EEG-fMRI Fusion Based On Bayesian Theory

Electroencephalogram (EEG) and functional magnetic resonance imaging (fMRI) are the mostly used two predominant techniques for their ability to reveal noninvasive brain mapping of the mental process. Simultaneous EEG-fMRI recording provides complementary information about the cerebral activity, and the EEG-fMRI fusion enables a better understanding of the brain with the enhanced spatiotemporal resolution. The work in this dissertation concentrates in the EEG-fMRI fusion based on Bayesian theory and its application in cognitive and clinic problemes. We start with an introduction of the fMRI-constrained EEG source imaging and its applications in multi-modal face study and epileptic discharges. We then develop an EEG-informed fMRI analysis and a novel spatial-temporal EEG-fMRI symmetric fusion. The proposed symmetric fusion is applied in the study of partial epilepsy. Finally, we introduce a systematic perspective of the EEG-fMRI fusion, which is inspired by the proposed symmetric fusion and the harmonic balance between yin and yang. The main contributions of this dissertation are as follows:1. NEtwork-based SOurce Imaging (NESOI) is a new method to reconstruct neuroelectric sources based on empirical Bayesian model. In NESOI, multiple functional networks derived from fMRI are employed as constraints for EEG source imaging. In contrast with previous applications of empirical Bayesian model in source reconstruction with smoothness or sparseness priors, functional networks play a fundamental role among the priors employed by NESOI. Using synthetic and real data, we systemically compared the performance of NESOI with other source inversion methods when fMRI priors are used or not used. Our results indicate that NESOI is a potentially useful approach for understanding the electrophysiological signatures of fMRI resting state networks.2. Base on empirical Bayesian model, an EEG-informed hemodynamic response function (HRF) estimation is proposed to reconstruct the hemodynamic fluctuation related to EEG features. This estimation and NESOI combine into a parallel fusion, termed Spatial-Temporal Eeg-Fmri Fusion (STEFF), to symmetrically integrate the simultaneous EEG-fMRI recordings. STEFF enables information of one modality to be utilized as priors for the other and hence improves the spatial (for EEG) or temporal (for fMRI) resolution of the other modality. Simulations under realistic noise conditions indicated that STEFF is a feasible and physiologically reasonable hybrid approach for spatiotemporal mapping of cognitive processing in the human brain.3. STEFF is applied in simultaneous EEG-fMRI recording for the partial epilepsy study. As interictal epileptiform discharges related components are widespread, STEFF classifies the fMRI component as a function of response sign (positive or negative), peak delay of HRF and consistence of the spatial pattern. Our results indicate that the EEG-fMRI spatial consistent components with early HRF peaks would be the indictors of the epileptogenic focus. STEFF make possible the discripting of the dynamic responses of epileptic networks with bioelectric and hemodanimic information.4. Inspired by STEFF and the harmonic balance between yin and yang, a systematic framework is proposed to break the boundary between data level fusion and feature leve fusion. Based on this framework, we discuss many newly emerging fusion methods, including fMRI-constrained EEG imaging, EEG-informed fMRI analysis, and EEG-fMRI symmetric fusion. Our systematic perspective is helpful in explaining the relationship between different fusion schemes: the levels of signal abstraction and the complementary natures of EEG and fMRI. Moreover, some schemes that are little investigated but have great potential are also revealed in this framework.