NEW APPROACHES FOR ASSESSING TIME-VARYING FUNCTIONAL BRAIN CONNECTIVITY USING FMRI DATA

It was long assumed that functional connectivity (FC) among brain regions did not vary substantially during a single resting-state functional magnetic resonance imaging (rs-fMRI) run. However, an increasing number of studies have reported on the existence of time-varying functional connectivity (TVC) in rs-fMRI data taking place in a considerably shorter time window than previously thought (i.e., on the order of seconds and minutes). However, the study of TVFC is a relatively new research area and there remain a number of unaddressed problems hindering its ability to fulfill its promise of increasing our knowledge of human brain function. First, while it has previously been shown that autocorrelation can negatively impact estimates of static functional connectivity, its impact on TVC estimates has not been established. Understanding the influence of autocorrelation on TVFC is of high importance, as we hypothesize the autocorrelation within a time series can inflate the sampling variability of TVC estimated using sliding window techniques, leading to the increase of risk of misinterpreting noise as true TVC and negatively impact subsequent estimation of whole-brain time varying functional connectivity. We thus study the impact of autocorrelation on TVC and how to mitigate it. Second, there is a need for new analytic approaches for estimating TVC. Most studies use a sliding window approach, where the correlation between region is computed locally within a specific time window that is moved across time. A shortcoming of this approach is the need to select an a priori window length for analysis. To circumvent this issue, we focus on the use of instantaneous phase synchronization (IPS), which offers single time-point resolution of time-resolved fMRI connectivity. The use of IPS necessitates bandpass filtering the data to obtain valid results. We seek to show how bandpass filtering affects the estimates of IPS metrics such as phase locking value (PLV) and phase coherence. Further, as current metrics discard the temporal transitions from positive to negative associations common in IPS analysis we introduce a new approach within IPS framework for circumventing this issue. Third, the choice of cut-off frequencies when bandpass filtering in IPS analysis is to some extend arbitrary. We seek to compare standard phase synchronization using the Hilbert transform with empirical mode decomposition (EMD) which eliminates the need for bandpass filtering in a data driven manner. While the use of EMD has a number of benefits compared to the Hilbert transform, it has a couple shortcomings: the susceptibility of the EMD to the SNR of the signal and untangling frequencies close to one another. To circumvent this issue and improve the assessment of IPS, we propose the use of an alternative decomposition approach, multivariate variational mode decomposition (MVMD) for phase synchronization analysis.