Identifying nonlinear Functional Connectivity with EEG/MEG using Nonlinear Time-Lagged Multidimensional Pattern Connectivity (nTL-MDPC)

Investigating task- and stimulus-dependent connectivity is key to understanding how brain regions interact to perform complex cognitive processes. Most existing connectivity analysis methods reduce activity within brain regions to unidimensional measures, resulting in a loss of information. While recent studies have introduced new functional connectivity methods that exploit multidimensional information, i.e., pattern-to-pattern relationships across regions, they have so far mostly been applied to fMRI data and therefore lack temporal information. We recently developed Time-Lagged Multidimensional Pattern Connectivity for EEG/MEG data, which detects linear dependencies between patterns for pairs of brain regions and latencies in event-related experimental designs (Rahimi et al., 2022b). Due to the linearity of this method, it may miss important nonlinear relationships between activity patterns. Thus, we here introduce nonlinear Time-Lagged Multidimensional Pattern Connectivity (nTL-MDPC) as a novel bivariate functional connectivity metric for event-related EEG/MEG applications. nTL-MDPC describes how well patterns in ROI X at time point tx can predict patterns of ROI Y at time point ty using artificial neural networks (ANNs). We evaluated this method on simulated data as well as on an existing EEG/MEG dataset of semantic word processing, and compared it to its linear counterpart (TL-MDPC). We found that nTL-MDPC indeed detected nonlinear relationships more reliably than TL-MDPC in simulations with moderate to high numbers of trials. However, in real brain data the differences were subtle, with identification of some connections over greater time lags but no change in the connections identified. The simulations and EEG/MEG results demonstrate that differences between the two methods are not dramatic, i.e. the linear method can approximate linear and nonlinear dependencies well. HighlightsO_LInTL-MDPC is a bivariate functional connectivity method for event-related EEG/MEG C_LIO_LInTL-MDPC detects linear and nonlinear connectivity at zero and non-zero lags C_LIO_LInTL-MDPC revealed connectivity between ATL hub and semantic control regions C_LIO_LIDifferences between linear and nonlinear TL-MDPC were small C_LI