How deep learning algorithms could detect diabetic eye disease

Researchers at the Byers Eye Institute at Stanford University believe deep learning algorithms, a form of artificial intelligence, hold the key to detecting diabetic retinopathy and averting blindness in patients.

Diabetic retinopathy damages the blood vessels at the back of the eye and is the leading cause of preventable blindness globally. Up to 45% of diabetic patients are likely to have diabetic retinopathy at some point in their life, but fewer than half of patients are aware of the condition. Early detection and treatment are integral to combating this worldwide epidemic of preventable vision loss.

The team created deep learning algorithms based on more than 75,000 images from a range of patients representing several ethnicities. They then used the algorithms to teach a computer to discern healthy patients, and mild to severe stages of the disease. The study, which was published online in Ophthalmology, the journal of the American Academy of Ophthalmology, showed 94% accuracy in detecting mild to severe symptoms of the disease. Those patients with any stage of the disease can be referred to an ophthalmologist for evaluation and study.

Diabetic retinopathy isn’t the only area where artificial intelligence could help with diagnosis. Stanford University researchers, for example, say that they’ve trained a deep learning algorithm to identify skin cancer as well as dermatologists. And IBM has been touting its Watson supercomputer as a tool to assist health providers with diagnoses.

Ophthalmologists typically diagnose the presence and severity of diabetic retinopathy by examining the back of the eye and evaluating color photographs of the interior lining of the eye. Given the large number of diabetes patients globally, this process is expensive and time-consuming. Also, previous studies have shown that detection is somewhat subjective, even among trained specialists. This is why an effective, automated algorithm could potentially reduce the rate of worldwide blindness.

The researchers believe the method could reduce the workload of doctors and increase efficiency in areas with limited healthcare resources. It can be run on a personal computer or smartphone and doesn’t require specialized computer equipment

The deep learning algorithms still need pilot studies, which the team is planning for later this year. The system will eventually require FDA approval.

[Want to stay more on top of MDO content? Subscribe to our weekly e-newsletter.]

Engineering Better Devices at DeviceTalks Minnesota DeviceTalks Minnesota's engineering track is designed to provide attendees with with actionable lessons in a peer-to-peer format on topics such as: Human factors engineering Patient-centric device development R&D trends Manufacturing and scalability Navigating the product development process from start to finish "How We Built This" panels with top medical device groups Join industry leaders and medical device professionals at the most awaited medtech conference of the year, DeviceTalks Minnesota. Use code DT15 to save 15%! REGISTER NOW