Inter- and Intra-Grader Reliability of Computer-Assisted Measurement of Retinal Vascular Geometry

Abstract

Purpose: : Changes in the retinal vasculature are markers of systemic and ocular vascular diseases. We developed a semi-automated retinal vascular imaging program that quantitatively measures retinal vascular geometric parameters, and reported inter- and intra- grader reproducibility of the measures.

Methods: : This computer-based program quantifies from retinal images a range of retinal vascular parameters (retinal arteriolar and venular caliber, branch angles, and simple and curvature tortuosity). Two trained graders independently graded twice for 23 right eye images of participants from the Singapore Malay Eye Study according to a standardized protocol. Inter- and intra-grader intraclass correlation coefficients (ICC) were estimated. The Bland-Altman plots were used to depict limits of agreement by showing mean difference with 95% confidence interval.

Results: : Means (±;SD) of the parameters measured within a zone ½ to 1 disc diameter away from the optic disc margin by graders A and B were: central retinal arteriolar equivalent (136.4±;13.9 and 136.2±;13.5 µ) and central retinal venular equivalent (207.2±;18.4 and 204.0±;14.0 µ). Corresponding means (±;SD) for parameters measured not limited to any zone and by the two graders were: arteriolar (80.4±;10.3° and 83.3±;10.2°) and venular (80.6±;9.3° and 80.7±;9.5°) branching angle; arteriolar (1.100±;0.022 and 1.103±0.024) and venular (1.107±;0.018 and 1.106±;0.021) simple tortuosity; and arteriolar (0.030±;0.009 and 0.041±;0.011) and venular (0.034±;0.11 and 0.040±;0.011) curvature tortuosity. Intra-grader ICC ranged from 0.77-0.99 for all parameters, and inter-grader ICC ranged from 0.77-0.97 in most parameters except for venular branching angles (ICC 0.60) and arteriolar curvature tortuosity (ICC 0.63). Limits of agreement were within 15% of mean value for retinal vessel diameters and simple tortuosity, but within 18.5% to 58.1% of the means for branching angles and curvature tortuosity.

Conclusions: : Quantitative assessment of retinal vascular geometry characteristics from retinal images using our new image program shows good intra- and inter-observer reproducibility on most measures. These parameters may be markers of vascular disease.

Keywords: retina • imaging/image analysis: clinical • clinical (human) or epidemiologic studies: systems/equipment/techniques