnosis and the planning of treatment, the advent of AI in handling so many patient data sets will require greater ethical scrutiny and appropriate governance. It is this data that gives machine learning its formidable power, and the NHS is in a unique position to offer huge, well-labelled datasets.
Also, questions such as how patient information which is so key to the use of AI is shared, who gets to use it and who gets to profit from it are questions that could fail to be properly answered in the rush to implement this important new technology. These questions will need to be answered as AI will soon be impacting on healthcare generally – for example, the hospital environment is such an expensive and complex system and clearly the humans working in it are simply overwhelmed by the scale and complexity of managing so many patients who are on so many different pathways and who need so many different tests and interventions. It becomes a massive co-ordination exercise and therefore AI can be applied so tasks in various areas of the hospital could be more efficiently and speedily prioritized to improve patient outcomes and care.