Background

Adaptive optics scanning light ophthalmoscopy (AOSLO) enables direct visualisation of the cone mosaic, with metrics such as cone density and cell spacing used to assess the integrity or health of the mosaic. Here we examined the interobserver and inter-instrument reliability of cone density measurements.

Methods

For the interobserver reliability study, 30 subjects with no vision-limiting pathology were imaged. Three image sequences were acquired at a single parafoveal location and aligned to ensure that the three images were from the same retinal location. Ten observers used a semiautomated algorithm to identify the cones in each image, and this was repeated three times for each image. To assess inter-instrument reliability, 20 subjects were imaged at eight parafoveal locations on one AOSLO, followed by the same set of locations on the second AOSLO. A single observer manually aligned the pairs of images and used the semiautomated algorithm to identify the cones in each image.

Results

Based on a factorial study design model and a variance components model, the interobserver study's largest contribution to variability was the subject (95.72%) while the observer's contribution was only 1.03%. For the inter-instrument study, an average cone density intraclass correlation coefficient (ICC) of between 0.931 and 0.975 was calculated.

Conclusions

With the AOSLOs used here, reliable cone density measurements can be obtained between observers and between instruments. Additional work is needed to determine how these results vary with differences in image quality.

Background Adaptive optics scanning light ophthalmoscopy (AOSLO) enables direct visualisation of the cone mosaic, with metrics such as cone density and cell spacing used to assess the integrity or health of the mosaic. Here we examined the interobserver and inter-instrument reliability of cone density measurements. Methods For the interobserver reliability study, 30 subjects with no vision-limiting pathology were imaged. Three image sequences were acquired at a single parafoveal location and aligned to ensure that the three images were from the same retinal location. Ten observers used a semiautomated algorithm to identify the cones in each image, and this was repeated three times for each image. To assess inter-instrument reliability, 20 subjects were imaged at eight parafoveal locations on one AOSLO, followed by the same set of locations on the second AOSLO. A single observer manually aligned the pairs of images and used the semiautomated algorithm to identify the cones in each image. Results Based on a factorial study design model and a variance components model, the interobserver study's largest contribution to variability was the subject (95.72%) while the observer's contribution was only 1.03%. For the inter-instrument study, an average cone density intraclass correlation coefficient (ICC) of between 0.931 and 0.975 was calculated. Conclusions With the AOSLOs used here, reliable cone density measurements can be obtained between observers and between instruments. Additional work is needed to determine how these results vary with differences in image quality. © 2014 by the BMJ Publishing Group Ltd.