Research On Tensor Atlas Of Tree-dimensional Human Brain On Diffusion Tensor Imaging

Diffusion tensor imaging is a non-invasive imaging technique and detecting thediffusion characteristics of the water molecules in different brain regions, thus providesthe direction information of white matter fiber tracts which the structural MRI isdifficult to capture. The imaging method extend the image collection, pre-andpoet-processing, and statistical analysis in the field of neuroimaging and contributes todevelop new representation theory of the micro structure of human brain.The brain DT atlas of diffusion tensor imaging is a very important direction in thestudy of the neuroimaging. The research direction through the acquisition of largesample set normal diffusion tensor imaging data, modeling on statistical significance,and construct contains the digital statistical atlas which contains sample concentrationof normal anatomic structures of common features, is the basis for the detection of brainwhite matter structure form. At present, the domestic and foreign related research is stillin a fledging period.Diffusion tensor image is a three-dimensional second-order tensor in mathematicalsense. So the construction of DTI atlas relates to the problem such as the tensor modelbuilding, tensor image registration, the average algorithm of tensor image. In this paper,techniques of atlas construction for traditional scalar MR images are improved and usedin construction of the DT atlas. The major contributions of the dissertation are asfollows.First, in the process of building a second order tensor field model, introduces a newEPI series correction method for image correction. Using this calibration method couldreduce the DTI image scanning process of dynamic error and the vortex noise a certainextent, and could enhance the image contour information at the same time, so that canconstruct more accurate second-order tensor field model.Second, the use structure of T1images for medical image registration, then theregistration parameters effect on DTI images. The structure of T1image has the same asthe DTI images geometry structure information of the brain, and relative to the DTIimages, T1image structure has better SNR, using T1structure image match on timecould get more accurate registration parameters.Again, we use two different tensor image average algorithm to build the atlas. Dueto the specificity of the second order tensor image structure, in addition to general linearaverage algorithm, our research introduces a average algorithm based on Frechet average to average tensor images. Then, we use a variety of evaluation methods toevaluate the two average algorithm.Finally, we use a group-based normal brain tensor atlas construction method tobuild the tensor atlas. This method can build the sample concentration field of eachindividual and average space transformation, and transformation in the process of fieldto image registration. To some extent, this method can reduce the sample concentrationdifferences between individual subjects, thus better reflecting the commoncharacteristics of sample concentration of normal anatomic structures. The specificresearch needs the tensor map can reduce the deviation of image registration, but alsoenhances the reliability of the results of statistical analysis.