Better eye screening

Image from retinal scan

“What is really needed, if you are thinking about prevention, is to diagnose the cases in-between. The average consistency of eye specialists to detect DR is 65%, and if you remove the obvious cases, it is not that different from flipping a coin.”

– IEOR professor Xin Guo

Eye exams at Chinese hospitalEye exams being performed at a hospital in China. (Photos courtesy the researchers)Diabetic retinopathy (DR) is the most common cause of vision loss among people with diabetes, and a leading cause of blindness. But thanks to cutting-edge machine learning techniques developed at the Risk Analytics and Data Analysis Research (RADAR) Lab, led by Xin Guo, professor of industrial engineering and operations research, millions of diabetic patients now have a cheaper and more accurate way to screen for eye disease.

Problem: Worldwide, about one-third of the estimated 415 million people with diabetes have DR. While blindness caused by DR is mostly preventable with early detection and treatment, it is difficult to diagnose, and many patients lack access to eye specialists. Moreover, diagnoses can be very inconsistent among eye specialists.

Solution: Using machine learning techniques, Guo and her team developed algorithms to detect features in retinal images and help with diagnosis. To train their algorithm, they worked with hospitals in China that provided over 100,000 retinal images from patients. A team of expert doctors labeled the images as healthy or diseased.

Result: The system can now detect DR with better than 97% accuracy, and researchers hope to apply these techniques to diagnosing other major eye diseases.

Impact: Over 100 systems that assist doctors in diagnosing DR have been installed in some of the poorest rural areas in China. Up to 3 million people, most of whom had no access to regular eye checkups, have received free diagnoses from these systems since 2016. Guo is currently working to expand the remote diagnosis program into other regions.


Topics: Industrial engineering, Health, Research


Reach the editors at berkeleyengineer@berkeley.edu