Correlation and Reproducibility of Semi-Automated Retinal Vascular Geometric Measurements Within Paired Stereoscopic Images

Abstract

Purpose: : Retinal vasculature geometry may be a marker of pre-clinical microvascular disease. We aimed to assess the robustness and reproducibility of a new computer imaging program in measuring retinal vascular geometry using retinal images of the same eye taken with different photographic angle of incidence.

Methods: : We used 30 stereoscopic pairs of colour optic disc centred photographs of the Blue Mountain Eye Study to assess the following geometric parameters of retinal arterioles and venules, using a new semi-automated software (SIVA) following standardized protocol: 1) retinal arteriolar/venular caliber (CRAE/CRVE); 2) arteriole-to-venule ratio (AVR); 3) branching angle; and 4) tortuosity. We used Pearson correlation (r) and Intra-class Correlation Coefficient (ICC) to assess within-pair correlation and agreement for each parameter measure.

Results: : Overall, moderate to high within-pair correlation and agreement was found, with r= 0.992, 0.995 and 0.994, and ICC= 0.997, 0.998, and 0.997 for CRAE, CRVE, and AVR, respectively. Moderate correlation (r= 0.621 and 0.492) and reproducibility (ICC=0 0.741 and 0.656) were found for arteriolar and venular branching angle, respectively. Moderate to high correlation (r = 0.981 and 0.953) and reproducibility (ICC = 0.980 and 0.957) were found for arteriolar and venular tortuosity, respectively. Excluding poorly focussed images showed identical results except for arteriolar and venular branching angles, which became less reproducible.

Conclusions: : Our new semi-automated software results in robust measures to the difference in photographic angle of fundus images. This property of the image program allows investigations of changes in retinal geometry over time in longitudinal studies.

Keywords: retina