blog

DeepMind is a British company that was formed in 2010 with the aim of creating artificial general intelligence and using it to tackle some of society’s toughest challenges.

DeepMind Health was launched in February 2016 at the Royal Society of Medicine. Their work is currently based around two strands: working with doctors, nurses and patients to develop clinical apps for direct patient care, and collaborating with clinical researchers to explore the ways artificial intelligence could improve patient care. They recently published two Study Protocols on applying machine learning to health conditions on F1000Research. In this blog post, they summarise both projects and talk about the collaboration with hospitals as well as the future application of those machine learning techniques.

 

Your Study Protocol on automated analysis of retinal imaging using machine learning techniques has recently passed peer review. Can you briefly summarise it?

Conditions such as diabetic eye disease and age-related macular degeneration (AMD), which both cause sight loss, affect more than 625,000 people in the UK. Yet it’s estimated 98% of severe sight loss caused by diabetes could be prevented with early detection.

At the moment, eye care professionals use digital scans of the fundus (the back of the eye) and scans called optical coherence tomography (OCT) to diagnose and determine the correct treatment for these serious eye conditions. But these scans are highly complex and take a long time for eye health professionals to analyse, which can have an impact on how quickly they can discuss diagnosis and treatment with a patient.

To date, traditional software analysis tools have been unable to explore scans fully. Our research project aims to investigate how machine learning could help analyse these scans much more quickly, leading to earlier detection and intervention and ultimately supporting ophthalmologists in providing their patients with the best care possible.

In the long term, diagnosis for eye diseases powered by machine learning could allow ophthalmologists to prioritise the most severe cases and provide immediate treatment to those patients who need it. Diabetes UK estimates that 4 million people in the UK have diabetes, and people with the condition are 25 times more likely to suffer from sight loss. Machine learning could therefore have a significant impact on the long-term prevention of debilitating diabetic eye disease.

 

You’ve recently submitted another Study Protocol on applying machine learning to automated segmentation of head and neck tumour volumes and organs at risk on radiotherapy planning CT and MRI scans. Can you explain the application for that?

Radiotherapy for treating head and neck cancers must be painstakingly planned because of the many delicate structures and vital organs that are concentrated in this area of the body. Before radiotherapy can be administered, clinicians have to produce a detailed map of the areas of the body to be treated, and the areas to avoid.

The process is known as segmentation, and is performed to reduce the risk of radiotherapy treatment for a cancer at the back of the mouth or in the sinuses from damaging any healthy tissue. In the head and neck, the concentration of these delicate structures means the process can take up to four hours.

We think machine learning methods could make radiotherapy planning more efficient by reducing the time radiotherapy planning takes. This has the potential to free up the clinicians’ time and significantly speed up the rate at which radiotherapy treatments are planned for the over 40% of cancer patients who receive radiotherapy treatment.

 

How did the collaboration with the Moorfields Eye Hospital and University College London Hospital respectively come about?

We’re particularly excited that we were approached by clinicians to lead these projects.

Pearse Keane, a consultant ophthalmologist at Moorfields, contacted DeepMind through our website with a direct clinical need to be able to better analyse the eye scans that are used to diagnose and determine the correct treatment for serious eye conditions. He wanted to explore whether machine learning could be applied to streamline the diagnosis and reduce the time between scan and treatment.

Following this initial contact, the DeepMind team met with Dr Keane and Professor Sir Peng Tee Khaw, Director of the National Institute of Health Biomedical Research Centre at Moorfields Eye Hospital NHS Foundation Trust and UCL Institute of Ophthalmology, to discuss how the project would work.

Our collaboration with UCLH began when a member of UCLH’s radiotherapy team and a clinical advisor to DeepMind met at a healthcare event. They agreed there was scope to investigate if radiotherapy planning could be improved with machine learning methods. The idea was explored and discussed in detail by both UCLH and DeepMind teams.

 

Why is this research important?

As incidence rates of diabetes-related blindness and head and neck cancers rise, the need for innovative technologies and methods to prevent, diagnose and treat these devastating diseases becomes increasingly acute.

Both of these exciting projects aim to show the role machine learning technology could play to help clinicians diagnose and treat diseases that affect the lives of millions of people across the world, and work towards solving the hard research problems that need to be overcome to make this possible.

We’re hopeful that this research could show a way to provide clinicians with highly innovative medical tools while also freeing up their time for direct patient care.

 

What are you hoping to apply the results to?

Machine learning could have a transformative impact on clinical care. Both of these initial research projects are proof of concept studies, but if they are successful, they would show the great promise that these techniques could have in clinical application in these areas, speeding up radiotherapy and eye disease treatment times.

Indeed, some clinicians are already excited by the potentially significant clinical impact of this. A breakthrough in the radiotherapy segmentation process, for example, has the potential to be applied to other cancers, as Professor Kathy Pritchard-Jones, the chief medical officer of London Cancer, has said. That could benefit healthcare professionals, and ultimately patients, across the country.