Transient Spectral Peak Analysis Reveals Distinct Temporal Activation Profiles for Different Functional Brain Networks

Employing functional magnetic resonance imaging (fMRI) data from a large schizophrenia study, we use wavelets and frequency-specific spectral thresholding to transform each of the functional network timecourses (TCs) produced by group independent component analysis (GICA) into sparse multivariate spectral timeseries that consist only of zeros except at time-frequency points where power exceeds the 95^th percentile for that frequency, i.e. masked to retain only frequency-specific extremal spectral events. From this characterization of network timeseries in terms of transient spectral peaks, we identify new distinctions between the temporal behavior of different functional networks and domains, new anomalies of meso-scale brain activation in schizophrenia patients and relationships between transient muti-network spectral extrema and time-varying network connectivity that have not been reported previously.