Most approaches to analyzing functional connectivity (FC) in resting-state fMRI characterize the interactions between brain regions as constant over time. This practice has yielded tremendous insight into functional organization, revealing the presence of resting-state “networks” that align closely with known anatomic and functional pathways [1-3]. However, FC is also known to vary across states such as sleep [4,5], as a function of caffeine intake , and over cognitive activities such as mind-wandering, learning, and focused attention [7,8]. Therefore, in addition to its remarkable consistency, the variability of resting-state FC can provide valuable information about state-dependent changes in large-scale network activity.
In recent years, resting-state fMRI studies have also moved toward examining changes in FC that occur over scales of seconds to minutes (reviewed in [9,10]). This perspective, along with novel approaches for “dynamic” analysis of resting-state data, are providing further insight into functional relationships between brain regions and their alteration with disorders such as schizophrenia (e.g. [11,12]). However, the study of dynamic FC is complicated; for instance, spaurious variability can easily arise from non-neural fluctuations (“noise”) in the fMRI signal , and as a consequence of particular analysis strategies (such as pointed out in [14,15]). Proper analysis, validation, and noise-reduction procedures are essential for obtaining an interpretable picture of FC dynamics.
We will first review sources of variability in functional connectivity as determined from experimentally induced contexts. We will then address the topic of spontaneous, time-varying changes in FC, with observations from both empirical and computational modeling work [16,17]. We will discuss studies into the neural basis of such changes , and how examining the dynamics of resting-state FC with multimodal imaging and physiological recordings can reveal neural correlates of endogenous shifts in the state of the brain/body [19,20]. We will show examples of recent literature proposing methods for analyzing dynamic FC, and the potential for temporal features of FC to serve as biomarkers for clinical applications. Finally, we will discuss the importance of reducing non-neural sources of variability, and illustrate how appropriate modeling and analysis techniques are critical when drawing inferences about FC dynamics.