Boundary Segmentation And Research Of SDOCT Retina Image Layer Based On 3D Graph Search

The Spectral Domain Optical Coherence Tomography (SDOCT) image plays an import role in modern medicine. While considering the different features of the macular and optic disc regions, this paper makes improvements for the 3D graph search algorithm to segment the retinal layers of SDOCT images. The works of this paper are as follows:(1) Segmentation of the Internal Limiting Membrane (ILM) layers in macular SDOCT images by gradient weaken algorithm and 3D graph search. Considering the traditional 3D graph search method cannot segment the ILM layer independently, this paper proposes the gradient weaken algorithm which can only retain the regions of strong gradients around the ILM layer while weaken the gradients of other regions. The improved method can avoid the error segmentation of the traditional methods.(2) Segmentation of multi-layers of macular SDOCT images by using multi-scale 3D graph methods. The chapter segment the Onner Nuclear Layer-Inner photoreceptor Segment (ONL-IS) layer, the Outer photoreceptor Segment-Retinal pigment epithelium(OS-RPE) layer and the Bruch’s Membrane (BM) layer. The traditional 3D graph search method has great time complexity to segment the three layers simultaneously. The chapter use the axial multi-scale to reduce the computing time. Also different cost function designs have been made for different layers. And the conception of combine layers is introduced and different scale takes different segment tactics.(3) Segmentation of multi-layers of Optic Nerve Head SDOCT images with application. The chapter improves the algorithm by using Thin Plate Spline(TPS). The projection image is produced by the data of segmented layers. The method of location of optic disc is proposed by using 3D graph search algorithm. The improved algorithm can segment the closed surface.