Utility of Partial Correlation for Characterising Brain Dynamics: MVPA-based Assessment of Regularisation and Network Selection

Abstract

Correlation and partial correlation are often used to provide a characterisation of the network properties of the human brain, based on functional brain imaging data. However, for partial correlation, the choice of network nodes (brain regions) and regularisation parameters is crucial and not yet well explored. Here we assess a number of approaches by calculating how each approach performs when used to discriminate different ongoing states of brain activity. We find evidence that partial correlation matrices, when estimated with appropriate regularisation, can provide a useful characterisation of brain functional connectivity.