Retinal OCT Image3-D Segmentation Based On The Combination Of Fuzzy C-means Algorithm And Expectation Maximization Algorithm

Abstract:Optical coherence tomography (OCT) allows high resolution and noninvasive imaging of the structure of the retina. The intraretinal layers of retina may be affected differently by disease. However, it is difficult to get the accurate segmentation of retina physical layers because of a large amount of noise in retinal OCT images. At present, the existing segmentation methods based on3-D information are very little. These methods are too simple to get the accurate result. Therefore, it is necessarily to research the3-D segmentation method of retinal OCT images, which is stronger antinoise and more accurate. The method provides the basis for quantitative analysis in the early diagnosis of eye disease.This paper sums up the general process of the segmentation methods, which are based on the characteristics of retinal OCT images. The typical algorithm of retinal OCT images segmentation methods is discussed in this paper. The typical method is tested by the experiment of the retinal OCT images segmentation. The experimental results show that the typical method can get6layer-interface from retinal OCT images. But it will be failure in the central fovea area, which can only get3layer-interface. The advantages of typical method is that it has a fast operation speed and simple algorithm. But, its poor accuracy and insufficient layer-interface of segmentation still need to be concerned.As the existing methods for retinal optical coherence tomography (OCT) images segmentation cannot get the exact multilayer segmentation, this paper proposes a retinal OCT image segmentation method. This method is combined the fuzzy c-means clustering (FCM) algorithm and expectation maximization (EM) algorithm. By using the fuzzy c-means clustering to fuzzy clustering for former images corrupted by noise, the boundaries of inner limiting membrane (ILM), inside and outside section layer (IS/OS), and pigment epithelium (RPE) were obtained in clustering images. The extraction retinal area is smooth by the anisotropic diffusion. The accurate locations of retinal physical layers are obtained by combining the EM algorithm with image gradient change, The obtained retinal physical layers are the nerve fiber layer (NFL),the inner nuclear layer(INL), the outer plexiform layer (OPL)and the position of the visual cell layer (PL).The experiments were carried out on299OCT images corrupted by noise(3-D image data of the retina). The accurate segmentation of seven layers mentioned-above and the retina thickness were obtained. The influence of speckle-noise and nearby physical layers can be avoided in segmentation process. Compared with the existing3-d segmentation method, the method proposed improves the accuracy of segmentation results, while maintaining the computational speed.