Fig 1.
Outline of our framework for capturing instantaneous spectra of ICA time-courses and its variation in time in the form of “frequency modes”.
(A) First, fMRI time-series is pre-processed and feed into the ICA to be decomposed into 50 ICA networks and the associated time-courses (detail of these ICA networks is provided in supplementary material of (Allen, 2014 #510)). Complex morlet wavelet is used to map the time-courses to the time-frequency domain. Finally, canonical patterns of power spectra are estimated by k-means clustering which we refer to as “frequency modes”. (B) "Frequency modes" as the representatives of the variation in spectral powers of networks time-courses, Each mode is formed by similar instantaneous frequency content of time-courses which have been clustered together.
Fig 2.
Analysis of occurrence the modes.
Boxplots of occurrence rates of each individual frequency mode in ICA networks. Networks with significantly higher (filled boxplots) or lower (dashed boxplots) occurrence of the given mode than majority (85%) of all networks are identified.
Fig 3.
Analysis of co-occurrence rates of the modes.
Cco-occurrence maps of frequency mode pairs. An entry (column m, row n)(1…50, 1…50) of a matrix at column i (1…4) and row j (1…4) of the figure shows cco-occurrence of frequency mode i in network m, given that frequency mode j is occurred at the same time-point in network n. Positive cc-occurrence (color coded as red) corresponds to reinforcement effect and negative cc-occurrence (color coded as blue) is corresponding to suppression effect.
Fig 4.
Analysis of age and gender effect on.
(A) occurrence rate of individual frequency modes and (B) cco-occurrence rate of pair of modes. In (A) specific networks and in (B) pairs of networks are highlighted [13] in which occurrence rate of given mode and cco-occurrence of pair of modes are significantly effected by age or gender