采用偏微分方程的磁共振弥散张量各向异性扩散滤波去噪

Diffusion Tensor Magnetic Resonance Imaging Denoising Using Anisotropic Partial Differential Function Filter

弥散张量磁共振成像(DTI)是无创研究大脑白质结构及其他人体纤维状组织结构的主要工具。由于合成DTI数据的弥散加权成像数据(DWI)易受噪声干扰,需要有效去噪以保证DTI数据精度和后续应用的实现。使用各向异性扩散滤波理论,综合考虑各方向通道DWI数据的几何形态和结构特点,重构其特征向量和特征值,获得统一平滑的结构张量,以期在有效去噪的基础上最大程度地保持DTI数据几何结构和特征。利用所提出方法在合成弥散张量数据上进行仿真,并在真实脑部DTI数据上进行实验。仿真和实验结果表明,该方法能有效减少噪声对DTI数据的影响,较之现有的时频分析去噪方法,可更准确地恢复DTI数据,减少主分量方向的偏差和部分各向异性值的误差。

Diffusion tensor magnetic resonance imaging（DTI） is the main non-invasive utility to reveal the information of local diffusivity of white matter and other fibrous human tissues.Because the diffusion weighted image（DWI） which is used to acquire DTI is sensitive to noise,effective noise removal is required to improve the accuracy of the DTI data and its subsequent applications.Instead of denoising the DWI in different direction separately,an anistropic filtering method synthetically considering the structure information and characteristic of the DTI was proposed in this paper.The eigenvalue and eigenvector of the smoothing structure tensor were reconstructed to remove the noise and keep the structure characteristic simultaneously.Simulations using a synthetic DTI dataset and experiments using an in vivo brain DTI dataset were performed.The results demonstrated that the noise could be removed significantly and the direction bias of the main eigenvector and the error of fractional anisotropy were reduced effectively compared with the common methods.