Characterisation of geometrical attributes of retina vasculature in healthy elderly individuals

Abstract

Purpose : Quantification and characterisation of the geometrical attributes of retina vasculature affords an opportunity for vascular disease risk assessment and potentially the monitoring of disease progression and as marker of therapeutic efficacy. In this study we utilize a newly developed tool for automated measurement of retina vascular parameters and describe higher order statistical distribution characteristics of these parameters in a large population database of elderly individuals from Personality and Total Health (PATH) Through Life study.

Methods : Optic disk centred images of 430 eyes of 240 healthy individuals from wave 4 of the 60’s cohort of PATH were taken. Subjects were 60.7% female aged 72 to 78 (74.7±1.36). Retinal vasculature parameters including the Central Retinal Artery Equivalent (CRAE) and Central Retinal Vein Equivalent (CRVE), arteriolar-venular ratio (AVR) were estimated with IVAN. The mean simple tortuosity (ST), total number of branching points (BP), average acute branching angle (ABA) and box-counting fractal dimension (FD) were measured using the automated Retinal Image Vasculature Assessment Software (RIVAS) platform. FDs were measured by taking three different approaches: binarized box count (BBC), skeletonized image (SI) and SI without cross overs (SINC). Sampling distributions of the parameters were estimated with goodness of fit test package (GOFT) in R to determine the best fitting statistical distribution for each parameter using normal (N), inverse Gaussian (IG), Generalized Pareto (GP) and beta distributions (BD) models.

Results : Only CRVE (169.94±24.49, R=0.99, p=0.49) and ABA (73.46±5.71, R=0.99, p=0.14) were normally distributed. A Beta distribution with shape parameters of α=24.8, β=140.6 was observed for SINC (1.14±0.02) after applying appropriate transformation to range between 0 and 1. For the shape parameter γ≥0, the test statistics revealed a GP distribution for CRAE (128.8±19.98, R=0.92, p=0.45), AVR (0.76±0.11, R=0.95, p=0.66), ST (1.08±0.01, R=0.96, p=0.95), BBC (1.45±0.05, R=0.94, p=0.92) and SI (1.22±0.05, R=0.94, p=0.86). No suitable distribution function was found for BP which had a bi-modal histogram.

Conclusions : The software was effective in batch processing a large population database. Only CRVE and ABA were normally distributed. These findings need to be considered in choice of statistical tests when considering studies utilizing these measures.

This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.