Research On Medical Image Pre-processing And Implementation Of 3D Reconstruction Software

At present,Computed Tomography and Magnetic Resonance Imaging technologies are able to take the images of the inner part of patients.Furthermore,In order to inspect the situation of cell's pathology,biologists or doctors cut cells and then put them on the slide,and finally use the transmission electron microscopic to take the image of the cell's slide.Adopting three dimensional visualization technology,three dimensional structure of organ or cell from a serial of consecutive images will benefit medical researchers and doctors to observe the pathology from any aspects of the three dimensional structure.In addition,it is easy for cells to be destroyed because of section part,such as like folding,tearing,deforming.Therefore,the rigid alignment of the consecutive images does not satisfy this situation.At the same time,since the automatic exposure and gain control of camera and inconsistent illumination,the intensity of serial images are different.Therefore,this thesis will focus on medical image alignment and intensity normalization.At last,we implement three dimensional reconstruction software based on above research results and medical image processing and visualization programing libraries.Firstly,we do a research on two-stage non-rigid alignment among serial medical images.At beginning,regarding the rigid alignment based on corresponding feature points,we research the fast method to extract corresponding feature points and remove the false matched points.By this way,the images are aligned coarsely.In the follows,as for cell microscopic images,we adopt modified optical-flow based non-rigid image alignment to align cell microscopic imaged because they may be folded,tear,deformed due to section processing.Secondly,we do a research on intensity normalization method for serial images.Since the automatic exposure and gain control of camera and inconsistent illumination when we take images of cell,the intensity of serial images are different.In this thesis,we modify intensity histogram based intensity normalization method to eliminate the variation of intensity.The experimental results illustrate our method will make the consecutive two images less difference,because the background and non-aligned part are removed.At last,we summarize some medical image segmentation and rendering method of three dimensional image.We implement a three dimensional software based on above research results and medical image processing and visualization programing libraries.