Based On Oct Image Automatic Detection And Segmentation Of Drusen

Research on the utility of automatic segmentation of drusen method has always been the goal, but also in recent years, been the focus of medical image segmentation. The automatic drusen detection methods are mainly based on the gradient and intensity information of images,, active contour, dynamic programming method and graph theory to extract target contours. This paper presents automatic drusen segmentation algorithms for two different imaging modes of retina OCT images. In this paper, the main work and research are as follows:(1) Considering the high brightness and high contrast imaging features in the first imaging pattern, an algorithm for segmenting retinal pigment epithelium (RPE) based on the local gray value is proposed, which makes full use of high brightness characteristic of RPE layer. The method based on local gray value is proposed to analyze RPE features, extract RPE edges and segment target regions. When RPE has retinal geographic atrophy (GA), the detection of RPE is often inaccurate. The problem can be solved by combining adjacent column edge’s information with RPE edge’s information of binarized images. The segmentation accuracy of RPE is improved. Experimental results demonstrate that this algorithm can effectively segment RPE layers.(2) Drusen detection is mainly based on threshold. According to drusen information in RPE regions, drusen can be detected by making full use of the threshold algorithm and comparing with normal RPE thickness. Experimental results demonstrate that this algorithm can effectively and accurately segment drusen.(3) Considering the strong noise, low brightness and low contrast imaging features in the second imaging pattern, an algorithm for reducing the impact of noise and retaining image details based on BM3D algorithm is proposed. The characteristics of the second imaging pattern which are not conducive to detect RPE can be solved by making full use of the priori target information of adjacent frames. The segmentation accuracy of RPE is improved.