Decomposing dynamic functional connectivity onto phase-dependent eigenconnectivities using the Hilbert transform

Dynamic functional connectivity (dFC) based on resting-state functional magnetic resonance imaging (fMRI) is a new avenue to explore brain network dynamics. Considering the time-varying evolution of the connectivity between spatially-defined regions increases drastically the dimensionality of the data with respect to static FC. Two of the more common approaches explored so far to analyze the large amount of data and to identify connectivity states are k-means clustering of the connections´ timecourses, revealing the average patterns of connections mainly recurring, and multivariate statistical methods like principal component analysis (PCA), identifying consistent connectivity patterns across time and population, so-called eigenconnectivities. In this work, we propose for the first time to explore the frequency content of dFC timecourses, with two objectives: 1) to decrease the computational load of the analysis by reducing the temporal redundancy intrinsically present in the data due to the sliding-window estimation; 2) to gain additional information about eigenconnectivities by using the Hilbert transform to explore phase information, that has been ignored until now. Results for resting-state fMRI data of 20 healthy subjects showed perfect consistency between the reduced and the original data, proving that we can safely ignore frequencies above the fundamental frequency of the window without discarding useful information. Furthermore, we highlighted interesting and clear patterns of connections with specific quadrature phase relationships in the new eigenconnectivities, which is promising to study interactions between functional networks.