Selection of spatially independent components to explain functional connectivity in fMRI

In functional magnetic resonance imaging (fMRI), functional connectivity of brain regions is defined as the temporal correlation of their average time courses. A key question is to determine which processes contribute to functional connectivity. Independent component analysis (ICA) is a recent data-driven method that has proven efficient to identify activation and a number of artefacts. We propose a flexible model to explain the functional connectivity in a network of brain regions. The method we propose is based on matching pursuit to select a small set of independent components calculated by ICA that explains most correlations in a given network. On a real dataset, we show that the number of components is small enough to allow for a systematic qualitative interpretation of the selected components. Our results suggest that functional connectivity is not only due to the activation signal and artefacts, but also to other components, sharing similarity with resting-state signal.