Google's DeepMind is embarking on a new research project to help doctors spot the early signs of sight-threatening eye diseases. The company's British-based artificial intelligence division will use machine learning to analyze more than one million anonymous eye scans, creating algorithms that can detect early warning signs that humans might miss. The project is DeepMind's second collaboration with the UK's National Health Service (NHS), but the first to use artificial intelligence.
"There’s so much at stake."
DeepMind is hoping to spot two eye conditions in particular: wet age-related macular degeneration and diabetic retinopathy, the latter being the fastest growing cause of blindness around the wold. "There’s so much at stake, particularly with diabetic retinopathy," DeepMind co-founder Mustafa Suleyman told The Guardian. "If you have diabetes you’re 25 times more likely to go blind. If we can detect this, and get in there as early as possible, then 98% of the most severe visual loss might be prevented."
The collaboration between DeepMind and the NHS specialist Moorfields Eye Hospital was reportedly sparked by an unsolicited request from a doctor at Moorfields. Pearse Keane, consultant ophthalmologist, told The Guardian he'd seen DeepMind's work training an AI agent to teach itself to play Atari games, and thought the company's machine learning skills could be used to analyze eye scans, known as Optical Coherence Tomography (OCT) images. "I had the brainwave that deep learning could be really good at looking at the images of the eye," said Keane. "Within a couple of days I got in touch with Mustafa, and he replied.”
The OCT data provided by Moorfields will be anonymized, meaning that the collaboration will face less scrutiny than DeepMind's previous collaboration with the NHS on its Streams app. This project was controversial as it gave DeepMind access to up to the full care history of some 1.6 million patients. Google was accused of not getting the proper authorization to handle this data, but both DeepMind and the NHS trust it worked with insisted that all the agreements were in order. The Streams app is still being tested, and is intended to alert doctors to when patients are at risk of kidney failure.
On the subject of the Moorfields' eye scans, Google says it's not possible to "identify any individual patients" from the data. "The data used in this research is not personally identifiable," said the company. "They’re also historic scans, meaning that while the results of our research may be used to improve future care, they won’t affect the care any patient receives today."