Correlation in brain networks at different time scale resolution

Assessing brain connectivity makes up a major issue in the field of network dynamics and neuroscience. Conventional experimental techniques are based on functional imaging and magnetoencephalography, allowing to reconstruct the activity of relatively small brain volume elements. A common approach to identify networks consists in singling out sets of elements that maintain a correlated activity over time. Despite the general consensus that these networks are detectable on a time window of 10 s, no study is presently available on the distribution and thus the reliability of this time scale. In this work, we describe a new method to assess time scales on which correlations between network elements occur and to consequently identify the underlying network structures. The analysis relies on the evaluation of quasi-zero-delay cross-correlation between power sequences associated with distinct volume elements. By changing the width of the running window used to analyze successive segments of time series, the behavior of cross-correlation at different time scales was investigated. The onset of connectivity was estimated to be observable at about 30 s. The method was applied to a set of volume elements that are supposed to belong to a known resting-state network, namely the Default Mode Network. Fully connected networks were identified, provided that a sufficiently long time scale is considered. Our method makes up a new tool for the investigation of the temporal dynamics of networks.