基于时间聚类分析和独立成分分析的癫痫fMRI盲分析方法

提出了一种基于时间聚类分析和独立成分分析的癫痫fMRI数据盲分析方法,并将两种方法有效联合,提取发作间期的癫痫fMRI激活时空信息.该方法首先由时间聚类分析得到与激活相关的时间峰度特征曲线,以此特征作为时间参考信息;再由空间独立成分分析分解fMRI信号得到空间独立成分;最后将每个独立成分所对应的时间曲线与参考曲线做相关分析提取相应脑激活图.提出的方法无需任何关于癫痫fMRI的先验假设信息,有效解决了独立成分的排序问题,实现了对数据的盲分析.仿真试验结果阐明了这一方法的有效性及可靠性,对癫痫数据的试验结果显示空间定位准确性优于统计参数图方法.

Blind analysis of fMRI data based on temporal clustering analysis and independent component analysis

A blind analysis procedure combined temporal clustering analysis and independent component analysis approaches for epileptic fMRI was suggested in this paper, by which tempo-spatial characteristics of interictal epileptic fMRI activities dynamic responses can be investigated simultaneously. First, temporal clustering analysis was utilized to obtain temporal kurtosis pattern of the brain activity,
and the pattern was considered as temporal reference function; subsequently, spatial independent component analysis was employed to decompose fMRI signals into independent spatial patterns, each pattern being associated with a temporal course; finally, only the component for which corresponding time course was the most correlated with the reference function was considered. The proposed method could carry out blind analysis of epileptic fMRI data, and robustly recognize the components represented with brain activation without any prior information. The ordering of independent components was resolved effectively. The validity and reliability of the presented method were confirmed by simulated results. In vivo epileptic fMRI data analysis, the method was superior to the statistic parameter mapping (SPM) method on the accuracy of spatial localization.