Purpose
To determine the repeatability and agreement of the human cone densities measured by two original automatic detection algorithms. The cone mosaic images recorded by an adaptive optics laser ophthalmoscope (AO-SLO) were analyzed, and the cone density was calculated by the originally developed software.
Methods
Five eyes of five healthy subjects were studied (27-38-years, mean 30.2 years). The cone mosaics were photographed by TOPCON prototype AO-SLO. The cone density was measured at 0.5 mm, 1.0 mm, and 1.5 mm temporal to the fovea. In one trial, approximately 400 sequential images were obtained at 30 frames/sec of the same retinal area (1.00 x 1.25 deg.). Ten recordings were made for each retinal area. Ten continuous frames were selected automatically to obtain an averaged image. The coordinates of the cones in each averaged image was identified using two different automated algorithms. In algorithm 1, an edge detection algorithm was applied. The averaged cone image was morphologically magnified, and an adaptive threshold process was applied to create a binary image. The edge of the cone mosaic in the binary image was detected and the cone density was calculated. In algorithm 2, an image enhancement algorithm in the spatial domain was used. The power spectrum of the averaged cone image by fast Fourier transform (FFT) was made with an appropriate band pass filter. Inverse FFT was applied to the filtered image and the local maximum was identified. The repeatability was calculated from the within-subject standard deviation (SDw). The repeatability was defined as 2.77 SDw (Bland and Altman, 1996). The agreement between the two algorithms was evaluated by a Bland-Altman Plot (Bland and Altman, 1986).
Results
The repeatability of algorithms 1 and 2 was high, and the agreement of the two methods is shown in a Bland-Altman Plot. The mean difference (bias) was 195 cones/mm2, and the 95% limits of agreement was 2503, -2114 cones/mm2.
Conclusions
We found good repeatability and agreement of the two original automatic detection algorithms. The cone densities determined measured by our two algorithms were in good agreement with a histological study (Curcio et al. 1990) and an in vivo AO-SLO study (Park et al. 2013).
Keywords: 648 photoreceptors •
689 retina: distal (photoreceptors, horizontal cells, bipolar cells) •
549 image processing