DeepMind and Moorfields Eye Hospital NHS Foundation Trust
Our work with Moorfields Eye Hospital NHS Foundation Trust (Moorfields) began in mid-2016 after Dr Pearse Keane, a consultant ophthalmologist at the hospital, asked us to collaborate on a project to explore whether artificial intelligence (AI) technology could help their clinicians improve the way sight threatening eye conditions are diagnosed and treated, in order to improve patient care.
Working in collaboration with clinicians at Moorfields Hospital, we have been able to develop AI technology which can automatically detect eye conditions in seconds and prioritise those patients in urgent need of care, matching the accuracy of expert doctors with over 20 years experience.
We believe this triaging process could drastically cut down the time taken between detection and treatment, making it much less likely that these conditions will lead to sight loss.
Why are we looking at eye disease research?
Eye disease is a significant global health problem, and one of the major causes of sight loss, which affects over 285 million people worldwide – a figure that is expected to triple by 2050.
Many of these eye diseases can be prevented with early detection and treatment. However, accurately detecting and diagnosing eye disease remains challenging.
Currently, clinicians, ophthalmologists and other eyecare professionals use optical coherence tomography (OCT) scans to help diagnose eye conditions. These 3D images provide a detailed map of the back of the eye, but they require highly trained, expert analysis to interpret the results. This can be a time-consuming process.
The time taken to analyse these scans, combined with the sheer number of scans that healthcare professionals have to go through (over 1,000 a day at Moorfields Eye Hospital alone), means that there can be lengthy delays in detection and treatment of eye conditions. These delays can cause sight loss.
We are working towards AI technology that can make the screening process much more efficient. This would allow clinicians to prioritise patients with the most serious conditions and treat them before irreversible damage sets in.
How does this AI technology work?
The AI technology we have developed is able to analyse OCT eye scans and correctly recommend how patients with a wide range of eye diseases, such as age-related macular degeneration, diabetic eye disease and severe myopia, should be referred for treatment. At 94% accuracy, this matches the accuracy of expert clinicians at Moorfields Eye Hospital with over 20 years’ experience in the field.
This alone is hugely promising, but we have also designed our AI technology so that it’s able to explain to clinicians how it arrives at its recommendations, rather than simply presenting them in isolation like a “black box”. This includes information about the features of eye disease the AI technology has identified on the OCT scan, so doctors and eyecare professionals can gain insight into its “thinking”, as well as the level of confidence the system has in its recommendations, in the form of a percentage.
We believe this functionality is likely to be critically important, since doctors, opthalmologists and other eyecare professionals are always going to play a key role in deciding the type of care and treatment a patient receives. Allowing them to scrutinise the technology’s recommendations and check they are accurate will therefore be key
Finally, we have developed the technology so it can be easily applied to different types of eye scanning devices, not just the specific type of device it was trained on at Moorfields. This could massively increase the potential number of people across the world that it could benefit from any resulting technology. This functionality also future-proofs the technology against new scanning devices that could be developed in the future.
Our technology is able to do all of these things using neural networks – complex mathematical systems for identifying patterns in images or data. Thousands of OCT scans from Moorfields Eye Hospital were fed into two types of neural networks, to teach these networks to perform two specific tasks.
The first “segmentation network” was taught to analyse the OCT scans and identify different types of eye tissue in the scan, which it would then use to automatically create a digital map of the eye and the features of disease it sees, such as haemorrhages, lesions, irregular fluid or other symptoms of eye disease. This map allows eye care professionals to gain insight into the system’s “thinking.”
The second “classification network” was taught to analyse this “map” to provide one of four referral suggestions currently used at Moorfields Eye Hospital.
By separating these two processes, the technology can be applied to different types of eye scanners easily. That is because only the first neural network needs to learn how the different types of tissue appear in images produced by a new scanner – the second network can simply be reused.
This separation also ensures the technology is able to provide information that helps explain to clinicians how it comes to its recommendations, since clinicians are able to review the relevant parts of the “map” on which the technology has based its decision, enabling them to understand the process and verify its accuracy.
What's the next step?
Our partners at Moorfields want our research to help them improve care, reduce some of the strain on clinicians, and lower costs - all at the same time. So we’ve also worked hard on what comes next.
If this technology is validated for general use by clinical trials, Moorfields’ clinicians will be able to use it for free across all 30 of their UK hospitals and community clinics, for an initial period of five years. These clinics serve 300,000 patients a year and receive over 1,000 OCT scan referrals every day – each of which could benefit from improved accuracy and speed of diagnosis.
We’re also proud that the work we’ve put into this project will help accelerate many other NHS research efforts. The original dataset held by Moorfields was suitable for clinical use, but not for machine learning research. So we’ve invested significantly in cleaning up, curating and labelling the dataset to create one of the best AI-ready databases for eye research in the world.
This improved database is owned by Moorfields as a non-commercial public asset, and it’s already been used by hospital researchers for nine separate studies into a wide range of conditions - with many more to come. Moorfields can also use DeepMind’s trained AI model for their future non-commercial research efforts.
Data & Security with Moorfields
Data security and integrity is hugely important to all of our partnerships. That is why every step is taken by our teams to protect data shared with us, audit it, and destroy it when it is no longer being used.
Moorfields have shared one million de-personalised digital eye scans, used by eye health professionals to detect and diagnose eye conditions. These scans are de-personalised, meaning that any information that could be used to identify individuals has been removed before DeepMind receives it.
All research projects go through rigorous regulatory and Trust approvals and are conducted only on de-personalised patient data. There is a lot more information on our agreements with Moorfields in our FAQs, and more on the way we handle data below.