Automatic retinal image analysis identifies diabetes early with high sensitivity

Screening of undiagnosed diabetes can be performed effectively with an automatic retinal image analysis (ARIA) technology developed by researchers at the Chinese University of Hong Kong (CUHK), instead of a finger-prink test.

The ARIA technology uses machine learning–based assessment to analyze retinal images. It is a convenient, noninvasive and painless method that rapidly assesses the risk of diabetes by analyzing fundus images.

Among 2,221 cases studied, the ARIA technology showed a sensitivity of 92.0 percent, a specificity of 96.2 percent, and an overall accuracy of 94.4 percent for diabetes screening. [Thomas RL, et al, European Association for the Study of Diabetes (EASD) 2021 Annual Meeting, oral presentation 289]

These results suggest that the ARIA technology can be an effective tool for early detection of undiagnosed diabetes. To validate these findings, the researchers tested the ARIA technology in two separate cohorts.

The first cohort involved 1,394 individuals from Hong Kong, including 660 diabetes patients and 734 healthy individuals from the community. In this cohort, the ARIA technology showed a sensitivity and specificity of 99.5 percent and 91.1 percent, respectively.

The second cohort involved retinal images from 1,682 patients with diabetes from the UK, which showed a sensitivity of 98 percent with the ARIA technology.

The sensitivity of ≥98 percent in both cohorts suggests that the ARIA technology is effective for diabetes screening in both Asian and Caucasian populations.

“This technology offers a convenient and accurate method for diabetes screening in the general population, especially for those at high risk,” noted the researchers.

The ARIA technology has been studied extensively for more than 10 years. In Hong Kong, it is now available for risk assessment of stroke (ARIA-stroke risk) and for screening of white matter hyperintensities (WMH) as an early risk factor of dementia (ARIA-WMH).

Currently, the researchers are working on applying the ARIA technology in mental health disorders, coronary heart disease, anaemia, diabetic kidney disease, and screening of autism spectrum disorder (validation study).

“Our eyes provide tremendously important clues and signals about our health. What we need is the right technology for interpretation,” concluded lead researcher, Professor Benny Zee, Director of the Centre for Clinical Research and Biostatistics, CUHK. “The ARIA technology will become a standard risk assessment tool, promoting disease prevention and translating research results into applications that will have a substantial impact on society.”

This October, the ARIA technology along with 30 new research projects by the CUHK were on display in three local large-scale technology exhibitions, including InnoCarnival 2021, the International ICT Expo 2021, and the Hong Kong International Medical and Healthcare Fair 2021.