A Study On Denoising Algorithm Of Diffusion Tensor Magnetic Resonance Imaging

Based on conventional magnetic resonance imaging(MRI), the principles ofDiffusion Tensor Magnetic Resonance Imaging (DT-MRI/DTI) involve theacquisition of diffusion-weighted images(DWI),DWI is sensitized in various bipolargradient directions, with one or more encoding levels in each direction, and itfollowed by pixel-by-pixel calculation and diagonalization of the diffusion tensor.DTI provides a non-invasive means of studying white matter fiber bundle in vivo,moreover it is the only one, it is an important tool used to observe white mattertracts of live humans for the study of connectivity of brain functional centers, braindevelopment, and white matter diseases. So it causes a great deal of attention in thetheoretical research and clinical applications.But during the acquisition of DWI processing, due to the impact of the noise,diffusion tensor also often contains noise components, and ultimately impacts fibertracking and fractional anisotropy (FA). The pre-processing of DWI image is a keystep in theoretical and applied research. At present, the denoising of DTI image hasnot formed a unified "gold standard", the purpose of this thesis is to find the bestdenoising methods of DTI image. The contents of this thesis are as follows:On the one hand, DWI image denoising algorithm is researched. According tothe characteristics of the DWI noise model, this thesis proposes a scheme thatLPG-PCA (Local Pixel Grouping-Principle Component Analysis) acts on DWIdenoising, and compare with the classic and the fashion algorithm, the simulatedand real data results show that the LPG-PCA method has a great advantage on peaksignal-to-noise ratio and edge preserve, this algorithm can effectively filter noiseand protect the image details, and the most important is that the direction of themain eigenvector of the diffusion tensor computed by the DWI of LPG-PCAdenoising is more consistent, the algorithm is good for fiber tracking. But at thesame time there are inconsistent eigenvector s of the diffusion tensor, the reason isthat there is still some noise in DWI map, there are some deviations with thecalculation of eigenvalues and eigenvectors, so there are some deviations in FA map.Thus, it is necessary to denoising FA map.On the other hand, FA denosing algorithm is researched. The smoothing filterused for VBA of DTI data is isotropic Gaussian filter, the size of filter is highlyvariable, and isotropic smoothing blurs the image features. Moreover it is based onFA denoising. This paper presents the first denoising DWI map, then FA map isdenoised by the anisotropic filter this paper presents, finally VBA analysis is conducted. From the two aspects of the peak signal-to-noise ratio and edge preserveindex, it is possible to provide a great quality data of FA for VBA.DWI denoising and FA denosing algorithm is researched, it can obtain greatquality of FA images. It is good for the subsequent use and analysis.