A comparison study of nonlinear and linear metrics in probing intrinsic brain networks from EEG data

Functional intrinsic brain networks (IBNs) has been widely studied due to its close relationship to different brain functions and diseases. In these studies, linear metrics, e.g., correlation, have been commonly used in identifying brain networks, especially on functional magnetic resonance imaging (fMRI) data. However, nonlinear mechanism is believed to exist in forming brain networks. In the present study, we investigated the performance of a nonlinear metric, i.e., phase coherence, in probing brain networks, as compared with a linear metric, i.e., power correlation. Specifically, individual IBNs were firstly obtained by a time-frequency independent component analysis (tfICA), and then the interaction among them were probed using either phase coherence (inter-component phase coherence, ICPC) or power correlation coefficient (PCC). We examined them using high-density resting-state electroencephalography (EEG) data from a group of patients with a balance disorder who received repetitive transcranial magnetic stimulation (rTMS) treatments. The results indicated that the use of ICPC indicated more detections of significant connectivity crossing multiple brain regions in various frequency bands than PCC. Moreover, consistent treatment-related network changes, as compared with previous neuroimaging findings, in this brain disorder were more successfully detected with ICPC. Therefore, it is important to use nonlinear metric in characterizing interactions between different brain regions and IBNs.