Integration of fMRI and MEG towards modeling language networks in the brain

Human language is a complex neurocognitive process that relies upon a widely-distributed network in the brain. With the advent of advanced neuroimaging techniques (i.e. fMRI and EEG/MEG), our understanding of language system is being transformed from functional segregation to functional integration over the last decade. Instead of asking "where' and "when" the task-related brain activity happened, researchers start to ask how the brain networks are modulated by the task. The main goal of this dissertation work is to elucidate language networks by integrating fMRI from a group of children who have participated annually in a longitudinal study from their childhood through adolescence and MEG data from the same group of children.;This dissertation consists of four main parts: (1) Neuroimaging data from each modality were analyzed separately and spatial maps were quantitatively compared. This is the initial step establishing a framework to integrate the two modalities, because it would provide confidence that fMRI could be used as spatial priors on MEG source localization. (2) Functional network connectivity supporting narrative comprehension was established using fMRI data only from two versions of the narrative comprehension task. FMRI provides us with the spatial information of the underlying network architecture supporting narrative comprehension. However, it is unclear that how the underlying mechanism of high-order cognitive processes modulates the brain networks during narrative comprehension task. Therefore, (3) in order to improve our understanding of these high-order cognitive processes, we integrated fMRI and MEG data within a Bayesian framework by applying a Multiple Sparse Prior (MSP) algorithm from Friston et al. 2008. Both simulated data and experimental data were examined. For experimental data, the group fMRI results were used as spatial priors in the MEG source reconstruction. As a result, we obtained fine spatiotemporal time courses from multiple elements of the brain-language networks. This step enables us to capitalize on the advantages of each modality and obviate the primary limitations of each, leading to an improved method for elaborating the complex network structure of language processing in the human brain. (4) Finally, we used Dynamic Causal Modeling (DCM), a recent network analysis technique, to study the fine spatiotemporal time courses from (3) in order to improve our understanding of how high-order cognitive processes modulate the pathways within the network architecture.;This dissertation makes several significant contributions to the neuroimaging field for better understanding of language networks. First, this is a first cross-modality validation study that qualitatively and quantitatively compares the fMRI and MEG data from the same subjects performing the same high-order cognitive tasks. Second, fMRI spatial maps were successfully incorporated into MEG inverse problem using MSP algorithm under a hierarchical Bayesian framework. Using both simulated data and experimental data, we provided evidence of improvements in the MEG source reconstruction by incorporating spatial priors. Finally, by using fine spatiotemporal time courses from functional active regions, we expanded our understanding of the language networks previously established from our fMRI data alone and found pathways within the language networks supporting narrative comprehension.