AI can predict diabetic kidney disease risk from retinal changes

AI analysis of routine eye screening images accurately predicts risk of developing kidney disease in the next five years, before symptoms develop or it can be detected by current tests.

In the new study, presented at Diabetes UK’s Professional Conference 2025, researchers at the Universities of Dundee and Glasgow explored whether AI could unlock hidden clues in images collected during annual retinal screening, which is recommended for everyone living with diabetes over the age of 12.

The team, led by Dr Alexander Doney, developed the AI tool using nearly 1 million eye screening photographs from almost 100,000 people with type 2 diabetes in Scotland. Photographs were linked with existing data on kidney health, and the AI tool was trained to distinguish between images from people with or without kidney disease. The tool was then validated with data from almost 30,000 other people with type 2 diabetes.

The AI tool detected existing kidney disease with 86% accuracy. In people without kidney disease, it was also able to predict who would go on to develop it in the next five years with 78% accuracy. Critically, the AI outperformed traditional kidney function tests, detecting future kidney disease risk in individuals where standard testing provided no warning.

Diabetes-related kidney disease can develop silently over many years, often going undetected until it becomes severe. One in five people with diabetes will need treatment for kidney disease during their lifetime. Diabetes is a leading cause of end-stage kidney disease, with almost one in three people who need dialysis or a kidney transplant having diabetes.