Google Sets Sights On Frontier Of Artificial Intelligence: Preventing Blindness
By Catherine Ho and Wendy Lee, SAN FRANCISCO CHRONICLE
May 7, 2018
Scientists in Google’s health division are developing technology they believe can help doctors better diagnose, treat and prevent vision impairment caused by diabetic retinopathy — a common eye disease among diabetics that can lead to blindness.
The technology is a cloud-based algorithm that analyzes photographic images of the eye for signs of the disease and grades the severity of the problem. Signs include abnormal growth of blood vessels in the back of the eye that causes scarring and detachment of the retina.
The algorithm is being tested in clinical trials and has not been approved by federal regulators. It is considered a medical device and would have to be approved by the U.S. Food and Drug Administration to be used in clinical care.
Google would not be the first to bring such a device to the masses. Last month, the FDA approved a similar device, called IDx-DR, that uses an artificial intelligence algorithm to analyze images of the eye to help determine the severity of diabetic retinopathy. It is made by an Iowa company, IDx, and is the first medical device approved by the FDA that uses artificial intelligence to detect the level of the disease in adults with diabetes.
Google’s tool is meant to be used by primary care physicians who frequently see patients with diabetes, to help catch the eye disease in its earlier, more treatable stages. By the time symptoms start to show, the chances of eventual blindness are much higher.
“The problem is people who have diabetes don’t get regular eye exams from an optometrist or ophthalmologist,” said Dr. Jorge Cuadros, an optometrist based in San Jose who is working with Google to test the algorithm. “What’s happening is they get a diabetic eye disease and it gets to the point where it starts to affect their vision, then they go see an optometrist or ophthalmologist.”
Cuadros is CEO of a telemedicine company, EyePACS, which supplied tens of thousands of images of patients’ eyes to Google to test its algorithm. Artificial intelligence is more than 90 percent accurate when diagnosing diabetic retinopathy — which is better than humans, Cuadros said.
The images were collected from patients who gave consent for their information to be studied, Cuadros said. The images are anonymized.
Diabetes is the main cause of blindness among people between the ages of 25 and 60, and it can largely be prevented by managing one’s blood sugar and by early detection. About one third of diabetics over age 40 have diabetic retinopathy, according to the U.S. Centers for Disease Control and Prevention.
Greg Corrado, a chief scientist who leads Google’s efforts in health and artificial intelligence, said he sees growth areas where machine learning can be applied in diagnostics, helping doctors make better decisions about treatment and in personalized medicine. He believes the technology may be maturing to the point where it can demonstrate scientifically that there is “real clinical impact.”
“If your phone can use AI to recognize every dog breed on the planet instantly, shouldn’t we also be asking how this technology can help doctors?” Corrado said.
Corrado said the technology used for diabetic retinopathy is being tested by doctors in the Bay Area as well as hospitals in India.
He doesn’t think the technology would eliminate jobs. He said it’s similar to when spreadsheets were introduced to accountants — disruption occurred, and aspects of the job changed, but it didn’t get rid of the field of accounting.
“The point of these technologies should be that they feel like an extension of the doctor’s ability and the extension of the nurse’s ability to care for patients, make good decisions and make good judgments,” Corrado said.
Google said its team working on health care and AI is independent of other teams focused on advertising at the company, and there are no plans to combine those efforts.
But John Simpson with privacy advocacy group Consumer Watchdog says he is “fundamentally skeptical” about how Google could use the information it collects, considering that much of its revenue comes from advertising.
“I don’t think you can trust Google when they go into personal health information at all,” Simpson said.