June 2023
Volume 64, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2023
Development of age prediction model using retinal fundus image and analysis of the results in patients with chronic disease
Author Affiliations & Notes
  • HYOJOO KANG
    Asan Medical Center, Songpa-gu, Seoul, Korea (the Republic of)
  • Miso Jang
    Asan Medical Center, Songpa-gu, Seoul, Korea (the Republic of)
  • Yoon Jeon Kim
    Asan Medical Center, Songpa-gu, Seoul, Korea (the Republic of)
  • Footnotes
    Commercial Relationships   HYOJOO KANG None; Miso Jang None; Yoon Jeon Kim None
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2023, Vol.64, 2396. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      HYOJOO KANG, Miso Jang, Yoon Jeon Kim; Development of age prediction model using retinal fundus image and analysis of the results in patients with chronic disease. Invest. Ophthalmol. Vis. Sci. 2023;64(8):2396.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Purpose : Aging is a significant risk factor for a wide range of chronic diseases. The aim of this study is to develop a deep learning (DL) model that predicts the biologic age from retinal fundus images (Reti-age) in healthy population and to investigate whether the difference between Reti-age and chronologic age reflects the health status of patients with chronic diseases.

Methods : A total of 200,820 normal fundus images taken from 125,408 participants in the health promotion center. Of these, 14,655 fundus images from 5,334 participants without medical history, with normal body mass index (BMI), HbA1c, lipid test and normal blood pressure (BP) were used to train and validate the DL model for age prediction. A regression model for age prediction was used ResNet50 with ImageNet pretrained weights. In addition, to analyze whether the prediction errors of Reti-age from the chronological age is correlated with the metabolic status, 14,657 patients who have chronic diseases were used. We classified the test dataset into three groups (based on the value of the Reti-age - chronological age (R-C) and compared the laboratory data of three groups: R-C>5 years, -5 years ≤ R-C ≤5 years, R-C <-5 years.

Results : The DL model achieved a strong correlation of R 2 0.91 between Reti-age and chronological age with an overall mean absolute error of 3.32 years. Older patients tended to have larger differences between Reti-age and chronologic age. Three groups classified according to (R-C) showed the significant differences in terms of age, BMI, diastolic BP (DBP), HbA1C, low density lipoprotein cholesterol (LDL-C), and triglyceride (TG). Patients predicted to be older more likely to be older in chronologic age and have higher BMI, DBP, HbA1C, LDL-C, and TG. In the meanwhile, patients predicted to be younger more likely to be younger in chronologic age and have lower BMI, DBP, HbA1C, LDL-C, and TG.

Conclusions : Using the Reti-age calculating DL model established with the normal fundus photography, we found that patients with chronic diseases have the larger errors between Reti-age and chronological age. Reti-age can be used as a useful biomarker in figuring out the patients at risk, particularly, patients with uncontrolled DM and hypertriglyceridemia.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.

 

 

×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×