Comparative quantification of geographic atrophy of the retinal pigment epithelium across color fundus photographs, red-free fundus photographs, and autofluorescence imaging by grader versus automated grading software using optical coherence tomography

Abstract

 

Purpose

 

To assess the correlation between total geographic atrophy (GA) area assessed by an experienced grader from color fundus photos, red-free fundus photos, or fundus autofluorescence (FAF) images with automated measurements derived from instrument grading software applied to optical coherence tomography (OCT) images.

 

Methods

 

Participants with at least 1 eye affected with GA secondary to age-related macular degeneration or retinal pigment epithelial abnormalities consistent with a pattern dystrophy gave informed consent for this IRB-approved study following which they had 30 degree digital color and red-free fundus photos, FAF (Spectralis) and spectral domain RTVue OCT imaging on the same day. Thirty-eight eyes from 27 study participants were included. One grader independently evaluated the total GA area for color and red-free fundus photos using digital imaging software and evaluated areas of absent autofluorescence using the Spectralis RegionFinder™ tool for the FAF images. Total GA area was extracted from RTVue OCT images using an automated grading software. All grading methods evaluated the total GA area within a 6 mm diameter grid centered on the fovea. Main outcome measures included correlation coefficients and mean absolute and relative difference of GA area measurement across the 4 imaging modalities.

 

Results

 

The Pearson’s correlation coefficient for total GA area based on color, red-free fundus photos, and FAF images relative to OCT images in the right eye and in the left eye are in Table 1. The mean absolute difference (95% confidence interval) of the total GA area (averaging 6.03 mm² or 2.27 disc areas on OCT) between OCT and color, red-free, and FAF images were -2.33 mm² (-3.61 to -1.05), -2.34 mm² (-3.88 to -0.79), and -0.58 mm² (-1.73 to 0.56), respectively.

 

Conclusions

 

The high correlation coefficients between each of the 3 manually-graded and the single automated modality suggests that the OCT-based automated modality for measuring GA area might be feasible for assessing GA area and its evolution. Further studies are needed to determine why the area of GA determined manually on either red-free or color fundus photos appears to be greater, on average, than the area determined by automated methods on OCT.