Modeling functional network dynamics via multi-scale dictionary learning and network continuums

Although there is growing recognition and interest in the dynamic organization of the brain's functional architecture, the characterization and systematic analysis of such dynamic patterns are still largely under-investigated. In this work, we proposed a novel multi-scale dynamic dictionary learning framework which decomposes the input fMRI signals into functional networks via sliding/growing time window network analysis approach based on a fast dictionary learning method. As the functional networks are obtained at every possible temporal position and multiple scales within the framework, links across functional networks are then defined by the direct or indirect correlations between their temporal variation patterns. The identified links between networks are then concatenated into network continuums, which are the dynamic combinations of functional networks at various temporal segments. By applying the proposed framework on the task fMRI data from the Human Connectome Project (HCP) Q1 release, we have found that the network continuum has the capability of capturing the dynamic transition patterns of functional networks and providing an intuitive characterization of the network stationarity across time scales.