AI taught to recognise signs of diabetic eye disease
Image credit: Dreamstime
Stanford University researchers have used machine-learning techniques to create an algorithm capable of rapidly detecting early signs of diabetic retinopathy, a common eye disease that can lead to blindness.
Diabetes – a chronic condition that causes high blood-sugar levels – affects one in every 11 adults worldwide. A common complication is diabetic retinopathy, an eye disease that can lead to severe eye damage and even blindness.
In order to see, light must pass from the front of the eye through to the retina. When the network of blood vessels supplying the retina with blood becomes blocked or leaky, the retina is damaged and cannot work properly. However, most people are unaware of their condition until it is too late.
Ophthalmologists normally diagnose the condition by directly examining the back of the eye and looking at full-colour photographs of the eye’s interior lining. This approach has been shown in studies to be subjective, even among highly trained specialists, and is a tedious, expensive approach.
Given that 45 per cent of people with diabetes are at high risk of developing diabetic retinopathy, there is a need for a more efficient method of screening patients in order to pick up early signs of the disease.
Researchers at Stanford University’s Byers Eye Institute set about using deep-learning methods to create an automatic algorithm capable of detecting diabetic retinopathy. They fed the programme more than 75,000 images from a range of patients, categorising them as healthy, or in different stages of the disease. Using this training data, the algorithm became capable of detecting all disease stages with an accuracy of 94 per cent.
The algorithm does not require specialised computer equipment to review images, and can run on an ordinary PC or smartphone.
Patients showing signs of the disease could then be seen by an ophthalmologist for further examination and possible treatment.
“What we showed is that an artificial intelligence based grading algorithm can be used to identify, with high reliability, which patients should be referred to an ophthalmologist for further evaluation and treatment,” said Dr Theodore Leng, lead author of the study.
Machine learning – the field of computer science that develops algorithms capable of learning from data without being explicitly programmed – could have far reaching uses in medicine. Computers can be left to rapidly search through data and flag up potential cases in need of human investigation.
“If properly implemented on a worldwide basis, this algorithm has the potential to reduce the workload on doctors and increase the efficiency of limited healthcare resources,” said Dr Leng. “We hope this technology will have the greatest impact in parts of the world where ophthalmologists are in short supply.”