London-based DeepMind, owned by Google’s parent company, Alphabet, focuses heavily on the specifics of using artificial intelligence in health care, and on Monday it released a study showing the progress it’s made in using AI to diagnose eye conditions.
Published in the science journal Nature Medicine, the study reports that DeepMind, in partnership with Moorfields Eye Hospital in London, has trained its algorithms to detect over 50 sight-threatening conditions to the same accuracy as expert clinicians. It’s also capable of correctly recommending the most appropriate course of action for patients and prioritize those in most urgent need of care.
In a project that began two years ago, DeepMind trained its machine learning algorithms using thousands of historic and fully anonymized eye scans to identify diseases that could lead to sight loss. According to the study, the system can now do so with 94 percent accuracy, and the hope is that it could eventually be used to transform how eye exams are conducted around the world.
AI is taking on a number of roles within health care more widely. In June, Babylon Health said that it gave its artificial intelligence technology the same test required of would-be general practitioners in Britain and that the AI performed better than humans. In March, researchers found that machine learning can classify heart anatomy on an ultrasound scan better than a human. AI is also being used to help emergency call dispatchers in Europe detect heart attack situations.
Diagnosing eye diseases from ocular scans is a complex and time-consuming for doctors. Also, an aging global population means eye disease is becoming more prevalent, increasing the burden on healthcare systems. That’s providing the opportunity for AI to pitch in.
“The number of eye scans we’re performing is growing at a pace much faster than human experts are able to interpret them,” said Pearse Keane, consultant ophthalmologist at Moorfields, in a statement. “There is a risk that this may cause delays in the diagnosis and treatment of sight-threatening diseases, which can be devastating for patients.”
Using AI instead could mean earlier diagnoses for patients and therefore earlier treatment, leading to less deterioration in eyesight down the line. “It gives us the best chance of saving people’s sight,” said Keane.
DeepMind’s AI has been trained using one particular type of eye scanner, but researchers say it’s compatible with any model. Not only does this mean it can be used widely and without hardware restrictions, but that it will remain useful in the future even when equipment is replaced and updated.
The AI can also explain to doctors how it arrived at a particular decision, which will allow the doctors to scrutinize whether it’s made the right call before going ahead with treatment.
Before the AI can be used in hospitals to diagnose real patients it must now go through clinical trials and gain regulatory approval.