Purchase this article with an account.
Kyungmoo Lee, Alexis K. Warren, Michael Abramoff, Andreas Wahle, S. Scott Whitmore, Todd E. Scheetz, Robert F Mullins, John H Fingert, Milan Sonka, Elliott H. Sohn; Repeatability and reproducibility of automated choroidal layer segmentations from SD-OCT and EDI-OCT scans. Invest. Ophthalmol. Vis. Sci. 2020;61(7):479.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To introduce a new automated 3D choroidal layer segmentation method for macular spectral-domain optical coherence tomography (SD-OCT) and enhanced depth imaging OCT (EDI-OCT) scans and to evaluate it in terms of repeatability and reproducibility.
To automatically segment choroidal layers, our LOGISMOS (layered optimal graph image segmentation for multiple objects and surfaces) method using a multiresolution coarse-to-fine approach was refined. An edge-based cost function was compared to a combined edge/vesselness-based cost function, as well as to our previous approach - enveloping the choroidal vessel segmentation. The repeatability between sequential SD-OCT scans and the reproducibility between SD-OCT and EDI-OCT scans of the choroidal layer thicknesses were estimated as intraclass correlation coefficient (ICC), coefficient of variation (CV), and repeatability coefficient (RC).
22 x 2 repeated macular SD-OCT scans (200 × 1024 × 200 voxels, 6.0 × 2.0 × 6.0 mm3) and 22 EDI-OCT scans (768 × 496 × 61 voxels, 9.1 × 1.9 × 7.7 mm3) from both eyes of 11 normal subjects were acquired on CirrusTM HD-OCT (Carl Zeiss Meditec, Inc., Dublin, CA) and Spectralis (Heidelberg Engineering, Germany). Choroidal layer segmentation results, thickness maps, Bland-Altman plots are shown in Fig. 1, and the mean layer thicknesses of the central 1 mm circular region, ICC, CV, RC values are shown in Table 1.
The LOGISMOS method using both edge-based and vesselness-based cost functions showed superior repeatability and reproducibility. Deep learning-based cost functions could be useful for more reliable choroidal layer segmentation.
This is a 2020 ARVO Annual Meeting abstract.
Figure 1. Fovea-centered B-scan images overlaid with the choroidal layers, thickness maps, and Bland-Altman plots obtained from our old method, new method using an edge-based cost function only, and new method using both edge-based and vesselness-based cost functions.
Table 1. Mean choroidal layer thicknesses of the central 1 mm circular region, ICC, CV, and RC values obtained from our old method, new method using an edge-based cost function only, and new method using both edge-based and vesselness-based cost functions.
This PDF is available to Subscribers Only