Abstract
Purpose :
Optical coherence tomography(OCT), being a noninvasive imaging modality, has begun to find vast use in the diagnosis and management of ocular diseases. However, due to speckle noise and irregularly shaped morphological features such as optic never head(ONH), accurate segmentation of individual retinal layers is difficult.
Methods :
11 normal subjects (6 used as training set, 5 used as testing set) were included and underwent Optic Nerve Head centered SD-OCT (Topcon 3D-OCT 2000, 512×128×885 voxels, 6mm×6mm×2.3mm). The proposed framework consists of two main steps: preprocessing and layer segmentation. In the preprocessing step, a curvature anisotropic diffusion filter was used to reduce the OCT speckle noise. Then the B-scans were flattened based on surface 5 (Fig.1). Subsequently, during layer segmentation, First, the Active Appearance Models(AAM)[3] was built, and the surface 1, 2, 3, 4, 5, 6, 7 and the optic disc was detected. Second, a multi-resolution graph-search algorithm[1], is applied for the further precise segmentation. Finally, the rim region of the optic disc was detected by shape prior model on the projection image from surface 6 to surface 7. The rim region is masked out because the layers are hard to define in this region.
Results :
The algorithm was tested against the ground truth which manually traced by two retinal specialists independently. The results are accordance with the ground truth well(Fig.2).
Conclusions :
In this paper, we proposed an automated method for retinal layers segmentation in ONH centered 3-D OCT images, which addresses the challenges posed by the presence of the large blood vessels and the optic disc. The preliminary results show the feasibility and efficiency of the proposed method.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.