Network neuroscience models the human brain as a collection of structural and functional networks. Brain network studies have broadened our understanding on how the brain works, revealing hubs of information transfer and a modular structure that reflects the balance between functional segregation and integration. Brain network structure depends on age, cognitive task, and health and disease. The most recent research paradigms investigate how brain networks change in time and use multilayer networks to simultaneously capture brain structure and function. However, a number of methodological questions related to network neuroscience remains unanswered. Importantly, the neuroscientific community lacks a standard definition of network nodes, although the properties of networks strongly depend on what their nodes represent. Similarly, the effects of preprocessing methods on brain network structure are not fully known. Regions of Interest (ROIs) are commonly used as nodes of functional brain networks. ROIs are brain areas that contain many measurement voxels in the case of functional magnetic resonance imaging (fMRI) or source vertices in the case of electro- and magnetoencephalography (EEG, MEG). ROIs are defined using anatomy, function, or connectivity. The time series of ROIs are typically obtained by averaging the time series of voxels or vertices of each ROI. The ROI approach is based on the assumption of functional homogeneity: all voxels or vertices of the ROI are assumed to behave similarly. In my Thesis, I explore methodological issues related to brain networks, in particular node definition, functional homogeneity, and preprocessing effects. In the first study, I show that signal mixing and inaccuracies of inverse modelling lead to low levels of functional homogeneity of ROIs in EEG and MEG studies. I introduce an optimized inverse collapse weighting operator as a possible solution. In the second study, I demonstrate that ROIs in three commonly-used fMRI parcellations display low levels of functional homogeneity. Therefore, they should not be used as nodes of functional brain networks without further consideration. In the third study, I investigate the nontrivial effects that a commonly used preprocessing method, spatial smoothing, has on the structure of fMRI brain networks. Finally, in the fourth study, I investigate time-dependent changes in functional homogeneity and local structure of fMRI networks. This leads me to ask if any set of static ROIs can be used to describe brain networks because they are dynamic. My results highlight possible pitfalls of some methods presently used in network neuroscience. Therefore, further careful work is needed to ensure that the methodological basis required for working in the frontier of network neuroscience is indeed solid.
|Publication status||Published - 2018|
|MoE publication type||G5 Doctoral dissertation (article)|
- network neuroscience, functional brain networks, brain parcellation, Region of Interest, functional homogeneity, dynamic brain networks, preprocessing