Computational Neuroimaging to Understand Dynamics of Human Brain Activity

Abstract: Over the past decade, approaches from signal processing, machine learning, and network science, have had a profound impact on the analysis and the interpretation of brain activity measured by functional magnetic resonance imaging (fMRI). Functional connectivity studies have given not only insights into how the brain supports coordinated cognition, learning, or stability in a changing environment, but also to what extent networks are altered in neurological disease and disorder. Recently, the quest for better understanding of brain dynamics has triggered new approaches to functional connectivity. In this talk, I will highlight two of our promising recent advances for fMRI analysis. First, a new sparsity-driven temporal deconvolution scheme (« total activation ») allows to study interactions between brain regions in terms of synchronized transient activity. This approach reveals a rich repertoire of functional networks and their temporal features (e.g., occurrence, duration, overlap) that can be fitted with systems-level temporal models (i.e., sparsely-coupled hidden Markov models), thus unraveling interdigitated and parallel organization. Second, graph signal processing has provided a new framework to analyze activity traces on top of a graph model—in this case, the brain’s structural backbone—and derive new relevant operations, for instance, filtering these signals according to being aligned versus liberal w.r.t. structure. Ultimately, these developments will contribute to build better, more mechanistic, models of brain function and will find application to disease diagnosis and prognosis.