基于独立成分分析的时间自相关方法在功能磁共振激活区自适应提取中的应用

Application of fMRI Activation Area Adaptive Extraction Based on Temporal Self-Correlation Method of ICA

传统基于ICA的激活区检测手段是将分离后的独立成分与参考信号做相关性分析。实际问题中,不同区域的脑血流动力学响应情况不同,因此往往得不到标准的参考信号。针对此类问题,提出时间自相关方法(TSC)与ICA方法结合,在不需要参考信号的情况下,通过检测体素点各周期的时间序列相关性,对fMRI数据进行激活区提取。应用5-邻域ICA方法对fMRI数据逐点处理,然后应用时间自相关算法检测各时间序列周期间的相关性,选择最大的自相关系数作为该体素点的信号值。再通过Z变换将相关系数分布转换为服从N(0,1)的Z分布,提取出具有显著性差异(a=0.05)的激活区。将自相关算法应用于仿真数据和12组双手握拳运动的真实fMRI数据的处理,结果表明该方法能够准确提取出仿真数据中的激活区。对真实数据的处理,该方法在空间准确性上与GLM方法无显著性差别(0.465 3±0.136 8 vs 0.490 5±0.134 1),在时间准确性上显著优于GLM方法 (0.636 4±0.011 1 vs 0.369 2±0.010 9),具有良好的脑功能激活区检测及空间定位能力。

The traditional ICA based activation zone detection is to analyze the correlation between the separated independent component and the reference signal. However, in practical problems, since the differences among the hemodynamic responses of the cerebral regions, the standard reference signal is often not available. Aiming at such problems, in this paper, the method of temporal self-correlation （TSC） combined with infomax-ICA was proposed. This method processed fMRI data point by point with 5-adjacent voxels based ICA, then detected the correlation between each time series period with temporal self-correlation algorithm and selected the maximum autocorrelation coefficient as the signal value of the voxel. After that we conversed correlation coefficient distribution to Z distribution which obey N （0, 1 ） by Z-transform, extracted the active regions with significant difference （ α = 0.05 ） in the statistical parametric mapping. The algorithm was applied to deal with simulation data and 12 set of real fMRI data of fist movement with both hands. Results： The method can accurately extract the active region of the simulation data. For real data processing, results of this method have a high similarity with GLM method in the spatial domain （ 0. 465 3 ± 0. 136 8 vs 0. 490 5 ±0. 134 1） and better than GLM method in the temporal domain （0. 636 4± 0. 011 1 vs 0. 369 2 ± 0. 010 9）. These results have statistical significant. Experimental results showed that, this method has good capacities of detection of functional brain activation areas and spatial orientation.