Reproducibility in peak foveal cone density characterization in AOSLO images.

Abstract

Purpose : To examine the inter-grader reproducibility of characterizing peak foveal cone density in normal subjects.

Methods : 300x300 μm foveal AOSLO images from the right eye of 44 normal volunteers were included. Semi-automated cone identification was performed by three observers, with resulting cone coordinates used to generate cone density maps. Two approaches were employed to estimate the location of peak cone density from these maps: 1) the point of absolute maximum density and 2) the center of an ellipse created by the 95th percentile isodensity contour. Cone density values at these locations were also compared. For each method, the repeatability and measurement error of peak cone density estimates were assessed. The variability in peak cone density location between observers was assessed by calculating confidence ellipses (CE) for each peak density location estimation method described above, and intraclass correlation coefficients (ICCs) were calculated to characterize interobserver variability in density estimates.

Results : The interobserver ICC of the maximum cone density value was 0.805 (95% CI = 0.718-0.893). The ICC using the isodensity contour method was slightly lower (0.748, 95% CI = 0.640-0.856). Accordingly, the measurement error for the maximum cone density method was 21,441 cones/mm² (12.3%) whereas the measurement error for the isodensity contour method was 24,549 cones/mm² (14.4%). Despite this, there was no significant difference between the average peak cone density value extracted using the two methods (p=0.09). The location of maximum density was significantly more reproducible between observers when using the isodensity contour method (mean±SD CE area = 231.93 ± 775.15 μm²) compared to the maximum density method (mean±SD CE area = 905.62 ± 1214.42 μm², p<0.0001).

Conclusions : Peak foveal cone density is important, not only as a quantitative metric of cell packing, but also as an anchor for calculation of retinal eccentricities. Understanding measurement errors, both in peak cone density estimation and in its retinal location is critical, to inform the approach to its identification. The superior reproducibility of the retinal location with the isodensity contour method may best serve comparison of parafoveal metrics, while the maximum density technique could be used to give the best estimate of the highest cone density within the retina, despite variability in the resulting exact retinal location.

This is a 2021 ARVO Annual Meeting abstract.