A novel hidden Markov approach to studying dynamic functional connectivity states in human neuroimaging

How can we better understand the underlying neural mechanisms of brain dynamics? Many groups have examined brain states as a means of identifying recurring patterns of activity or connectivity. Although there is an abundance of methods of establishing and interpreting brain states, hidden Markov models are becoming an increasingly popular choice to extract recurring patterns of intensity or connectivity in neuroimaging data. These models not only recognize spatial patterns of brain states, but also ascertain their temporal progression. An assortment of hidden Markov model instantiations has arisen for diverse purposes, and these have been applied to a variety of neuroimaging datasets. However, most of these instantiations have focused on intensity-based states, i.e. states defined by the activity levels of one or more nodes, rather than connectivity-based states, i.e. states defined by patterns of functional connectivity between nodes. The intensity-based approach is problematic if we want to understand connectivity dynamics, since there is no reason to believe that the resultant states can provide any useful information about dynamic connectivity patterns. Here we aimed to remedy this methodological challenge by introducing a new hidden Markov model approach based on identifying states defined as full functional connectivity profiles among brain regions, and applying this approach to directly extract connectivity-based states in functional magnetic resonance imaging (fMRI) data. We then empirically explore the behavior of this new model in comparison to existing approaches based on intensity-based states and summed functional connectivity states, utilizing the widely-available HCP unrelated 100 functional magnetic resonance imaging "resting state" dataset. Results show that our newly-introduced 'full functional connectivity' model discovered connectivity states with more distinguishable patterns than those derived from previously-employed approaches, and demonstrated clear superiority in recovering simulated connectivity-based states. These findings suggest that if our goal is to extract and interpret connectivity states in neuroimaging data, our new model can reveal more insights than previous methods, and intensity-based or summed functional connectivity-based approaches miss crucial information about the evolution of functional connectivity in the brain. ### Competing Interest Statement The authors have declared no competing interest.