Abstract
Purpose: :
Adaptive optics (AO) imaging enables visualization of individual photoreceptors in the living human retina. As the clinical relevance of these images grows, the issue of how to analyze them becomes more important. Metrics like cell density and cell spacing are frequently used, though there remains inconsistency in how they are measured and there is virtually no data on their repeatability. Here we sought to assess the repeatability of measurements of cone density in images of the cone mosaic obtained with an AO scanning light ophthalmoscope (AOSLO).
Methods: :
Twenty-one participants with no ocular pathology were recruited. Four retinal locations, at approximately 0.5° eccentricity from the center of fixation were imaged using an undilated pupil 10 times in randomized order with an AOSLO (775 nm, 1 degree field-of-view). A chin/forehead rest was used to stabilize the head. Cones were identified using semi-automated software in each image, from which cone density was measured. Owing to fixational instability, the 10 images recorded from a given location did not overlap entirely. We thus analyzed images both in the presence of this variance and after additional processing to ensure precise co-alignment.
Results: :
Initial estimates of cone density were generally unreliable, with a coefficient of repeatability of 12,094 cones/mm2 (~16%, or 17 cones in our sampling field). The primary reason for this variability appears to be fixational instability, as when we aligned the 10 images before computing cone density, we found an improved reliability of 4,537 cones/mm2 (~6%, or 7 cones in our sampling field). Repeatability improved further by manually identifying cones missed by the automated algorithm, with a coefficient of repeatability of 2,050 cones/mm2 (~2.5%, or 3 cones in our sampling field). Mean cone density in our subject group was 77,725 cones/mm2, consistent with previous reports for this retinal eccentricity.
Conclusions: :
As our data were collected in a normal population, this likely represents a best-case estimate for corresponding measurements in patients with retinal disease. Nevertheless, we demonstrate good repeatability of an automated cone density algorithm. Similar studies need to be carried out on individual imaging systems, as repeatability is highly sensitive to initial image quality and the performance of cone identification algorithms, both of which are highly variable. Additional work is also needed to develop ways of automating calculation of rod density.
Keywords: photoreceptors • imaging/image analysis: non-clinical • imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound)