Abstract
Abstract: :
Purpose: A scanning laser ophthalmoscope (SLO) using adaptive optics to correct the aberrations of the human eye has enabled in vivo imaging of the photoreceptor mosaic. This study investigates a computer-automated algorithm for computing photoreceptor packing in the living retina from digital video data acquired by the adaptive optics SLO. Future applications relate to quantification of photoreceptor loss in diseased retinas. Methods: Video images of the in vivo photoreceptors of four normal subjects were acquired with an adaptive optics scanning laser ophthalmoscope operating at a 660 nm wavelength. Twenty seconds videos at 30 frames per second with a 1.5x1.4 degree field of view were collected at 512x480 pixel resolution. The left eye of each subject was imaged at four locations: 1.5, 2.25, 3.0, and 3.75 degrees temporal to the fovea. At each location, ten consecutive frames were registered and summed to improve the signal-to-noise ratio. The two regions from each summed image that exhibited the best-defined photoreceptors were manually extracted. An iterative rule-based segmentation algorithm was applied to these segments to automatically detect photoreceptors and compute the resulting photoreceptor packing density. Results were compared to those from manual analysis of the summed regions. An exponential model for photoreceptor packing versus eccentricity from the fovea was computed for the four subjects. Inter-subject variation was examined. Results: The automated algorithm achieved an accuracy of 90% as compared to manual analysis of the extracted regions. Based on an exponential model, the mean photoreceptor packing for the four normal subjects at locations of 1.5, 2.25, 3.0, and 3.75 degrees temporal to the fovea was respectively 34900, 29200, 24500 and 20500 cones per square mm. The standard deviation of the measurements about the model was 3050 cones per square mm. Conclusions: Use of the automated algorithm resulted in a slight negative bias for photoreceptor packing estimates; but it reduced human effort, more accurately identified the centers of individual cones, and reduced user-dependent estimate variability.
Keywords: image processing • photoreceptors