DTI Image Denoising Algorithm Based On Anisotropic Filtering

Diffusion tensor imaging(DTI)is a well-known brain imaging method,which plays a very important role in the research of brain neural network,medical diagnosis and treatment.Due to the limitation of imaging mechanism,the DTI images are seriously interfered by noise.On the one hand,the noise in DTI image will cause different types of deviations;on the other hand,it will cause distortion of DTI image,which limits the development and application of DTI to some extent.Therefore,the research of new DTI image denoising algorithm is of great significance to the later applications of DTI image.The DTI image denoising method can be summarized into three types: the first one is to denoise the DTI image directly.Although these algorithms can remove the noise in DTI image effectively,the boundary is blurred.The second way is to process the characteristic information of the diffusion tensor.Despite these methods alleviate the phenomenon of blurred boundary to a certain extent,the result of filtering is still unsatisfactory.The third scheme is to estimate and denoise the diffusion tensor at the same time.This method obviously reduces the noise in DTI image and achieves the preservation of boundary,but the algorithm is time consuming with large amount of computation.Aiming at the limitations of the above-mentioned algorithms,in order to decrease the influence of noise in DTI image and preserve the information of boundary effectively,this dissertation proposes three new DTI image denoising algorithms with the combination of the structure tensor,Riemannian geometric framework and complex shearlet transformation.The main research is as follows:(1)DTI image denoising based on structure tensor and anisotropic filteringIn order to reduce the influence of noise in DTI image and preserve the edge information effectively,a new DTI image denoising method is proposed by combining structure tensor and anisotropic filtering technique.Firstly,the pixels of the DTI image are divided into homogeneous and boundary areas by structure tensor.Then,the isotropic filtering is performed in the homogeneous region,and the anisotropic smoothing is performed in boundary.Finally,we obtain the denoised DTI image.The algorithm combines the structure tensor that can distinguish structure of the image with the anisotropic filtering that has the capabilities for retention of boundary,which shows good performance in retaining theboundary of the image and removing noise.(2)DTI image denoising based on Riemannian geometric frameworkUp to now,algorithms and tools that were developed to deal with diffusion tensor treat tensor as linear entities,with the neglect of the symmetric positive definite constraint of the tensor under practical physical meaning.To this end,our approach is grounded on the differential geometrical properties of the space of multivariate normal distributions,where it define an affine-invariant Riemannian metric and express statistics on the manifold of symmetric positive definite matrices,thus a novel diffusion tensor denoising algorithm based on Riemannian geometric framework is introduced.The proposed denoising algorithm not only covers the image local characteristics of DTI data,but also considers the specific Riemannian structure of the DTI image,which extending the concept of anisotropic diffusion filtering from Euclidean space to Riemannian geometry space.(3)DTI image denoising based on anisotropic filtering and complex shearlet transformationIn order to preserve the information of edge effectively while suppressing noise in DTI image,a new DTI image denoising algorithm is proposed by applying complex shearlet transformation in multi-scale geometric transformation with the combination of the anisotropic filtering based on structure tensor and Riemannian geometric framework.Firstly,the DTI image is decomposed into the high frequency coefficients and low frequency coefficients by complex shearlet transformation.Then the DTI image denoising algorithm based on structure tensor and Riemannian geometric framework is used to remove the noise in high frequency coefficients and low frequency coefficients respectively.Finally,the denoised DTI image is obtained by conducting inverse transformation of the complex shearlet transformation.The algorithm not only suppress the noise in DTI image effectively,but also preserve the texture information of boundary in DTI image efficiently.