Automatic detection of modal spacing (Yellott's ring) in adaptive optics scanning light ophthalmoscope images

Authors

  • Robert F Cooper,

    1. Department of Biomedical Engineering, Marquette University, Milwaukee, USA
    Search for more papers by this author
  • Christopher S Langlo,

    1. Department of Cell Biology, Neurobiology and Anatomy, Medical College of Wisconsin, Milwaukee, USA
    Search for more papers by this author
  • Alfredo Dubra,

    1. Department of Biomedical Engineering, Marquette University, Milwaukee, USA
    2. Department of Ophthalmology, Medical College of Wisconsin, Milwaukee, USA
    3. Department of Biophysics, Medical College of Wisconsin, Milwaukee, USA
    Search for more papers by this author
  • Joseph Carroll

    Corresponding author
    1. Department of Cell Biology, Neurobiology and Anatomy, Medical College of Wisconsin, Milwaukee, USA
    2. Department of Ophthalmology, Medical College of Wisconsin, Milwaukee, USA
    3. Department of Biophysics, Medical College of Wisconsin, Milwaukee, USA
    • Department of Biomedical Engineering, Marquette University, Milwaukee, USA
    Search for more papers by this author

Correspondence: Joseph Carroll

E-mail address: jcarroll@mcw.edu

Abstract

Purpose

An impediment for the clinical utilisation of ophthalmic adaptive optics imaging systems is the automated assessment of photoreceptor mosaic integrity. Here we propose a fully automated algorithm for estimating photoreceptor density based on the radius of Yellott's ring.

Methods

The discrete Fourier transform (DFT) was used to obtain the power spectrum for a series of images of the human photoreceptor mosaic. Cell spacing is estimated by least-square fitting an annular pattern with a Gaussian cross section to the power spectrum; the radius of the resulting annulus provides an estimate of the modal spacing of the photoreceptors in the retinal image. The intrasession repeatability of the cone density estimates from the algorithm was evaluated, and the accuracy of the algorithm was validated against direct count estimates from a previous study. Accuracy in the presence of multiple cell types and disruptions in the mosaic was examined using images from four patients with retinal pathology and perifoveal images from two subjects with normal vision.

Results

Intrasession repeatability of the power spectrum method was comparable to a fully automated direct counting algorithm, but worse than that for the manually adjusted direct count values. In images of the normal parafoveal cone mosaic, we find good agreement between the power-spectrum derived density and that from the direct counting algorithm. In diseased eyes, the power spectrum method is insensitive to photoreceptor loss, with cone density estimates overestimating the density determined with direct counting. The automated power spectrum method also produced unreliable estimates of rod and cone density in perifoveal images of the photoreceptor mosaic, though manual correction of the initial algorithm output results in density estimates in better agreement with direct count values.

Conclusions

We developed and validated an automated algorithm based on the power spectrum for extracting estimates of cone spacing, from which estimates of density can be derived. This approach may be used to estimate cone density in images where not every single cone is visible, though caution is needed, as this robustness becomes a weakness when dealing with images from patients with some retinal diseases. This study represents an important first step in carefully assessing the relative utility of metrics for analysing the photoreceptor mosaic, and similar analyses of other metrics/algorithms are needed.

Ancillary