Neural-network-measured retinal vascular area is associated with pulse wave velocity

Abstract

Purpose : Retinal vessels reflect the alteration of physical status related to hypertension and arteriosclerosis. Previously, we reported a deep learning algorithm to detect retinal vessels full-automatically and to measure the total retinal vascular area in fundus images (Fukutsu et al. Ophthalmology Science. 2021). In the present study, we investigated the relationship between the total retinal vascular area and brachial-ankle pulse wave velocity (baPWV), which is known as a gold standard for arterial stiffness assessment in clinical practice.

Methods : Retinal photographs (n=696) obtained from 372 individuals who visited Keijinkai Maruyama Clinic for a regular health checkup were used to analyze the total retinal vascular area. PWV was also measured in each patient. Automatic retinal vessel segmentation was performed using our deep learning algorithm, and the total arteriole area (AA) and the total venule area (VA) were measured. The correlation between PWV and several parameters including AA and VA were assessed.

Results : The baPWV correlated negatively with AA (R = -0.40, n = 696, P < 2.2e-16) and VA (R = -0.36, n = 696, P < 2.2e-16). Each of the independent variables (AA, gender, age, and systolic blood pressure) selected using the stepwise method showed a significant correlation with the value of baPWV. The predicted value of baPWV, calculated from regression equation using the variables including AA, showed a better correlation with the measured baPWV value (R=0.70, n=696, P < 2.2e-16) than the predicted value without AA (R=0.68, n = 696, P < 2.2e-16).

Conclusions : AA and VA showed a significant correlation with baPWV. In addition, the predicted value of baPWV using AA well correlated with the actual baPWV value. The present data indicate that the retinal vascular area serves as an alternative biomarker for evaluating arteriosclerosis.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.