Multivariate Mutual Information Measures Functional Connectivity Accurately

Abstract

Most studies use linear correlation as an estimator of functional connectivity. This measure does not detect the nonlinear dependence between two variables. During resting state, there are nonlinear relations among time series discarded by common functional connectivity measures such as Pearson correlation. Another limitation of linear correlation is the inability of calculating the association between two multivariate variables. Typically, a dimension reduction such as averaging is applied to each region time series. This reduction leads to a loss of spatial information across voxels within the region. Here, we propose to use a new information-theoretic measure as an interaction estimator between brain regions. Using simulated data, we show that this measure, multivariate mutual information (MVMI), overcomes the above mentioned limitations.