One of the many complications of diabetes is diabetic retinopathy (DR). This condition results in damage to the retina that can eventually lead to the loss of sight. And the longer a patient has diabetes, the more likely they are to develop DR. It is the leading cause of blindness for people between the ages of 20 and 64. According to some sources, as many as 95% of all cases could avoid loss of vision through early treatment. The problem is that screening requires an expert clinician who carefully examines images of a patient’s retina. The process requires specialized training, and is time-consuming and subjective.
IBM researchers have taught a computer system to make a DR diagnosis on its own. Using a deep learning artificial intelligence (AI) system, they used more than 35,000 sample retina images in the EyePACS database. The system was trained to identify lesions and other signs of damage, and to make an assessment based on a five-level clinical scale: no DR; mild; moderate; severe; proliferative DR. In testing the system, the researchers found that it achieved an 86 percent accuracy score in classifying the severity of DR, and it only took about 20 seconds to analyze an image.
Automating the screening process for DR among diabetic patients could lead to fewer cases of vision loss, which would reduce healthcare costs and improve the quality of live for these patients. It’s possible that patients might be able to take their own images of their retinas with a smartphone app, and submit them to a machine system for initial screening.
