Research On The Intra-retinal Layer Segmentation Methods And Measurement Of Optical Coherence Tomography Images

The optical coherence tomography(OCT)is a powerful,noninvasive and high resolution imaging modality,which can fast reconstruct the micron scale biological structure of the intra-retinal layers and remarkable pathological features in 2D or 3D images without contact,trauma.With the rapid increase of retinal patients around the world,the OCT images has been widely applied to diagnose and screen various ocular diseases such as glaucoma,age-related macular degeneration and central serous retinopathy by the ophthalmologists,through changing in the size of the retinal layer thickness or the size of the lesion.However,in the local region of the OCT image,there existences some disturbed factors such as a large number of intrinsic speckle noise,fuzzy and uneven gray,weak contrast,retinal vascular that cause the discontinuity of some layer,some vitreous artifacts and lesions that cause the irregular from the intra-layer or interlayer.Not only the automatic segmentation of the many layers can be affected in the macular OCT image,but also the layer thickness or volume of some lesions can not be measured effectively.In this paper,two key problems were focused,namely,automatic segmentation of many layers,thickness measurement and lesion volume in macular OCT images.The main points are as follow:In order to overcome the large number existence inherent speckle noise and vascular artifacts in the OCT image,which interfere the detection of the boundaries between the inter layers.Based on the graph optimal technology,a novel automated intra-retinal layers segmentation method was proposed in this paper.By referring to the object structural integrity principle of the Gestalt psychology,firstly,the proposed method defined a connected component as the vertex in the graph,constructed a complete weighted graph that could well represent the OCT image,and fused local region information that could improve the quality of image processing.Then,we applied the current popular graph optimization techniques to extract each boundary.Finally,the research demonstrated that the proposed approach could effectively and robustly segment eleven boundaries of ten retinal layers in OCT images,and avoid well the disturbed factors of the speckle noise and vascular artifacts.In order to overcome some irregular structure factors such as local some vitreous artifacts,which affect the detection of the boundaries such as the internal limiting membrane.Based on the graph-based a simple linear iterative clustering super-pixels and manifold ranking technology,a novel automated intra-retinal layer segmentation method was proposed.Instead of considering the intensity or gradient features of the single-pixel in most existing segmentation methods,the proposed method focused on the texture information and spatial relationship of the super-pixels and the connected components.Firstly,the proposed method used super-pixels to eliminate the vitreous artifacts disturbance.Then,each vertex was optimally ranked by using the manifold ranking technology for the three-stage scheme,and each boundary was detected.Finally,the experiments demonstrated that the proposed method could fast and robustly extract and segment eleven boundaries and ten intra-retinal layers in OCT images,effectively.In order to aviod the influence of the central serous retinopathy on detecting the boundaries around the pigment epithelium layer.In this paper,an edge super-pixels method was proposed to segment intra-retinal layers of macular OCT images with central serous retinopathy.Firstly,the proposed method defined the detected the connected component into the edge super-pixel,and made the regularization of the edges super-pixel,which could avoid the disturbance from the discontinuous edges and the irregular structures.Then,the proposed method constructed only the weight graph by the edges super-pixel,and optimized the boundaries detection of the three stages into two stages by manifold ranking technology.Finally,the proposed could rapidly and effectively segment automatically intra-retinal layer in the weak contrast images with the central serous retinopathy.For the weak contrast macular OCT images with central serous retinopathy,since there existences some disturbed factors such as fuzzy and uneven gray,and irregular spatial structure.In this paper,a coarse-grained diffusion mapping method was proposed to segment the retinopathy region in OCT images.Based on the intra-retinal layers segmentation of the central serous retinopathy OCT images using the edge super-pixels method,firstly,the proposed method segmented the intra-retinal layers of OCT images with the central serous retinopathy,then,segmented the super-pixels around the pigment epithelium layer.Finally,we used the coarse-grained diffusion mapping method to detect the lesion target area accurately and effectively,and achieved the 3D visualization of the retinopathy and the quantitative measurement of the retinopathy volume.We also compared successfully the 9 sectors of each intra-retinal layer of macular OCT images between the healthy and central serous retinopathy.