Abstract
Purpose:
Reliable automatic cone detection and cone mosaic assessment using adaptive optics (AO) is of the key importance in tracking of macular pathologies, especially ones related to the inherited retinal degenerations (e.g. Usher syndrome). The automatic cone detection (especially fovea centered) in AO is an unsolved issue mainly due to the dense cones concentration both in healthy retina and in retinal pathologies. An automatic cone quantification tool was developed with the aim to provide quick and reliable cone counting in fovea centered AO images. The tool was first tested in healthy subjects. In addition cone mosaic profile was compared between healthy subjects and patients with Usher syndrome.
Methods:
A flood-illumination AO retinal camera (rtx1, Imagine Eyes, France) was used to acquire fovea centered images of the cone mosaic. The cone mosaic in 10 images of 8°x8° (1500x1500 pixel) corresponding to the macular area of 1181x1181 µm were analyzed in healthy subjects. The automatic tool in LabVIEW visual programming language (National Instruments, USA) with sequenced image filtration was employed to obtain clearly discerned cell mosaic. The reliability of the automatic tool was compared to an expert based manual counting using a set of 30 images with 120x120 pixel (95x95µm), which were randomly selected from the initial large images. The repeatability of manual and automatic counting was tested in 3 randomly selected images. In addition, healthy subjects’ cone mosaic profile was compared with ones having genetically confirmed Usher syndrome.
Results:
The mean amount of cones per 1 mm2 was 23120(21760 - 23530) [median(1st quartile - 3rd quartile)]. There was no significant difference in cones count comparing automatic counting: 248(213 - 282) cells/picture, and manual counting: 248(214-273) cells/picture (p = 0.8, Wilcoxon test). The repeatability amounted to 100% (automatic counting) and to 98% (manual counting). Comparing healthy subjects’ cone mosaic profile with ones having Usher syndrome significant decrease of 30% in cones quantity per area was shown.
Conclusions:
Our developed automatic cone detection tool seems to be reliable for automatic cone quantification and mosaic assessment in the normal fovea using AO images. In addition, the automatic tool could be potentially used for tracking macular pathologies, especially ones related to inherited retinal degenerations.