Technical factors influencing cone packing density estimates in adaptive optics flood illuminated retinal images

Abstract

Purpose: To investigate the influence of various technical factors on the variation of cone packing density estimates in adaptive optics flood illuminated retinal images.

Methods: Adaptive optics images of the photoreceptor mosaic were obtained in fifteen healthy subjects. The cone density and Voronoi diagrams were assessed in sampling windows of 320×320 µm, 160×160 µm and 64×64 µm at 1.5 degree temporal and superior eccentricity from the preferred locus of fixation (PRL). The technical factors that have been analyzed included the sampling window size, the corrected retinal magnification factor (RMFcorr), the conversion from radial to linear distance from the PRL, the displacement between the PRL and foveal center and the manual checking of cone identification algorithm. Bland-Altman analysis was used to assess the agreement between cone density estimated within the different sampling window conditions.

Results: The cone density declined with decreasing sampling area and data between areas of different size showed low agreement. A high agreement was found between sampling areas of the same size when comparing density calculated with or without using individual RMFcorr. The agreement between cone density measured at radial and linear distances from the PRL and between data referred to the PRL or the foveal center was moderate. The percentage of Voronoi tiles with hexagonal packing arrangement was comparable between sampling areas of different size. The boundary effect, presence of any retinal vessels, and the manual selection of cones missed by the automated identification algorithm were identified as the factors influencing variation of cone packing arrangements in Voronoi diagrams.

Conclusions: The sampling window size is the main technical factor that influences variation of cone density. Clear identification of each cone in the image and the use of a large buffer zone are necessary to minimize factors influencing variation of Voronoi diagrams of the cone mosaic.

Figure 1. Correlation between the ocular biometry variables and the individual corrected retinal magnification factor (RMF_corr).

A 2-predictor model incorporating axial length and spherical equivalent refraction was developed in order to understand the correlation of biometry variables with the RMF_corr. The model explains 70% of the variance of RMF_corr across the population (r = 0.70; P = 0.02).

Figure 2. Images of the cone mosaic and corresponding Voronoi maps at 1.50 degree temporal eccentricity.

Photoreceptor mosaic images acquired at 1.50 degree temporal eccentricity and corresponding Voronoi maps estimated within the three sampling areas of different size in case W10_S14. Scale bar is 50 µm. In this case, the differences in the % of hexagonal arrangement between sampling windows of different size were ≤1.7%. The black dots highlight the same Voronoi tiles that change their relative arrangement across sampling areas of different size.

Figure 3. Images of the cone mosaic and corresponding Voronoi maps at 1.50 degree superior eccentricity.

Photoreceptor mosaic images acquired at 1.50 degree superior location and corresponding Voronoi diagrams obtained from cone coordinates estimated within the three sampling areas of different size in case W10_S11. Scale bar is 50 µm. In this case, the difference in the % of hexagonal arrangement between sampling windows of different size were ≤6.5%. The black dots highlight the same Voronoi tiles that change their relative arrangement across sampling areas of different size. Presence of dark areas in the image of the cone mosaic (white arrows), the boundary effect and the manual selection of cones missed by automated counting influence the accuracy of Voronoi diagrams.

Figure 4. Spectacle corrected Retinal Magnification Factor (RMF_corr) plotted as a function of the spherical equivalent spectacle correction.

The solid line represents linear regression to data from the present work (black squares). The interrupted black line represents linear fit to aggregate data from three of our previous studies (total 62 eyes, gray dots; Lombardo M. et al. Retina, 2013; Lombardo et al. OPO, 2013; Lombardo et al. JCRS, 2013); the interrupted and dotted grey lines represent linear fits to data from Li et al. (18 eyes; IOVS, 2010) and Coletta & Watson (18 eyes; Vis Res, 2006) respectively.

Figure 5. Effect of manual checking of cone identification on Voronoi diagrams.

AO retinal images from case W10_S14 (from fig. 2). We revealed the effect of manual re-selection of cones missed by automated identification on variation of the relative arrangements of Voronoi tiles in the same mosaic. Panel A: the central inset shows a 160×160 µm overlapped to a 320×320 µm area. The violet crosses represent the cones that have been identified at exactly the same position in both cases (93%); the red and blue crosses show the cones that have been identified only in 160×160 µm and 320×320 µm area respectively. Panel B: the central inset shows a 64×64 µm overlapped to a 160×160 µm area. 72% cones have been identified at exactly the same position.