Purchase this article with an account.
Teresa Torzicky, Stefan Zotter, Michael Pircher, Bernhard Baumann, Christoph Hitzenberger; Investigation of the Choroid-Sclera Interface and Choroidal Thickness in 3D Polarization Sensitive OCT Images of the Human Eye. Invest. Ophthalmol. Vis. Sci. 2013;54(15):1474.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To extend existing algorithms for 2D images for automatically segmenting the choroid-sclera interface (CSI) and choridal thickness (ChT) based on polarization sensitive optical coherence tomography (PS-OCT) data of healthy human volunteers, in order to make them applicable for 3D date sets.
In this work we are using 3D data sets, generated with a PS-OCT prototype with a central wavelength of 1040 nm and an A-scan rate of 100 kHz to segment the CSI and determine the ChT, based on polarization characteristics. Since ChT varies highly within small areas, comparing thickness measurements in B-scans can lead to inaccurate results, if the compared positions vary. Generating 3D thickness maps is one possibility to overcome this issue. The RPE can be segmented by its depolarizing effect as in the 2D case. For the CSI, however, applying the same algorithms as used in averaged 2D images (searching for retardation gradients in the region posterior to the RPE, in order to detect the birefringence of the sclera) is not sufficient for 3D data sets, since the sensitivity of a single frame is sometimes too low for detecting the retardation increase caused by the birefringent sclera. Therefore different approaches to alter the algorithm, like fitting a polynomial of the fourth order through the segmentation points found for the CSI, averaging in 3D windows that where shifted over the data set before searching for the retardation gradient, etc., were investigated. In a second step the results for ChT created with the different algorithms were compared with each other and with results acquired in 2D images.
The approach with the polynomial fitting was tested in different human volunteers and provided reproducible results, which were comparable in most areas with the results generated from 2D averaged B-scans. In certain areas where the polarization characteristics are changing strongly and not homogenously, the algorithm neglects certain structures and provides a smooth segmentation line, which is not really following the observed polarization characteristics.
First tests in healthy human volunteers showed, that automated segmentation of the CSI using PS-OCT data is not only possible for averaged B-Scans, but also for 3D data sets. The results show similar trends as the results from automated polarization based segmentation in 2D scans.
This PDF is available to Subscribers Only