Group Analysis Of FMRI Data Based On IVA And Group-ICA

Functional magnetic resonance imaging (fMRI) is widely used in brain function research and clinical diagnosis due to its high spatial resolution and non-invasive characteristics. Analysis of multi-subject fMRI data is called group analysis which can obtain the required brain components for brain studies and clinical diagnosis. Two typical data-driven methods for group analysis of fMRI data are independent vector analysis (IV A) and group independent component analysis (GroupICA). However, these methods exhibited some limitations in analyzing in the real-valued and complex-valued fMRI data. As such, we carry out the following work in this thesis:(1) An adaptive IVA algorithm is proposed since the existing IVA algorithms assume fixed and known, probabilistic model of source component vector (SCV) Our algorithm utilizes the alterable multi-dimensional exponential power (EP) model to adaptively match the SCV distribution, thus needs little prior information about the probabilistic model of source data and has a wide range of application. Meanwhile, this method utlizes both second order statistics and higher order statistics of SCV to further improve separation performance. The experiment results with simulated and real fMRI data both show the efficiency of our algorithm.(2) The thesis also proposes an improved complex IVA method aiming as the existing algorithm did not perform well. Specifically, the real-valued IVA-L is extended to complex domain by adding the stability evaluation of non-circularity and non-gaussianity, and the EP distribution is also used as multi-dimensional probability model The experiment results with simulated and real fMRI data demonstrate that the proposed algorithm performs better than the complex-valued IVA-G method.(3) Since the inherent phase ambiguity of ICA affects the group analysis performance to some extent, the phase correction algorithm is incorporated to Group ICA algorithm, which can effectively improve the group analysis of complex-valued fMRI data. Experiment results verify the validity of the phase correction.(4) To further verify the efficiency of the different IVA and group ICA with phase correction, the thesis compares their effect on functional connectivity analysis. Results indicate that the proposed algorithms not only improve the accuracy of functional connectivity analysis, but also avoid misjudgment caused by phase ambiguity problem.