Google’s DeepMind Prepares Clinical Device For Eye-Scan Diagnoses


London-based AI unit demonstrates real-time diagnosis of eye health using cloud-based prototype, as it plans clinical trials

Google’s London-based DeepMind artificial intelligence unit has created a working prototype of what would be its first commercial medical device, the result of the unit’s three-year collaboration with Moorfields Eye Hospital.

In a live demonstration of the system at an event last week, DeepMind performed a retinal scan and real-time diagnosis on a patient who had agreed to be examined publicly.

The scan was analysed by DeepMind’s algorithms in Google’s cloud, which provided an urgency score and a detailed analysis in about 30 seconds, the Financial Times reported.

The prototype – which Moorfields ophthalmologist Pearse Keane, speaking at the Wired Health event in London last week, compared to a “concept car” – is able to diagnose diseases including glaucoma, diabetic retinopathy and age-related macular degeneration with an accuracy comparable that of experienced specialists.

DeepMind co-founder and chief executive Demis Hassabis. Image credit: DeepMind


The research has already resulted in a paper published the journal Nature Medicine last August, with the prototype more recently developed in collaboration with Moorfields.

DeepMind said the device, which must still undergo clinical trials and obtain regulatory approvals, would then be available for use by doctors at Moorfields for free for an initial period of five years.

The unit also makes the Streams medical diagnosis app and tools used to analyse medical scans in the US and the UK, but the retinal scan device would be its first stand-alone device.

Speaking at last week’s event, Keane called DeepMind’s results “jaw-dropping”, and predicted that “ophthalmology will be the first field transformed by AI”.

The system is designed to supplement human practicioners, rather than replacing them, in an environment in which eye scans quickly are becoming more prevalent, but the availability of the human experts needed to interpret them is not.

It is intended to refer urgent cases to specialists as quickly as possible, while reducing the number of false positives, which Keane said was currently creating a “huge logistical challenge” for the hospital.

DeepMind’s planned headquarters in London’s Kings Cross. Image credit: DeepMind

London expansion

DeepMind project lead and senior clinician scientist Alan Karthikesalingam told the Wired event the tool was an “opportunity to make the lives of clinicians easier”.

It also represents an effort by Google to get an early lead in the area of AI healthcare services, which are expected to be worth $6.6 billion (£5bn) in two years’ time.

Google said in November it would transfer control of DeepMind to a new Google Health division in California, as it invests more in commercialising its medical research efforts.

DeepMind is, however, intending to retain and expand its London base, with plans to move into a new flagship building in the Kings Cross area in the first half of next year.

“London is the artificial intelligence capital of Europe, with the capital housing more than twice as many companies as its closest rivals, Paris and Berlin, combined,” said Laura Citron, chief executive of London & Partners, when the new headquarters was announced last month.