Purchase this article with an account.
Milan Sonka, Xinjian Chen, Meindert Niemeijer, Li Zhang, Kyungmoo Lee, Mona K. Garvin, Andreas Wahle, Stephen R. Russell, James C. Folk, Michael D. Abramoff; Automated Segmentation of Fluid Regions In Choroidal Neovascularization In SD-OCT. Invest. Ophthalmol. Vis. Sci. 2012;53(14):4085.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Intraretinal and subretinal fluids are one of the primary parameters guiding anti-VEGF injection treatment of patients with choroidal neovascularization (CNV). We report performance evaluation of our new automated method for segmenting fluid and associated abnormalities in the retina, so-called Symptomatic Exudate Associated Derangements (SEADs).
Treated eyes of 10 subjects undergoing anti-VEGF injections were imaged using SD-OCT (Zeiss, 200×1024×200 voxels, 30.0×2.0×30.0µm/voxel). A retinal specialist manually segmented the intra- and sub-retinal fluid in each slice of each eye using Truthmarker software on iPad (Fig. b green).Our SEAD segmentation method started with segmenting the retinal layers and detecting SEAD footprints using our previously reported approach . A surface was fitted to the RPE layer while ignoring locations within the SEAD footprint, this fitted surface was used to flatten the original OCT images. Once flattened; a voxel classification based method was applied to get the approximate SEAD regions in 3D (Fig. a). Employed features features included Gaussian derivatives at various scales describing texture, structure features of Hessian eigenvalues at various scales, and locational features derived from the computed distance to certain layers of the layer segmentation. In the final segmentation step (Fig. b red)., a probability-constrained graph cut method was employed that used the approximate SEAD segmentation as the source and background seeds; and integrated the other features into the graph cut cost function as probability constraints. A leave-one-out strategy was used to evaluate the performance of the automated SEAD segmentation approach. The true positive volume fraction (TPVF) and false positive volume fraction (FPVF) were used as performance indices.
The average TPVF and FPVF between the automatically segmented SEAD region and the expert-traced independent standard were 85.1% and 2.5%, respectively.
Automated fluid segmentation is reported that has a comparable performance to a human expert. Our approach has the potential to improve the management of patients with CNV. Quellec, G. et al.: IEEE Trans Med Im, 29 (6), 2010, 1321 - 1330.
This PDF is available to Subscribers Only