Identifying task-related dynamic electrophysiological brain connectivity

How does human cognition emerge from neural dynamics? A proposed hypothesis states that efficient neuronal communication between brain regions through oscillatory synchronization gives the basis for cognitive processing. These synchronized oscillatory networks are transiently forming and dissolving at the timescale of milliseconds to support specific cognitive functions. However, unlike resting-state networks, there is still no appropriate technique for characterizing the complicated organization of such cognitive networks during task performance, especially naturalistic tasks (e.g., music listening). In this thesis, we exploit the high spatiotemporal resolution of electro- or magnetoencephalography (EEG/MEG) to match the rapid timescales of synchronized neural populations and develop EEG/MEG analysis tools to probe the reconfiguration of electrophysiology brain networks during cognitive task performance. In the first study, we applied CANDECOMP/PARAFAC (CP) tensor decomposition to single-trial wavelet-transformed representations of sourcelevel EEG data recorded in a simplified and controlled task, to extract the stimuliinduced oscillatory brain activity. In the second study, by combining spatial Fourier independent component analysis with acoustic feature extraction, we probed the spatial-spectral signatures of brain patterns during continuously listening to natural music. In the third study, we examined the connectivity dynamics during natural speech comprehension via performing principal component analysis on envelope-based connectivity measurements concatenated across time or subjects. In the fourth study, we introduced a novel approach based on CP decomposition to investigate the task-related functional networks with a distinct spectrum during self-peace movement and working memory tasks. Then, we extended this tensor-based method of analyzing network dynamics during natural music listening in the fifth study. In conclusion, these studies introduce novel approaches based on matrix or tensor decomposition to allow for multi-way connectivity analysis considering its non-stationarity, frequency-specificity, and spatial topography. Keywords: naturalistic stimuli, brain networks, functional connectivity, dynamics, frequency-specificity, tensor decomposition