Automated identification of cone photoreceptors in adaptive optics retinal images

J Opt Soc Am A Opt Image Sci Vis. 2007 May;24(5):1358-63. doi: 10.1364/josaa.24.001358.


In making noninvasive measurements of the human cone mosaic, the task of labeling each individual cone is unavoidable. Manual labeling is a time-consuming process, setting the motivation for the development of an automated method. An automated algorithm for labeling cones in adaptive optics (AO) retinal images is implemented and tested on real data. The optical fiber properties of cones aided the design of the algorithm. Out of 2153 manually labeled cones from six different images, the automated method correctly identified 94.1% of them. The agreement between the automated and the manual labeling methods varied from 92.7% to 96.2% across the six images. Results between the two methods disagreed for 1.2% to 9.1% of the cones. Voronoi analysis of large montages of AO retinal images confirmed the general hexagonal-packing structure of retinal cones as well as the general cone density variability across portions of the retina. The consistency of our measurements demonstrates the reliability and practicality of having an automated solution to this problem.

Publication types

  • Evaluation Study
  • Research Support, N.I.H., Extramural

MeSH terms

  • Algorithms*
  • Artificial Intelligence*
  • Cluster Analysis
  • Humans
  • Image Enhancement / methods*
  • Image Interpretation, Computer-Assisted / methods*
  • Lenses
  • Pattern Recognition, Automated / methods*
  • Reproducibility of Results
  • Retinal Cone Photoreceptor Cells / anatomy & histology*
  • Retinoscopes
  • Retinoscopy / methods*
  • Sensitivity and Specificity