Xie Wuxiang's team at the Clinical Research Institute combined artificial intelligence technology and fundus photo information to identify high-risk groups of dementia

On December 19, 2022, the team of Xie Wuxiang of Peking University Clinical Research Institute cooperated with researchers from Beijing Eagle Pupil Science and Technology Development Co.
, Ltd.
, University of Washington, PLA General Hospital, iKang Group, Beijing Tongren Hospital, Shanghai North Hospital and Beijing Anzhen Hospital to publish the latest research results
.
The study is based on the physical examination data of more than 290,000 people in 19 provinces and cities in China (modeling: 258,000, internal verification: 14,000, external verification: 2.
1 10,000 people), developed an artificial intelligence algorithm based on fundus photos to accurately estimate CAIDE (Cardiovascular Risk Factors, Aging, and Incidence of Dementia).
Dementia risk score to identify people at high risk of dementia, this is the first international study
to combine artificial intelligence technology and fundus photo information to identify people at high risk of dementia.

As the population ages, the burden of disease caused by dementia is expected to continue to grow rapidly
this century.
How to accurately and efficiently screen people at high risk of dementia from the population, so as to actively and effectively intervene early, is the key to the
success of dementia prevention and treatment.
Professor Kivipelto's team published the CAIDE Dementia Risk Score in the journal Lancet Neurology in 2006 and is currently international The most recognized dementia risk prediction tool, but its calculation requires blood collection and multi-dimensional health information, which is invasive, time-consuming, and not easy to be grasped by doctors, and it is difficult to promote and apply
.
Retinal microvascularity is the only blood vessel
in the human body that can be non-invasively observed with the help of fundus cameras.
Previous studies have found that various retinal microvascular indexes such as fundus arteriovenous diameter, arterial stenosis, and degree of retinopathy are significantly correlated with the incidence of dementia, and have the potential to be applied to the early identification
of high-risk groups of dementia.
In recent years, artificial intelligence fundus detection technology has matured, and fundus photos and result interpretation can be taken without a professional ophthalmologist, and has been recognized and promoted
in outpatient and community physical examinations.
Based on the above research background, Xie Wuxiang's research team cooperated with a number of studies to collect fundus photos and health-related data of 271,864 physical examination participants in 19 provinces (cities) across the country, and randomly divided them into development sets (95 %) and internal validation sets (5%), using convolutional neural network technology to develop artificial intelligence algorithms to estimate CAIDE dementia risk scores and in an independent external cohort (20,690 medical examination participants).
) to verify
it.

The results showed that in both internal and external validation, the algorithm could accurately identify people at high risk of dementia (defined as CAIDE dementia risk score ≥10 points), with AUCs of 0.
944 ( 95% CI: 0.
939–0.
950) and 0.
926 (95% CI: 0.
913–0.
939).

In addition, the research team evaluated the cognitive function
of 1512 middle-aged and elderly subjects in the externally validated population.
The results showed that after adjusting for a number of confounding factors such as lifestyle, genetic factors, and chronic disease prevalence, the higher the CAIDE dementia risk score estimated by the algorithm, the subjects' comprehensive cognitive function (MoCA) and cognitive function in various domains ( Memory function, executive function, attention) performance were significantly worse
.
The study provides a new idea for the early screening of dementia, and the artificial intelligence algorithm based on fundus photos has the advantages of fast (30-40 seconds), non-invasive, simple and cheap, and has been widely used in China's physical examination population to identify high-risk individuals
with dementia early.

The research results were published in Age and Ageing, an international high-level journal in the field of geriatrics.

Hua Rong, a master's student at the Clinical Research Institute of Peking University, and Dr.
Xiong Jianhao, an algorithm engineer at Beijing Eagle Pupil Science and Technology Development Co.
, Ltd.
, are co-first authors of this paper, and Professor Zeng Qiang, director of the Health Management Research Institute of PLA General Hospital, and researcher Xie Wuxiang are co-corresponding authors
of this paper.

About the author

Xie Wuxiang, researcher, doctoral supervisor
.
He has been engaged in clinical research and population research of heart and brain diseases for a long time
.
As first or corresponding author (including co-author) in JACC (2019, 2022), Science Bulletin (2021, 2022), JAMA Network Open (2020, 2022) and other journals published more than 50 academic papers
。 He has presided over the National Natural Science Foundation of China, the Newton International Scholar Fund of the British Academy of Medical Sciences, and the Irma and Paul Milstein Geriatric Health Special Research Funds
in the United States.
He currently serves as the Secretary-General of the Health Risk Assessment and Control Professional Committee of the Chinese Preventive Medicine Association, the executive editorial board member of Science Bulletin magazine, and the deputy editor-in-chief
of Frontiers in Cardiovascular Medicine.

(Clinical Research Institute)