Abstract
Purpose:
To present the reproducibility and repeatability of cone distribution and density measurement, comparing the automatic count to the manual count. The imaging quality and individual differences of operators are also considered using by adaptive optics scanning laser ophthalmoscopes (AOSLO) images and preprocessing with software of ARIA system, Canon Inc.
Methods:
Total 22 eyes of 11 healthy subjects were included for repeatability test of AOSLO imaging system, prototype II Canon. The S1(340X340um) image was taken at each different meridian of macula of the same 1.0mm distance from fovea by the same operator in 3 different days. Another 18 eyes of 9 healthy volunteers were recruited for reproducibility test and P0 (680X680um) images were also obtained by the same method above, but performed by 3 different skilled operators following totally randomized order within 30 minutes. The 100X100um area was selected by choosing the entirely same cone mosaic location of 3 different images from different times (repeatability) and different operators (reproducibility). All the selected 100X100um images were measured automatically by ARIA and point-by-point by manual counting. Each selected image was also evaluated using the subjective score grading level from 0 to 10. Using blocking analysis of variance (ANOVA) and regression methods, image quality, counting method, and meridian of macula were evaluated on the measurements at statistical significant level of 0.05.
Results:
Results of repeatability analysis: within each meridian, there is no significant difference between automatic and manual counting (p = 0.294). We also observed no statistically significance for the interaction effect between meridian and image quality (p = 0.408). However, when the quality of image increased the average difference in auto counting and manual counting measurements shrunk (p < 0.01). Results of reproducibility analysis: we found no significant difference between two counting methods and each meridian of macula (p = 0.27). There is also no statistical significance between both methods and different operator (p = 0.94).
Conclusions:
The repeatability and reproducibility analyses of AOSLO showed that both automatic and manual counting methods were reliable, but as image quality decreased manual counting remained accountable. Bad image quality induced counting error.