Abstract
Purpose :
To investigate the potential of using computer-based hypertensive retinopathy (HR) detection software as a tool in estimating the 10-year cardiovascular disease (CVD) risk.
Methods :
A clinical study was performed to acquire retinal fundus images and clinical data from patients with chronic hypertension and potential risk of CVD, as determined by a primary care physician (PCP). The PCP’s determination was based on evaluation of risk factors such as obesity, smoking, blood pressure, body-to-mass index, cholesterol levels, lipid panel, family history of CVD, and pre-existing conditions. The clinical data acquired from patients included demographics, health record, disease history, and metabolic panel tests based on blood and urine samples. The clinical data was utilized to calculate the 10-year risk of CVD using the ASCVD clinical calculator, developed by the American College of Cardiology and American Heart Association.
The automated HR detection software, “CARV” (Comprehensive Assessment of Retinal Vessels), was developed for the identification of HR abnormalities such as artery-venous ratio, vessel tortuosity, copper/silver wiring, and retinal emboli. A partial least square classifier combined the individual algorithms to quantify the likelihood of presence of HR. The software performance was tested against the ground truth provided by a retinal grader for presence or absence of HR, to calculate sensitivity and specificity. Pearson’s correlation was calculated between the 10-year CVD risk estimation and HR likelihood results provided by the software and the retinal grader.
Results :
A dataset of 25 patients was utilized, with N=10 patients with HR and N=15 controls with no retinopathy. The CARV software achieved sensitivity of 90% and specificity of 73% in detecting HR against the ground truth provided by the retinal grader. The Pearson’s correlation coefficient between the 10-year ASCVD risk estimation and the likelihood of HR as determined by the software, was 0.64. The correlation between the retinal grader and the 10-year ASCVD risk estimation was 0.62.
Conclusions :
Our results show significant correlation between detection of HR and estimated risk of CVD, demonstrating the potential of HR detection system, as a tool for estimating the associated risk of CVD, which may aid in improving the screening of HR as well as predicting the future complications occurring due to CVD.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.