Purchase this article with an account.
A. Ruggeri, F. Scarpa, E. Grisan, J. W. McLaren, J. C. Erie, W. M. Bourne; Evaluation of Automatic Corneal Nerve Tracing in Normal and Refractive Surgery Subjects. Invest. Ophthalmol. Vis. Sci. 2007;48(13):2753.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To evaluate an algorithm for the automatic tracing of nerves in confocal microscopy images of the corneal subbasal plexus.
Corneal images were acquired with either of two clinical confocal microscopes (ConfoScan3 or ConfoScan4; Nidek Technologies Srl, Padova, Italy) in 20 normal subjects at two visits one year apart, and in 10 patients before and at 1, 3, 6, and 12 months after LASIK. Each cornea was scanned twice at each visit (180 total scans). One image of the subbasal nerve plexus was selected from each scan. Nerves fiber bundles were manually identified and their lengths were measured in all images by a cornea expert by using a computer-aided tool (NeuronJ, NIH and Eric Meijering). A new procedure, based on modern digital image processing and analysis, was developed for the fully automatic identification of nerves and measurement of their length (Ruggeri, et al., ARVO, 2006). The nerve lengths (um per image) determined by the automatic procedure were compared with the nerve lengths determined after manual identification in the same images.
Total nerve length per image ranged from 0 um/image to 2213 um/image, determined manually. Nerve lengths per image determined by the automatic analysis system were well correlated with those determined manually (r = 0.92, see figure).In images from the ConfoScan3, nerve density determined automatically was 165 ± 230 um/image (mean ± SD, n = 96) lower than that determined manually. In images from the ConfoScan4, nerve density determined automatically was 180 ± 241 um/image (n = 84) lower than that determined manually.
Nerve density was on the average underestimated by the algorithm in images from both instruments by approximately 30%. Automatic and manual densities from the same corneas were, however, well-correlated, suggesting that the automatic procedure is capable of reproducing the differences in nerve length estimated manually in different subjects. Refinement of this program is ongoing.
This PDF is available to Subscribers Only