Analysis of simultaneous EEG-fMRI data for discovering the association between EEG activities and fMRI functional connectivity

Abstract

Simultaneous electroencephalogram (EEG) and functional magnetic resonance imaging (fMRI) (EEG-fMRI) is a powerful neuroimaging technique that provides high temporal and spatial resolution for probing brain functioning. In particular, EEG-fMRI allows for identifying the association between EEG activities and fMRI functional connectivity (FC), which is attracting increasing attention because such EEG-FC association could shed light on the neural mechanism and functional relevance of brain networks. This thesis aims to address two challenging problems in processing EEG-fMRI data for identifying EEG-FC association: (1) how to improve the signal-to-noise ratio of EEG simultaneously collected with fMRI; (2) how to estimate the association between EEG and FC metrics in specific experimental paradigms.

First, simultaneously collected EEG is severely contaminated by two types of artefacts: gradient artefact (GA) and ballistocardiogram (BCG) artefact. To remove these artefacts, two novel signal processing methods were developed in this thesis. For GA removal, a least across-segment variance (LASV) method was proposed to correct the desynchronization between simultaneously recorded EEG and fMRI data, which could drastically hamper the performance of GA removal. The results of simulation and real data tests demonstrated that, by minimizing across-segment variance of EEG, the LASV method is able to remarkably reduce the negative influence of EEG-fMRI desynchronization on GA removal. For BCG artefact removal, a multi-channel-EEG-based heartbeat-timing estimation (MEHE) method was proposed to estimate heartbeat timings (which are essential for BCG artefact removal) without electrocardiography recording. The results of real data test demonstrated the high accuracy of MEHE method.

Second, EEG-FC association was typically studied in one specific resting condition, but the methods for studying one specific resting condition may not be suitable for investigating EEG-FC association in two other important experimental paradigms: (1) event-related paradigm; (2) resting paradigm with two different conditions. This thesis developed new analysis methods to explore EEG-FC association in the above two paradigms. For the event-related paradigm, a new data analysis framework based on task-regressed FC was proposed to analyze the association between spontaneous FC and event-related EEG response. This new framework was applied to an auditory EEG-fMRI dataset and found a significant correlation between FC of bilateral Heschl’s gyrus (HG) and N1 magnitude of auditory evoked potentials, suggesting an important role of FC in auditory onset processing. As for resting paradigm with two conditions, a novel multivariate method named combined covariance difference maximization (C-CDM) was proposed to explore between-condition difference in EEG-FC association. The results of simulation and real data tests demonstrated that, by maximizing between-condition difference in latent covariance between time-varying EEG and FC metrics, the C-CDM method is able to extract meaningful information regarding EEG-FC association with high accuracy, holding great promise to disclose the neural basis of FC dynamics.

In summary, this thesis contributes to two critical problems in analyzing EEG-FC association, namely, EEG de-noising, and analysis methods for two important paradigms. These methods have high potential for exploring EEG-FC association in a large variety of new experimental designs or conditions, providing powerful tools for improving our understanding of brain networks. (495 words)