Our data show higher average intergrader agreement on nonconfocal AOSLO than HMM images, although confidence intervals overlap. This may be explained by the principles underlying split-detection and HMM imaging modalities. HMM imaging has a similar appearance to confocal AOSLO, which depends on the waveguiding of the photoreceptor. In the parafovea, larger cones have less uniform intensity profiles caused by passing of higher waveguide modes,
54 which can make them more challenging to distinguish from the surrounding rods.
55 In contrast, split-detection images are not thought to rely on waveguiding, as split-detection imaging in patients with non-waveguiding cones reveal inner segment structures.
8,56 This accounts for known superior intergrader agreement in split-detection than confocal AOSLO images outside the macula.
8,52 It is worth considering that our intergrader ICCs are lower than those reported by Cunefare et al.,
57 which may be explained by the fact that 85% of images included in the current study were captured at 1.5° field of view, compared to images exclusively captured at 1° by Cunefare et al. Our intergrader ICC for cone density on HMM was much higher than that previously reported by Mendonça et al.
38 Their low ICC of 0.22 is likely to be secondary to differing internal rules for cone identification. Naïve graders of retinal images with single-cell resolution have been shown to have measurably lower repeatability
55; thus, the prior experience of both of our graders with analyzing confocal AOSLO images is likely to have contributed to our higher ICC of 0.739 (95% CI, 0.611–0.868). Although ICC values comparable to ours with overlapping 95% confidence intervals have been reported in studies with graders inexperienced with AOSLO imaging (0.891 (95% CI, 0.696–0.952), these values were based on raw cone counts rather than bound cone density, as examined in our study.
39 Interestingly, the intergrader ICC of our raw cone counts was higher at 0.921 (95% CI, 0.861–0.983), indicating interobserver variability in identification of cells at the edge of an ROI.