Distribution and determinants of choroidal vascularity index in healthy eyes from deep-learning choroidal analysis: a population-based SS-OCT study.

Abstract

To quantify the profiles of choroidal vascularity index (CVI) using fully artificial intelligence (AI)-based algorithm applied to swept-source optical coherence tomography (SS-OCT) images and evaluate the determinants of CVI in a population-based study. This cross-sectional study included adults aged ≥35 years residing in the Yuexiu District of Guangzhou, China, a follow-up population-based study. All participants (n=646) underwent comprehensive ophthalmic examinations, including SS-OCT for quantifying choroidal parameters. The CVI and subfoveal choroidal thickness (SFCT) were measured by a novel AI-based system. A total of 556 participants were included, with a mean age of 56.4±9.9 years and 44.96% women. The average CVI and SFCT of the overall population were 69.7% (95% CI 69.2 to 70.3) and 263.0 µm (95% CI 257.2 to 268.8), respectively. After adjusting for other factors, older age and longer AL were significantly associated with a lower CVI. The CVI decreased by -0.13% (-0.19 to -0.06, p<0.001) with each 1-year increase in age, -2.10% (-3.29 to -0.92, p=0.001) with each 1 mm increase in AL. Furthermore, significantly positive correlation between CVI and SFCT has been observed, with coefficient of 0.059 (0.052 to 0.065, p<0.001). Using new AI-based choroidal segmentation software, we provided a fast, reliable and objective CVI profile for large-scale samples. Older age and longer AL were independent correlates of choroidal thinning and CVI decline. These factors should be considered when interpreting SS-OCT-based choroidal measurements.