DeepMind, the British artificial intelligence group, has built a working prototype of a device that can diagnose complex eye diseases in real time, in a major step towards the Alphabet-owned company’s first medical device.
In a live demonstration this month of its AI system, where a patient agreed to be examined publicly, DeepMind performed a retinal scan and real-time diagnosis of her eye. The scan was analysed by a set of algorithms in Google Cloud, which provided an urgency score and a detailed diagnosis, all in roughly 30 seconds.
The system is capable of detecting a range of eye diseases — including glaucoma, diabetic retinopathy and age-related macular degeneration — with the same level of accuracy as the world’s leading specialists. It was developed in conjunction with London’s Moorfields Eye Hospital over the past three years. Details about the research were published in Nature Medicine, the scientific journal, last August.
In recent months, the company has collaborated with the team at Moorfields to build a working product, although it has not yet received any regulatory approval.
A DeepMind spokesperson said that if the research results in a product that passes clinical trials and regulatory approvals, doctors at Moorfields will be able to use the product for free for an initial period of five years.
Alan Karthikesalingam, project lead and senior clinician scientist at DeepMind, said it was a “major milestone” towards a bedside tool that could be used by GPs.
“What we’ve been working on really hard is how to take this type of early-stage research system and start to move it into a cloud technology, building a prototype of a system really used in practice,” Mr Karthikesalingam said at the Wired Health event in London where the demo took place.
He said the goal of the AI tool is to figure out “should someone call a specialist and if so, how urgently and why?”
DeepMind founded its health division, as part of its applied team led by co-founder Mustafa Suleyman, in 2016. The goal of this team has been to develop and deploy AI technologies in real-world settings, including GP surgeries and hospitals, but so far it has no commercial products and the division has yet to generate any revenues.
Other health projects include a medical diagnosis app, Streams and programs using AI to analyse medical scans in the UK and the US.
From the archive
In November, DeepMind announced that it would transfer control of its health unit to a new Google Health division in California, an indication of its plans to expand and commercialise its efforts.
The market for AI healthcare services is rapidly expanding, with analysts expecting it to reach $6.6bn by 2021.
“This is likely part of a larger strategy to make Google a reliable outsourced R&D partner. Demonstrating that its applications of AI can have a significant impact on outcomes builds necessary credibility as they move deeper into healthcare, which as a whole could be a significant business for them,” said Nikhil Krishnan, former healthcare analyst at CB Insights.
“However as a business, eye screening alone probably wouldn’t move the needle relative to Google’s core search/ad business, especially considering the risk they’d be taking on it,” he added.
In the US, the Food and Drug Administration’s approvals of medical AI algorithms has expanded dramatically, from two in 2017 to one or two a month in 2018. One is an algorithm from Iowa-based start-up IDx that, like DeepMind, can scan and diagnose diabetic retinopathy in real time and is already used in clinical care across Europe. The FDA said it was the first AI system to provide a medical decision “without the need for a clinician to also interpret the image or results, which makes it usable by healthcare providers who may not normally be involved in eyecare”.
DeepMind claims its product will not just offer diagnoses, but also be able to explain exactly how it arrived at its conclusion and how certain it is of the result, which is crucial for healthcare professionals.
“For an ophthalmologist, this is jaw-dropping. What you can see is [the AI] has segmented every single point, about 65m data points in this scan [creating] super high-resolution images,” said Pearse Keane, a consultant ophthalmologist at Moorfields. “We have to bring the same levels of rigour [to] how we validate the algorithms that we would with any medical device, but my personal prejudice is that ophthalmology will be the first speciality of medicine that is fundamentally transformed by AI.”