Neuro-X Seminar: Connectivity informed interpretation of brain activity with graph signal processing and spectral residual networks

Network neuroscience is the application of graph theory to model the complex structure and function of the brain. While network neuroscience enables many new insights on brain organisation, still relatively few methods exploit brain connectivity in the analysis of brain activity. Recent attempts in this direction have leveraged on the one hand graph spectral analysis (to de compose brain connectivity into eigenmodes or gradients) and graph signal processing (to decompose brain activity “coupled to” an underlying network in graph Fourier modes). In this talk, I will describe two ongoing works that attempt at integrating knowledge from brain connectivity in order to decode and interpret brain activity. In the first contribution, we use functional connectivity graphs to define spectral convolution operators in a deep residual network trained on task decoding. We show how paramete r pruning can be used to select the most important connectivity gradients for the task. In the second study, we analyze brain activity measured using high density EEG, and perform an analysis using graph signal processing to estimate coupling and decouplin g of source localized electrophysiological activity on a structural connectivity graph. We discuss the similarity between structure function coupling during resting state and video watching at the individual level. The overarching goal of this line of work is to explore whether connectivity informed analysis of brain activity can contribute to a better understanding of brain complexity as multimodal signals over networks.