Purchase this article with an account.
Zhihong Hu, Ziyuan Chris Wang, Srinivas R Sadda; Automated 3D Choroidal Segmentation Using Multimodal Complementary Information. Invest. Ophthalmol. Vis. Sci. 2017;58(8):19.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Alterations in the choroid have been suggested to be of relevance to the appearance and progression of geographic atrophy (GA) in age-related macular degeneration (AMD). Although choroidal segmentation algorithms on spectral-domain optical coherence tomography (SD-OCT) have been developed, they may be confounded by the zones of increased choroidal reflectivity due to presence of GA. The purpose of this study is to evaluate a novel automated 3D choroidal segmentation approach for OCT which utilizes information from companion fundus autofluorescence (FAF) images.
The choroid was defined as the layer between outer border of retinal pigment epithelium (RPE)/Bruch’s membrane (BM) and choroid-sclera (C-S) junction. A 3D graph search algorithm was applied to OCT images to simultaneously identify these inner and outer choroidal surfaces. To utilize complementary information from FAF images, GA lesions on FAF were segmented using a supervised pixel classification approach. The FAF images and segmented GA masks were aligned with OCT projection maps using a feature-based image registration algorithm. By incorporating FAF-derived GA information into cost function and surface interaction constraints of 3D graph search, the choroidal boundaries on OCT were refined in these regions. Fig.1 illustrates the major steps of this 3D choroidal segmentation process. To evaluate system performance, 43 cases of GA with OCT and FAF data were randomly chosen from the reading center AMD database. Masked human graders independently segmented the choroidal borders. The automatically segmented choroidal borders with/without FAF-derived GA information were compared with manual segmentation, and mean difference and mean absolute difference for each boundary/surface were computed.
An example of the automated 3D choroidal segmentation performance with/without FAF-derived GA information is shown in Fig.1. Quantitative border differences are provided in Table 1. The mean border position for both the inner and outer choroidal surfaces were significantly improved (all p<0.01) by utilizing the FAF-derived information, with sub-micron mean differences for the RPE/BM border.
Utilization of multimodal information, in particular the location of GA on FAF images, can enhance the accuracy of automated choroidal segmentation on OCT images in eyes with atrophic AMD.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.
Fig.1. Overview of major steps of the 3D choroidal segmentation
This PDF is available to Subscribers Only