The detection and quantification of co-activation patterns is a longstanding problem in neuroscience. We have extended a method (Lopes-dos-Santos, Ribeiro, and Tort 2013) to detect co-activation in fMRI signals, allowing both spatial and temporal characterization of interregional interactions at the whole-brain level. Combined with whole-brain models, this method allows us to analyze the functional richness of arbitrary timescales in the human brain. We use this framework to detect the timescale of maximum information transmission in the human brain and to determine this frequency’s relationship with consciousness state.
We utilize the z-score point process (Tagliazucchi et al. 2016) to convert the instantaneous regional activations in the BOLD signal into an event matrix. We then e extract principal components of the event covariance matrix and define an assembly space spanned by the significant components (Lopes-dos-Santos, Ribeiro, and Tort 2013). These significant components define the coactivation patterns, and significance is determined using the Marčenko-Pastur distribution as a lower bound (Marčenko and Pastur 1967). Projecting the covariance matrix onto this assembly space and extracting independent components shows regional membership weights in each assembly, and projecting the event matrix onto the resulting assembly vectors gives activation timings. We thus extract precise functional spatiotemporal connectivity patterns.
This method can be run on both empirical and simulated data. Simulating data allows us to explore the spatiotemporal network dynamics at arbitrary timescales. We sweep exhaustively over a range of global coupling strength G to find the value which produces the lowest Kolmogorov-Smirnov distance between simulated and empirical functional connectivity dynamics. The model is then run with this optimal value in order to generate simulated neural data, which may be sampled to detect the spatiotemporal assembly dynamics over a wide range of timescales. The dynamical richness, as quantified by the Shannon entropy, can then be examined as a function of both timescale and consciousness state.
We have detected significant differences in the dynamical richness of N2 and N3 sleep. We are currently running the analysis using different measures for interregional communication, namely the cosine coherence (Deco and Kringelbach 2016), in the belief that this will better capture the differences between functional states.