Abstract
In this dissertation novel methods are proposed for analyzing EEG and fMRI. These include an advanced technique developed to integrate these modalities. Majority of the proposed methods in this thesis are based on blind source separation (BSS) concept. Artifact removal is an essential task to prepare EEG signal recorded simultaneously with fMRI for further processing. This is tackled using two new approaches. In the first method, a hybrid independent component analysis (ICA) plus discrete Hermite transform (DHT) is developed. The second method employs a new cost function to perform source extraction based on joint short-and-long term prediction. The main objective of this work is to incorporate the prior information about the temporal structure and periodicity of ballistocardiogram (BCG) artifact into separation procedure. The main objective in fMRI analysis is detection of blood oxygenation level dependent (BOLD). General linear model (GLM) is a widely used technique for this purpose relying only on stimuli onset times and predefined haemodynamic response function (HRF). In this thesis, several BSS methods mainly based on non-negative matrix factorization (NMF) are developed for fMRI analysis. The main superiority of these techniques over GLM is their model-free nature. However, we demonstrate that the performance of these techniques can be improved by exploiting some statistical and physiological prior information. This leads to an advanced approach that does not entirely rely on a predefined model (in contrast to GLM) while taking advantages of all existing information. An important step in our approach for EEG-fMRI fusion is through estimating the fMRI time course using the EEG signals. One approach is by detection of movement onset from brain event-related oscillations. Extracting these oscillations is challenging due to their non-phase-locked nature and inter-trial variability. In this research, a novel method based on linear prediction is proposed to extract rolandic beta rhythm from multi-channel EEG recording. This technique employs a spatio-temporal constraint to effectively extract beta rhythms to study post-movement beta rebound. The results are used to construct a regressor for fMRI analysis in a combined EEG-fMRI paradigm. The last contribution in this thesis is development of a novel technique for EEG-fMRI fusion. This method combines the reconstructed time course using the extracted beta rhythm and fMRI using a partially constrained algorithm. PARAFAC2 is used for this purpose. The obtained results identify the voxels which are involved in post-movement beta rebound due to performing a motor task.