314 - Development of a Semi-Automated ImageJ Plugin to Quantify Oxygen-Induced Retinopathy

Publication Number: 4308.314

Husaam Qureishy, University of Minnesota Medical School, Eagan, MN, United States; Amelia Morikuni, University of Minnesota Medical School, Itasca, IL, United States; Rehan Krishnan, University of Minnesota Medical School, Collierville, TN, United States; Ellen Ingolfsland, University of Minnesota, Minneapolis, MN, United States

Background: The incidence of retinopathy of prematurity (ROP) is increasing as survival rates improve for extremely preterm infants, underscoring the need for more research to improve prevention and treatment strategies. The rat oxygen induced retinopathy (OIR) model is a well-validated method for preclinical study of ROP. Its primary outcome is the quantified avascular and neovascular areas of retinal flat mounts. The gold standard method to measure these outcomes is by hand, a time-consuming process subject to variability between reviewers and laboratory groups.

Objective: To develop and validate a semi-automated ImageJ macro plugin to measure retinal avascular and neovascular area in the rat OIR model

Design/Methods: Previously generated Isolectin B4-stained rat retinal flat mounts from the rat 50/10 OIR model at postnatal day (P)15 (the avascular stage) and P20 (the neovascular stage) were used to develop three plugins: total retinal area, avascular area, and neovascular area in ImageJ. These plugins used filters, segmentation tools and limited user input to isolate the retina, remove non-retinal areas, and identify the avascular area. Due to high variability in neovascularization morphology, the neovascular area plugin failed and was not pursued. For validation of the total retinal area and avascular area plugins, 30 retinal images, including images from both time points and across multiple experiments and years of research were measured with the plugins. Vascular outcomes were compared to the gold standard, the average measurements from two manual reviewers, and intraclass correlation coefficients, coefficients of determination, linear regression, and Bland-Altman measures were calculated.

Results: Manual reviewer measurements demonstrated high correlation, as expected (Table 1). Measurements of total retinal area and avascular area from the plugins showed similarly strong correlation with the gold standard (Table 2). Bland-Altman analysis showed close agreement, with greater scatter only when avascular areas approached zero (Fig. 1).

Conclusion(s): We successfully developed ImageJ plugins to semi-automate measurement of total retinal area and avascular area in rat OIR flat mounts. These tools correlate strongly with manual measurements and have the potential to reduce processing time and improve reproducibility across laboratories studying OIR.

Table 1. Statistical analysis of two hand measured quantifications (gold standard)
Quantifications of total retinal area (TRA) and avascular area made by two hand measurers across n=30 retinal flat mounts were compared with Pearson and intraclass correlation coefficients.

Table 2. Statistical comparison of plugin performance and gold standard
The average of the two hand reviewers across the thirty flat mounts served as the gold standard, compared against a single plugin user. Coefficients of determination, intraclass correlation coefficients with 95% confidence intervals, and linear regression values were calculated for comparison.

Figure 1. Log-transformed Bland-Altman plots comparing a) total retina area and b) avascular area quantification of the plugin versus the gold standard
The x axis plots the average measurements between the plugin and gold standard, while the y axis plots the difference between the two.