Dynamic fMRI Connectivity Tensor Decomposition: A New Approach to Analyze and Interpret Dynamic Brain Connectivity.

As brain network organization likely fluctuates over time to react to internal and external stimuli, the validity of conventional static brain connectivity models are being questioned. Thus, there is a growing interest in using so-called dynamic network analyses. Brain network analyses yield complex network data that is difficult to analyze and interpret. To deal with the complex structures, data reduction techniques that simplify the data are often used. For dynamic network analyses, data simplification is even of greater importance, as dynamic connectivity analyses result in a time series of complex networks. Decomposition/factorization methods that identifying the main components underlying the data are gaining popularity for dynamic brain network simplification. A new challenge that must be faced when using these data reduction techniques is how to interpret the resulting network components. Thus, the primary goal of this paper is to specifically address this challenge and discuss issues that must be considered when interpreting the network components. Based on simulated and real fMRI data analysis, we argue that the network components cannot be interpreted in the same manner as the original functional networks, e.g. the connections in the components represent complex relationships between nodes not simple associations between the time series. We also demonstrate the associated weight that varies across time and across participants in a study population that accompanies the network components. The network components should always be interpreted in conjunction with these weights that denote how any one component contributes to overall brain network connectivity at given time or in any given participant.