Quantification of Resting-State fMRI Networks Driven by Hemodynamically Informed Spatiotemporal Regularization

The brain's spontaneous fluctuations measured by functional magnetic resonance imaging during rest cluster into recurrent activity patterns known as resting-state networks (RSNs). The spatial organization of RSNs in health and disease has been immensely investigated by conventional correlational analyses of fMRI time series. Recent findings of time-resolved analyses have provided evidence of reoccuring activation patterns that are accessible at instantaneous time points enabling the dynamic characterization of RSNs. We have proposed a method to recover spatially and temporally overlapping RSNs, which we named innovation-driven co-activation patterns (iCAPs), to study the dynamic engagement of RSNs unconstrained by the slow hemodynamic response. The iCAPs are extracted by temporal clustering of sparse innovation signals recovered from Total Activation (TA) framework, which is cast as a variational problem with sparsity-promoting spatial and temporal priors for fMRI data deconvolution. The temporal prior uses the inverse of the hemodynamic response function as a general differential operator and exploits sparsity of the innovation signals. In this work, we perform a quantitative analysis to assess the stability of iCAPs recovered from a group of patients with mood disorders and healthy volunteers.