静息态功能磁共振数据的非局部空间平滑方法研究

After nearly two decades, resting-state functional magnetic resonance imaging (R-fMRI) is becoming matured and rapidly emerging as a highly powerful tool in discovering various clinical neuropsychiatric disorders. It can measure the intrinsic functional activity of the human brain in vivo. Based on the R-fMRI signal, we can compute various functional indices to characterize the functional architecture of the human brain. In such a procedure, raw functional imaging data must be preprocessed with some algorithms, among which Gaussian spatial smoothing is the most common and critical for subsequent analyses to: 1) increase signal-to-noise ratio, 2) reduce the effects of ringing in images due to the restriction of sampling to a finite k-space region, 3) improve the errors on group-level statistics caused by imperfect spatial normalization, and 4) ensure the assumption on Gaussian random field during correction for multiple comparisons. However, contrasting with advances in fast MR imaging technology, as a local spatial averaging method, Gaussian smoothing is limited because of to: 1) increase partial volume effects, 2) reduce the spatial resolution, and 3) hurt the spatial accuracy of functional localization and barrier structure-function association studies. In this proposal, the non-local diffusion or smoothing from the theory of partial differential equations will be employed to overcome above shortcomings of Gaussian spatial smoothing. The impact of non-local smoothing on R-fMRI data processing will also be systematically investigated. At both individual and group levels, we will compare the performance of the two spatial smoothing methods. This study will provide novel methods for R-fMRI data preprocessing, improve the current image processing pipeline and provide new perspectives on processing R-fMRI data with high temporal and spatial resolutions as well as vehicle R-fMRI clinical applications.

近20年来，静息态功能磁共振技术(R-fMRI)逐显成熟并成为探索人脑功能和神经精神疾病的有力工具。R-fMRI能测量活体大脑内在功能活动。基于R-fMRI信号，可以计算各种指标来刻画人脑内在功能特性。在此过程中，原始数据必须经过一些预处理，高斯空间平滑是其中最常见和重要的一步，用来：增强信噪比、降低环效应、改善组分析误差和满足高斯随机场需求。随着快速磁共振成像技术发展，高斯平滑的局限凸显：加剧部分容积效应、降低空间分辨率、损害功能定位精度和限制结构功能关联。本项目采用偏微分方程理论中非局部扩散方程克服上述缺陷，研究非局部空间平滑对R-fMRI图像处理的影响。在个体和群组水平上，比较非局部平滑与高斯平滑在R-fMRI计算中的不同。本研究将为R-fMRI图像处理提供新方法，改进目前的R-fMRI图像处理流程，为高时空分辨率的R-fMRI计算提供新思路，推动其在神经和精神疾病研究中的应用。

静息态功能磁共振技术(rfMRI)逐显成熟并成为探索人脑功能和神经精神疾病的有力工具。RfMRI原始数据必须经过一些预处理，空间平滑是其中最常见和重要一步，用来：增强信噪比、降低环效应、改善组分析误差和满足高斯随机场需求。随着快速磁共振成像技术发展，高斯平滑局限凸显：加剧部分容积效应、降低空间分辨率、损害功能定位精度和限制结构功能关联。本项目采用偏微分方程理论中非局部扩散方程克服上述缺陷，系统地研究了非局部空间平滑对R-fMRI图像处理影响，开发了基于时间点的rfMRI非局部平滑的降噪算法，将其整合到人脑连接组计算系统（CCS）。项目团队参与研发rfMRI计算软件一套，作为共同通讯作者在5年影响因子超过10的杂志上发表论文一篇，并成为该刊Top25最受关注论文之一，提出了rfMRI数据处理与分析的标准规范。本研究成果为rfMRI图像处理提供了新的降噪方法，改进了目前的rfMRI图像处理流程，提高了基于rfMRI的人脑功能连接组学的重测信度。