Research On Key Issues Of Vessel Segmentation And3D Reconstruction Of OCT Medical Imaging

With the development of medical science, optical techniques are of particularimportance in the medical field, and offer the therapeutic and diagnostic potentials.Especially optical coherence tomography (OCT) imaging technology, can providemicrometer-resolution, high sensitivity and non-contact depth resolved cross-sectionalimages of biological samples. Therefore, how to combine the knowledge of imageprocessing to analyze OCT images have become the research topics of optical image.Among many OCT technologies, cmOCT (correlation mapping optical coherencetomography) is the center of attention in research, which is a proposed technique thatenable mapping of vasculature networks and achieved as a processing step on OCTintensity images that does not require any modification to existing OCT hardware.In this thesis, we have studied traditional cmOCT technique and proposeimproved methods to overcome pre-existing problems based on this technique. Weimplement some feasible image processing methods to obtain in vivo vascular imagesof mouse ear and show the vascular tissues reconstructed using three-dimensional (3D)reconstruction system we developed in this research. The main works of this thesis areas follows:1. Image enhancement in OCT. Due to OCT images exist a large number ofspeckle noise, is not conducive to the extraction of vessel contours. Therefore,we need to use image enhancement for image pre-processing. We applytraditional Gaussian filtering, median filtering, and mean filtering to enhanceimage individually. Novel anisotropic diffusion filtering algorithm is alsoadopted for image enhancement. Median filtering is sensitive to salt and peppernoise, so can effective at removing them. Anisotropic diffusion method can’tonly reduce speckle noise without removing significant parts of the imagecontent, typically edges, lines or other details that are important for theinterpretation of the image but enhance the effects of edge. It is more obvious forimage enhancement.2. Binarization threshold selection. We implement Otsu algorithm, KSW algorithm,Niblack algorithm, and expectation algorithm as binary threshold selectionmethods. After analysis, Otsu and KSW algorithms belong to global thresholdmethod, is only suitable for processing a single vascular image. However, defacto in vivo blood vessels are complex and different sizes, so we propose themethod based on gray-level expectations with classical Niblack algorithm asbinary threshold selection method and acquire better results. 3. Ameliorative cmOCT method. cmOCT algorithm is achieved by calculating thecorrelation coefficient value of pixels between two adjacent frames at the sameposition. This is necessary to choose an appropriate kernel size and slide overtwo images for different sizes of vessels. In this research, we propose anameliorative cmOCT algorithm. On the basis of traditional cmOCT technique,we combine image enhancement with image segmentation to improve and solvepre-existing problems. Then we employ improved algorithm to in vivo OCTimages of mouse ear and obtain the contour of blood vessels.4.3D reconstruction of cmOCT. After acquiring2D vessel contour from OCTimage in previous step, and then we will carry out3D vessel reconstruction. Weput forward different3D reconstruction methods, and compare the results of3Dmodel with commercial software.