Purchase this article with an account.
Li Zhang, Kyungmoo Lee, Xinjian Chen, Meindert Niemeijer, Michael D. Abramoff, Milan Sonka; Semi-automated Segmentation of Choroidal Vasculature in SD-OCT images. Invest. Ophthalmol. Vis. Sci. 2012;53(14):2117.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Imaging of the choroidal vasculature is clinically relevant in choroidits and retinal degeneration, but the classic approach to choroidal imaging, Indocyanine Green (ICG) angiography, contains an overlay of retinal vessels and is hard to interpret with large intra- and interobserver variability. We report a novel interactive graph-theoretic method to segment choroidal vessels from 3D spectral-domain optical coherence tomography (SD-OCT) scans.
5 subjects with normal choroid were imaged with SD-OCT (Zeiss Cirrus, 200×200×1024 voxels, voxel size of 30x30x2µm). The method constructed 3 sub-graphs for applying a maximum flow algorithm to determine in 3D possibly discontiguous region of vasculature and the two closest surfaces identifying the choroid (the lower surface of the retinal pigment epithelial (RPE) complex and a reference surface under the choroid). Using regional intensity information inside interactively identified seeds, regional costs were used to build the first sub-graph. Additional two gradient-based cost function images were used to build the remaining sub-graphs. All three sub-graphs were connected by intra-column edges, which introduced a distance constraint between the region of vasculature and the two surfaces. The final solution was obtained by computing the minimum-cut in the overall graph. The performance of this method was compared with that of a conventional graph-cut segmentation without the simultaneous dual-surface detection.
Our approach successfully identified the choridal vasculature in all 5 analyzed datasets, proving the feasibility of fully-3D detection of choroidal vasculature in SD-OCT. Figure below shows four examples of the segmentation results as 3D renderings.
Overall, our new approach successfully imaged the choroidal vasculature in 3D and produced markedly more accurate results with less over-segmentation compared to the conventional graph-cut approach. Though expert-assisted, this technique has the potential to advance choroidal imaging in retinal disease.
This PDF is available to Subscribers Only