Time series modeling of non-invasive measurement of brain functioning

点击翻译按钮获取中文摘要

In the present project we have developed a ToolBox for fMRI Connectivity Analysis. Here we employ NN-ARX model which we developed in our previous JSPS project. We use the model to whiten the 140,000 dimensional fMRI time series into 140,000 dimensional innovation series. Here all the obvious correlations between the neighboring voxels and the activated voxels by the stimulus are removed by the whitening. We proved that it can be effectively used for discovering dynamic connectivity between remaining remote voxels. The test version of the toolbox has been already used by our collaborators in National Institute of Physiological Science and Tohoku University Medical School for a few experimental data of fMRI. A method for estimating the dynamic causality between a few voxels is also implemented in the toolbox.We also developed a statistical method for decomposing the EEG/MEG data into component signals by using state space modeling techniques. Advantage over non-dynamic signal decomposition method such as ICA is confirmed.