Abstract
Purpose: :
Measures derived from optic disc photographs have been used in multi-center clinical trials and reading centers. The variability of the measures have limited the sensitivity of detecting glaucomatous structural changes. There is thus a need for methods yielding less variable results.
Methods: :
We have developed a prototype software algorithm to assist a reader in deriving quantitative measures of optic nervehead (ONH) morphology. Digital monochromatic images (in green) obtained from film, digital color photographs or ‘red-free’ photographs can be used. The reader generates a cropped image of the ONH by pointing at its center with a cursor. The reader points at the margin of the disc and the algorithm fits a polynomial curve which can be refined by adding more marks. The disc tracing is used for subsequent images of that eye and for the delineation of the cup. The latter is derived by an algorithm based on the assumption that the image density at each location on the rim is a function of the amount of neural tissue present between the lamina cribrosa and the ONH surface. For each radial segment of the disc, a density profile is generated and its derivative is used to locate the point of inflection, defined as the cup margin. The reader is presented with the computed marks and required to delete artifacts due to the presence of blood vessels. A polynomial cup contour is fitted to the remaining marks and can be altered by additional ones. The intra-session variability was estimated with 34 pairs of ONH stereo color photographs obtained during the Glaucoma Imaging Longitudinal Study (GILS). The inter-session variability was derived by presenting the reader with a random series of photographs obtained in 24 eyes at two sessions a year apart.
Results: :
The difference in vertical C/D ratio was -0.012 ± 0.051 and -0.006 ± 0.044 for the intra-sessions and inter-sessions, respectively. The absolute difference exceeded 0.2 and 0.1 in 0% and 4% of the cases, respectively. The values indicate that, theoretically, five instances of progression could be detected with 97.5% confidence in a C/D ratio range of 0.5 to 0.9.
Conclusions: :
This pilot study indicates that a low variability in C/D ratio measurements may be obtained by implementing a computer algorithm that presents a reader with computed data and allows deletion of artifacts
Keywords: imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound) • image processing • detection