Research On Layer Segmentation Of OCT Fundus Images And Detection Of Optic Disc Center

As China has entered the ageing population period, the probability for the elderly to suffer from diseases, especially eye diseases is increasing each year. The way to rapidly detect eye diseases has become a task demanding an early solution.The fundus of the eye is the interior surface of the eye, opposite the lens, and includes the retina, optic disc, macula and fovea, and posterior pole. Medical signs that can be detected from observation of eye fundus, and doctors can distinguish the possible diseases that a patient might have through inspecting those signs and even predict the early phase diabetes.Generally, in order to identify the possible fundus sickness, doctors would take fundus image for the patient. There're mainly Color images, IR image, ultrasound image, and the new OCT image.In this paper, medical image processing about eye and fundus is mainly concerned, including: OCT fundus image layer segmentation, detection of optic disc center in color and IR fundus image.The main works and innovation points of this thesis are described as below:1. An OCT fundus layer segmentation based on bilateral filter, canny operator, morphological operations and dynamic programming for shortest path; The bilateral filter greatly reduce the noises in the original OCT image while keeping the edge information of each visible layer; Canny operator is used to search edges with different brightness in the result of bilateral image, and then enhance those edges; Morphological operations are proceeded to eliminate the trivial information; And finally, on the weighing image of original image, direction gradient of Gaussian filtered image, canny image, a dynamic programming algorithm is executed to find a path with least cost, so as to be defined as the optimum path, in another word, one layer. By using different parameters, different layers would be segmented.2. In the part of optic disc center detection, 2 methods are used:For color fundus image, by using Gaussian Vessel detector, the vessels are separated as the region of interest, and on the corresponding area of the original image, the vessels are replaced with the surrounding non-vessels pixels, so that the vessels are eliminated; next, a Prewitt edge detection algorithm is used to detect the edge of optic disc, and finally, by using the tangent transformation like Hough transformation and the trait of ellipse, the optic disc center is detected.For IR grey level fundus image, a large kernelled Gaussian filter is firstly applied to get rid of the vessels, and then a designed line operator which is only sensitive to the change of luminance is used to process the vessel-free image. After that, a pre-defined cycle mask is searched on the resulted image, and the point with least difference is chosen as the center of optic disc center.The algorithms proposed in this paper have achieved ideal results in both OCT and IR images from TopCon and Color images form the standard S.T.A.R.E fundus image database.