Development of AI to Predict Chronic Kidney Disease Risk from Retinal Examination Results

An artificial intelligence (AI) capable of predicting the risk of chronic kidney disease based on retinal examination results has been developed.

Professor Joo Young-soo of the Department of Nephrology at Yongin Severance Hospital, Professor Park Jung-tak of the Department of Nephrology at Severance Hospital, and the research team led by Lim Hyung-taek, Chief Medical Officer at Mediwhale, announced on the 30th that they have created an AI that predicts the likelihood of chronic kidney disease occurrence through retinal examination.

Chronic kidney disease is usually diagnosed when there is a persistent decline in kidney function or proteinuria lasting more than six months. The main causes are diabetes and hypertension. Diabetes and hypertension cause damage to kidney tissues and blood vessels, leading to chronic kidney disease. It is often called a silent disease because symptoms frequently do not appear until it reaches a severe stage. When the glomerular filtration rate falls below 60, it corresponds to severe chronic kidney disease and requires active treatment. However, since there is no clear treatment to restore already deteriorated kidney function, prevention is most important.

Until now, the risk of onset has been assessed by estimating the glomerular filtration rate through blood tests. The method of estimating the glomerular filtration rate is easily influenced by external factors such as age and physical activity, making it difficult to measure risk in people with normal kidney function.

The research team developed an AI that can predict the risk of chronic kidney disease through retinal examination and analyzed its effectiveness. They focused on the fact that the retina, like the kidney, is composed of microvessels and that fundus examination can easily provide results. They also considered the accessibility advantage, as diabetic patients, a major cause of chronic kidney disease, need to undergo regular retinal examinations to check for complications.

Actual incidence of chronic kidney disease according to the risk level of chronic kidney disease occurrence analyzed by the AI developed by the research team. The yellow solid line represents the highest risk group. [Data provided by Yonsei Medical Center]

The research team first trained the AI with retinal examination and glomerular filtration rate data from 80,000 people screened at Severance Hospital, then created an algorithm to analyze the risk of chronic kidney disease occurrence. The AI evaluates the risk of chronic kidney disease by examining the blood vessels in retinal images to predict the decline in glomerular filtration rate. The AI’s predictive utility was then validated on 35,000 subjects combining participants from the UK Biobank study and Korean diabetic patients. The retinal images of the subjects were reviewed to confirm the occurrence of chronic kidney disease over a period of up to 10.8 years.

As a result, when subjects were classified into four groups according to the risk of chronic kidney disease occurrence, the actual incidence of chronic kidney disease increased with the risk level. The AI showed higher accuracy than the glomerular filtration rate estimation method. Professor Park Jung-tak said, "It was difficult to predict future chronic kidney disease occurrence when kidney function was normal. With the AI developed this time, we have created an opportunity to prevent the disease by selecting high-risk groups for chronic kidney disease."

The results of this study were published in 'npj Digital Medicine' (IF 15.357), a prestigious international journal in the healthcare services field.

This content was produced with the assistance of AI translation services.