Google, Verily use machine learning to detect diabetic eye disease

For the past three years, Google and Verily—Alphabet’s life sciences and healthcare arm—have been researching the use of machine learning to screen for diabetic retinopathy, a leading cause of preventable blindness in adults.

The result: A machine learning-enabled screening tool that is now in real-world clinical use through Verily and Google’s retinal diagnostic program at Aravind Eye Hospital in Madurai, India, they revealed this week.

The two companies been conducting a global clinical research program with a focus on India where studies demonstrated their algorithm performed on par with general ophthalmologists and retinal specialists assessing the images for disease, according to a Verily blog post.

Verily received the CE marking for the algorithm, signifying researchers have met the European Union Directive’s standards for medical devices and validating their approach, wrote Sunny Virmani, product manager at Verily and Kasumi Widner, program manager at Google.

The machine learning algorithm is being applied to screen for both diabetic retinopathy and diabetic macular edema, which can cause blindness but are preventable if caught early enough, they said. 

The screening tool could be particularly valuable in India, where there is a shortage of more than 100,000 eye doctors and only 6 million out of 72 million people with diabetes are screened—which leads to many individuals going undiagnosed and untreated, Google and Verily said. Technology could help to fill the gap.

Thousands of patients come through the doors of Aravind Eye Hospital and vision centers every day.

According to a Google blog post, Dr. R. Kim, chief medical officer and chief of retina services, said that by integrating the machine learning algorithm into their screening process, physicians have more time to work closely with patients on treatment and management of their disease while increasing the volume of screenings that can be performed.

At Aravind Eye Hospital, one of the sites where the technology was tested, officials trained technicians to take one image per eye. The algorithm assesses the image for DR and DME. The results offer quick feedback so the technicians can immediately determine whether patients need to be referred to an eye care physician, reducing the number of individuals who go without screening or clear follow-up, Virmani and Widner wrote.

An article in The Wall Street Journal back in January tracked the progress of the technology project and noted the companies ran into challenges transferring the technology from the lab to the doctor’s office.

According to the WSJ article, Google’s Automated Retinal Disease Assessment (ARDA) screening tool detects signs of diabetic retinopathy—lesions, burst blood vessels and patches of yellow on the retina—and grades their severity.

“While ARDA is effective working with sample data, according to three studies including one published in the Journal of the American Medical Association, a recent visit to a hospital in India where it is being tested showed it can struggle with images taken in field clinics. Often they are of such poor quality that the Google tool stops short of producing a diagnosis—an obstacle that ARDA researchers are trying to overcome,” the WSJ reported.

The two companies see potential uses for the machine learning algorithm in other areas of the world where there are not enough eye doctors to screen a growing population of people with diabetes. Researchers have launched a prospective study with Rajavithi Hospital, which is operated by the Ministry of Public Health in Thailand, studying the impact of the algorithm in a national DR screening program.

“We are also collaborating with Nikon and its subsidiary Optos to explore additional geographies and settings where machine learning tools for screening for diabetic eye disease could have an impact,” Virmani and Widmer wrote.