AI predicts risk of kidney disease from eye-screening images

Researchers at the Universities of Dundee and Glasgow have developed an innovative approach to predicting whether people with type 2 diabetes are likely to develop chronic kidney disease. Insights into kidney health are provided using artificial intelligence (AI) to analyse images taken during routine diabetes eye screening. This enables kidney problems to be detected long before current tests can identify them or before symptoms arise.

Diabetic kidney disease can often go undetected until it becomes severe. It represents a great burden for the individuals affected and to health services. One in five people with diabetes needing treatment for kidney disease within their lives.

The retina provides the only opportunity for the fragile networks of capillaries that are critical to the health of organs to be conveniently visualised. In the UK, everyone with diabetes aged 12 years and above is invited to attend regular eye screening, during which photographs are taken of the retina to spot signs of damage.

The researchers took nearly one million screening photographs from 100,000 people with type 2 diabetes and linked them with existing data on kidney health. The AI tool was trained to distinguish between images from people with and without kidney disease. The tool was then validated with data from around 30,000 other people with type 2 diabetes.

It was able to detect existing kidney disease with 86% accuracy. Furthermore, in people without kidney disease, it was able to predict who would go on to develop it in the next 5 years with 78% accuracy. By detecting future risk where standard kidney function tests provide no warning, it is hoped that AI will allow for earlier interventions, so that kidney disease progression can be slowed or halted.

By detecting future risk where standard kidney function tests provide no warning, it is hoped that AI will allow for earlier interventions, so that kidney disease progression can be slowed or halted.