Purchase this article with an account.
M. A. Mayer, R. P. Tornow, R. Bock, J. Hornegger, F. E. Kruse; Automatic Nerve Fiber Layer Segmentation and Geometry Correction on Spectral Domain OCT Images Using Fuzzy C-Means Clustering. Invest. Ophthalmol. Vis. Sci. 2008;49(13):1880. doi: https://doi.org/.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To propose a novel automated method for nerve fiber layer (NFL) thickness measurement on high resolution spectral domain OCT B-scan images.
Circular B-scans (diameter 3.4mm, 512 or 768 A-scans) around the optic disc were acquired using a Spectralis HRA+OCT (Heidelberg Engineering). After removing the bias the extrema of the first derivative along all A-Scans of the OCT image are detected. For each extremum a feature vector is calculated using intensity and gradient information from the neighborhood in the B-Scan. The feature vectors are clustered automatically using fuzzy C-means clustering, yielding in separated classes for different retina layer borders. The classes corresponding to the lower and upper NFL boundaries and retina pigment epitel (RPE) are identified. Next, an interpolation over outliers and unreliable regions is applied. In order to improve further image analysis and to enhance the comparability between subjects the OCT image is flattended by taking a smoothed version of the lower RPE boundary as a baseline. The reliability of this method is evaluated by comparing the results to a manual segmentation carried out by 7 persons.
The method was applied on data sets from 5 normal and 7 glaucoma eyes. Algorithm parameters were the same for each image and due to the automatic clustering no previous knowledge about the image is needed. 97% of the upper and 75% of the lower nerve fiber layer boundary points lie within the double standard deviation of the hand segmentation. Computation time is 30s on a 2Ghz Pentium IV for a 512x496 circular B-scan using a Matlab implementation.
We present a NFL segmentation method on circular OCT scans that is applicable to normal as well as pathological data, different patients and varying scanner settings without parameter adaption. An ophthalmologist is assisted in his diagnosis by this new objective and reproduceable measurement. In addition a visually more informative image with corrected geometry is generated during the same process.
This PDF is available to Subscribers Only