Purchase this article with an account.
Padraig Joseph Mulholland, Juliane Matlach, Marketa Cilkova, Tony Redmond, David F Garway-Heath, Roger S Anderson; Adaptive Optics Free Photoreceptor Imaging – Comparison of Manual and Automated Cone Counts. Invest. Ophthalmol. Vis. Sci. 2016;57(12):68. doi: https://doi.org/.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To compare cone density reported by automated analysis software and a manual analysis technique when examining in-vivo images of the human photoreceptor mosaic acquired with a modified Heidelberg Retina Angiograph-2 (HRA-2).
Forty in-vivo images of the photoreceptor mosaic (768x768 pixels, 0.65 mm2) were acquired at 8.8° retinal eccentricity in 33 healthy subjects of varying age with a modified HRA-2 (scan angle 3°). Images were cropped to 200x200 pixels (figures a-b). Cones were manually counted by an experienced observer (figure c). The same images were analysed with an automated counting technique (figure d), programmed in MATLAB (2014b, The Mathworks, USA). The technique was used to (i) estimate the number of cones within the image, (ii) estimate inter-photoreceptor spacing, and (iii) analyze the packing arrangement (hexagonality) of the photoreceptor mosaic. Inter-method agreement for the localization of identified cone centers was assessed for all images.
Mean cone number (and cone density, cells/mm2) ± SD across all images were 405.1 ± 51 cells (8775 ± 1109 cells/mm2) and 405.9 ± 33 cells (8795 ± 705 cells/mm2) with manual and automated techniques, respectively. The mean difference between the automated and manual counts for all images included in this study was 8.3 ± 39.5 cells. With the automated technique, a median of 88% of cells identified were within ~4 µm of a corresponding manually labeled cone. The mean inter-photoreceptor separation was 14.2 ± 0.6 µm. Voronoi analysis revealed 47% of all identified cells to be hexagonally arranged.
Manual and automated cone counts were very similar in the images examined in this study. Further research is required to examine the accuracy of the analysis algorithm when applied to images of the photoreceptor mosaic acquired in subjects with retinal disease.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.
Steps in image analysis: (a) 200x200 pixel region (red border) cropped from 3x3o in-vivo image of the photoreceptor mosaic, (b) example cropped image analyzed in study, (c) manually identified cones (green crosses), and (d) cones (red spots) identified using automated analysis algorithm for example image.
This PDF is available to Subscribers Only