功能磁共振数据结构性噪声分析

本文提出了一种非参数化功能磁共振数据噪声分析方法.该方法根据噪声时域特性将其分为结构性噪声和非结构性噪声两类,它们对统计检验功效造成的影响在文中分别进行了讨论.使用正则相关分析提取数据中的结构性信号,利用基于降阶自回归模型的surrogate检验从结构性信号中确定出神经响应信号;利用随机化方法在保持非结构性噪声能量不变的情况下消除它们时域中的自相关性,使得非结构性噪声谱低频部分的能量下降.利用神经响应信号和经过随机化处理后的非结构性信号重构数据.重构数据基本满足了多种统计推断方法中白噪声的假设.构造了2种仿真数据,使用基于多窗口谱估计的F检验来验证算法的有效性.最后用此方法处理了20组真实的功能磁共振数据,成功提取到了一些在未降噪数据中检测不到的任务相关脑功能区.

Structured Noise Analysis for Functional Magnetic Resonance Image

A nonparametric fMRI data noise analysis procedure is proposed in the paper.The fMRI noise is classified into structured and unstructured noise by their temporal characteristics,and their impacts to efficiency of statistical tests are discussed separately.The canonical correlation analysis technique is exploited to extract the underlying structured signals from which the neural response signals are picked up by surrogate test based on reduced autoregression model.The temporal autocorrelation of unstructured noise is eliminated by the randomization method without changing the spectral power.The spectral power of noise in low frequency range descends and the "white noise" assumption applies better after the application of the procedure.Two kinds of simulation data sets are generated and to which the F-test based on multitaper spectral analysis is applied to test the validity of the technique.Twenty sets of fMRI data are then processed and some task-related areas go undetected in original data are gotten.