Abstract
Purpose: :
To develop an automated method of estimating keratocyte density in images from the ConfoScan 4.
Methods: :
We modified a program, that was developed to assess cell density in images from the Tandem Scanning confocal microscope (Tandem Scanning, Inc., Reston, VA; McLaren, et al. J. Microscopy, in press), to determine cell density in images from the ConfoScan 4 (Nidek, Inc., Fremont, CA). Fifteen eyes of 15 normal volunteers were examined by using the ConfoScan 4 equipped with a Z-ring adapter. Two non-blurred frames were selected from each of 5 layers of stroma, and cell density was assessed manually and automatically in a central rectangle (239 µm x 212 µm) to avoid the peripheral decrease in image brightness. Selection variables in the program were adjusted to optimize automatic identification of nuclei. The program was tested (using the pre-determined selection criteria) in 15 corneas from a second group of 15 volunteers and was compared to density in the same frames determined by manual assessment. Cell densities were also compared to densities determined from scans of the same corneas by the Tandem Scanning confocal microscope. The depth of field was 25.9 µm and 11.9 µm for the ConfoScan 4 and Tandem Scanning microscopes respectively. Densities were compared within layers by using paired t-tests and in combined layers by using Generalized Estimating equation models. Statistics were Bonferroni-adjusted for 2 comparisons.
Results: :
In the 15 test corneas, mean cell densities determined by automatic assessment of images from the ConfoScan 4 were 28,725 ± 3,501 cells/mm3 and 23,724 ± 3,781 cells/mm3 in the anterior 10% and 10-100% of the stroma respectively. Cell densities in each layer were not significantly different from those determined by manual assessment (p > 0.8). The mean difference between automated and manual assessments was -395 ± 3,385 cells/mm3 (auto - manual, p=0.8) and the minimum detectable difference was 1,250 cells/mm3 (α=0.025, β=0.20, n=75 frames). Mean cell densities in the anterior 10% of the stroma were significantly less than those determined from the same region in scans from the Tandem Scanning microscope (43,747 ± 6,824cells/mm3, p<0.001), but were not significantly different in deeper layers (p>0.4).
Conclusions: :
This program provides an efficient and objective method of determining cell density in the corneal stroma in images from the ConfoScan 4. Cell densities determined from the ConfoScan 4 in the anterior-most region of the stroma are considerably lower than those from the Tandem Scanning microscope, possibly because of the difference in the depth of field between these instruments.
Keywords: cornea: stroma and keratocytes • microscopy: confocal/tunneling • imaging/image analysis: non-clinical