Hypertension, diabetes affect age prediction from retinal fundus imaging
Click Here to Manage Email Alerts
A convolutional neural network model accurately and reliably predicted age and sex using retinal fundus images; however, accuracy dropped among patients with underlying systemic vascular-related conditions.
“We expected the [convolutional neural network (CNN)] to find the proper feature for the task from the retinal fundus images and to produce accurate results,” Yong Dae Kim, MD, from the Seoul National University Bundang Hospital in the Republic of Korea, and colleagues wrote.
The study comprised 219,302 fundus images from healthy patients and patients with hypertension, diabetes and any smoking history. The CNN accurately predicted age in healthy patients with a mean absolute error (MAE) of 3.06 (R2 = 0.92). Mean absolute error was similar but had a lower correlation among those with hypertension (R2 = 0.74), diabetes (R2 = 0.75) and smoking history (R2 = 0.86).
Results showed acceptable accuracy for fundus-predicted sex (area under curve higher than 0.96) in all sets.
“Retinal fundus images from participants with underlying vascular-altered conditions (hypertension, DM or smoking) indicated similar MAEs and low coefficients of determination between the predicted age and chronologic age, thus suggesting that the ageing process and pathologic vascular changes exhibit different features,” Kim and colleagues wrote. “In the process of fundus change, systemic vascular diseases are thought to have a different effect from aging.”
Additionally, a subgroup analysis revealed that MAEs increased and accuracy declined significantly among patients older than 60 years in both the healthy patient group and those with vascular-altered conditions.
“These results suggest that pathologic retinal vascular changes occurring in systemic vascular diseases are different from the changes in spontaneous aging process, and the aging process observed in retinal fundus images may saturate at age about 60 years,” the researchers concluded. “Deep learning-based fundus image reading may be a more useful and beneficial tool for screening and diagnosing systemic and ocular diseases after further development.”
Perspective
Back to Top
Kuniyoshi Kanai, OD, FAAO
Eyes are a window to the body. By looking at the fundus, we may be able to extract information about the individual’s current health status and even to predict future health events. This concept has been validated for decades by many community-based cohort studies. For example, retinal arterial caliber is known to have an association with future cerebral vascular events. However, despite numerous data, this idea has not been put into use in medical practice.
Artificial intelligence is an emerging tool which can assess fundus images differently than a human examiner. After going through the “learning” process, the program can potentially provide a wealth of information about the individual from a fundus image, even without knowing a patient’s systemic history. This study looked at the specific effect of diabetes, hypertension and smoking on the artificial intelligence’s ability to predict the individual’s age and sex. The study concludes that more work is needed to refine its ability. Though we are not at the stage of the everyday application of artificial intelligence to predict health risks from fundus images, a study like this could lead to the eventual implementation of fundus imaging for health assessment in various medical fields.
Collapse
Read more about
Click Here to Manage Email Alerts