ARVO Meeting Abstracts
 QUICK SEARCH:   [advanced]


This Article
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Cooper, R. F.
Right arrow Articles by Carroll, J.
Right arrow Search for Related Content
Right arrow Articles by Cooper, R. F.
Right arrow Articles by Carroll, J.
Invest Ophthalmol Vis Sci 2011;52: E-Abstract 1321.
© 2011 ARVO


Examining FFT and Direct Counting Estimates of Photoreceptor Density in Adaptive Optics Retinal Images

Robert F. Cooper1, Jungtae Rha2A, Alfredo Dubra3 and Joseph Carroll2A,2B

1Biomedical Engineering, Marquette University, Milwaukee, Wisconsin
AOphthalmology, BCell Biology, Neurobiology, & Anatomy, 2Medical College of Wisconsin, Milwaukee, Wisconsin
3Flaum Eye Institute, University of Rochester, Rochester, New York

Commercial Relationships: Robert F. Cooper, None; Jungtae Rha, None; Alfredo Dubra, None; Joseph Carroll, None

Support: Research to Prevent Blindness, NIH grants P30EY001931 & R01EY017607. AD holds a Career Award at the Scientific Interface from the Burroughs Wellcome Fund.


Purpose:Adaptive optics (AO) imaging systems permit direct visualization of the photoreceptor mosaic. One of the more common metrics derived from these images is density, however, in cases where the mosaic is incomplete, either due to poor image quality, pathology, or interruption by another cell type, FFT-estimates of density can be inaccurate. Here we sought to analyze the efficacy of FFT techniques in deriving accurate estimates of density in AO images of varying density and disruption.

Methods:Eleven subjects with normal vision and five subjects with retinal pathology were imaged using an AO ophthalmoscope and/or an AO scanning laser ophthalmoscope. Photoreceptor density was calculated in all 16 images using manual and semi-automated direct counting methods as well as from the radial average of the FFT. Synthetic mosaics were constructed using photoreceptor coordinates from 10 of the normal cone mosaic images. This allowed us to examine the effect of system noise and background artifacts on the FFT, and also to examine the effect of mosaic disruption by randomly removing varying amounts of cones from the synthetic mosaic (10-70%).

Results:With the original cone mosaic images, we found that FFT-derived estimates of density matched those from the synthetic images within 5%, validating the use of the synthetic images for further modeling. Compared to both of these FFT-derived estimates, the direct count density was 12% greater on average. Furthermore, FFT-derived estimates of density degraded in some synthetic images when as few as 30% of the cones were randomly removed. This discrepancy was also seen in the pathological mosaics. In images containing both rods and cones, 2 peaks are visible in the radial average of the FFT, however the FFT-derived estimate of rod density overestimates the direct count density by over 50%. Correcting for the area of the image occupied by interleaved cones results in agreement within 5% between the FFT-derived and direct-count density estimates.

Conclusions:The source of the discrepancy between FFT-derived and direct-count cone density estimates remains to be determined, though likely results from assumptions made in converting peak spatial frequency to a density value. As AO systems can now image rods and cones simultaneously, it is important to develop methods to reliably retrieve density estimates of both mosaics, without resorting to counting every cell, especially as 100% of the cells may not be distinguishable.

Keywords: image processing • photoreceptors • imaging/image analysis: non-clinical

© 2011, The Association for Research in Vision and Ophthalmology, Inc., all rights reserved. Permission to republish any abstract or part of an abstract in any form must be obtained in writing from the ARVO Office prior to publication.