Purchase this article with an account.
Y. Ojima, M. Hangai, A. Sakamoto, D. Iwama, A. Otani, A. Tsujikawa, N. Yoshimura; Errors in Three-Dimensional Imaging of Retinal Pigment Epithelium Surface by Using Spectral-Domain Optical Coherence Tomography. Invest. Ophthalmol. Vis. Sci. 2008;49(13):1837.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To determine errors in automatic segmentation of the anterior borders of the retinal pigment epithelium (RPE) in three-dimensional (3D) imaging of the RPE surface in macular diseases.
We imaged 22 eyes with macular disease [10 with central serous chorioretinopathy (CSC) and 12 with polypoidal choroidal vasculopathy (PCV)], using spectral-domain optical coherence tomography (OCT) (3D-OCT1000, Topcon). Raster scan protocol of 512 x 128 covering a 6 x 6 mm macular area was used. Automatic segmentation of the anterior borders of the RPE line in 128 B-scan images, followed by 3D topographic mapping of the resulting RPE plane, were performed using TMS software, version 2.10.(Topcon Medical Systems, Inc) Errors per B-scan image were graded and classified. The RPE surfaces depicted by the automatic and manual segmentation of the anterior borders of the RPE were compared in terms of sub-RPE lesions by color fundus photography and fluorescein or indocyanine green angiograms.
Sub-RPE lesions such as pigment epithelial detachment, polypoidal lesions, and branching vascular network were seen as varying levels of RPE surface elevation against the flat RPE surface. Errors in automatic segmentation of the RPE were classified into 3 groups. The first group included critical errors: lines were misplaced on intraretinal high-reflectivity in case the reflectivity of RPE lines was decreased due to the serous retinal detachment. or beneath highly reflective lesions such as hard exudates and fibrin. The second group included mild errors: small focal elevation of the RPE surface, representing bulges in CSC, was missed. The third group included minor errors: lines were misplaced on neighboring highly reflective lines such as photoreceptor inner and outer segment junction(IS /OS) and OS. First and second errors interfered with the depiction of subretinal lesions in 3D RPE surface imaging. We considered segmentation as successful when images did not include first and second errors. The average success rates were 86.4% in CSC and 79.9% in PCV.
Sub-RPE lesions can be effectively depicted by 3D RPE surface imaging by segmentation of its anterior borders. The segmentation algorithm should be improved to prevent errors that critically interfere with the depiction of sub-RPE lesions.
This PDF is available to Subscribers Only