Interactions Between Large-Scale Functional Brain Networks Are Captured by Sparse Coupled HMMs
T. Bolton, A. Tarun, V. Sterpenich, S. Schwartz, D. Van De Ville
IEEE Trans Med Imaging
Functional magnetic resonance imaging reveals two complementary facets of brain activation: how it distributes in space, and how it evolves over time. In the resting-state (i.e., in the absence of any external task or stimulus), disentangling the sets of brain regions that co-activate as networks, and characterizing the dynamics of the interactions between those networks, make two promising directions in terms of neuroscientific and clinical potential.
In this work, we considered a set of 13 previously extracted innovation-driven co-activation patterns (iCAPs) as our functional networks of interest (see top left part of the illustration), and introduced a novel framework able to model the temporal dynamics of those iCAPs, as well as possible cross-network interactions. The core of our modeling strategy consisted in describing each iCAP's activity profile as a Markov chain taking one of three states at each time point: deactive, baseline, or active (respectively blue, gray or red circles in the illustration middle part). In addition, we enabled (de)active iCAPs at time t to exert a modulatory impact on the dynamics of the other networks from time t to t+1 (blue and red modulatory arrows). To keep the problem computationally affordable and functionally meaningful, we restrained the set of existing cross-network interactions in data-driven manner through L1-regularisation.
On a battery of simulated examples, our approach outperformed simpler correlational tools and uncoupled HMM descriptions. On our set of 13 iCAPs, we could unravel a broad panel of significant interactions across iCAPs (bottom right part of the illustration), offering a directional representation of functional brain network interplays.