Abstract
Purpose :
The aim of this study was to create a gold-standard set of images and test an algorithm — A3ID — to automate, accelerate and standardize the process of avascular area segmentation in images from a rat oxygen-induced retinopathy (OIR) model.
Methods :
Within 6 hours of birth, full term pups born to Sprague Dawley dams that had undergone partial bilateral uterine artery ligation at embryonic day 19.5 were placed into a controlled oxygen environment (Oxycycler, Biospherix) at 50% oxygen for 48 hours, followed by cycling between 10% and 50% oxygen every 24 hours until day 15. Pups were then moved into room air until day 18.5. Ten lectin-stained retinal flat mounts were imaged in montage fashion at 10x magnification. Three human reviewers measured two parameters for each image using the Image J freehand selection tool: total retinal area (TRA) and total avascular area (AVA). The outputs of each read were measured as number of pixels. The gold standard (GS) value for each image was the mean of the three human reads. Interrater agreement for the measurement of TRA, AVA, and percent avascular area (%AVA) was calculated using type A intraclass correlation coefficients (ICC) with a two-way random effects model.
Automated Avascular Area Identification (A3ID) — is a method, written in ImageJ Macro, intended for use in the FIJI (FIJI is Just ImageJ) image processing platform. The input for A3ID is a rat retinal image with the ciliary body cropped. The output is the TRA and AVA (in pixels). A3ID utilizes a random-forest classifier with a connected-components algorithm and post processing filters for size and shape. We compared A3ID output to the GS by calculating ICC, performing linear regression, and constructing a Bland-Altman plot.
Results :
ICC for %AVA between human readers was 0.995 (CI: 0.974-0.999) p <0.001. ICC between A3ID and the GS was calculated for three image parameters: AVA:0.974 (CI: 0.899 - 0.993) p<0.001; TRA: 0.465 (CI: 0.0-0.851) p = 0.001; and %AVA — 0.94 (CI: 0.326 - 0.989) p <0.001. In linear regression analysis, the slope for prediction of the GS %AVA from A3ID was 0.98, R2 = 0.975. Bland-Altman analysis revealed a trend for computer underestimation of the AVA in images with low AVA and overestimation of AVA in images with large AVAs.
Conclusions :
A3ID reliably predicts %AVA from rat OIR retinal images, but hand-segmentation of images is superior, especially in retinas with extremes of %AVA.
This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.