Measuring the Consistency of Global Functional Connectivity Using Kernel Regression Methods

This paper describes a novel approach to estimate the consistency of global functional connectivity. We apply kernel regression methods, kernel ridge regression (KRR) and support vector regression (SVR), to predict the time-series from a target voxel using voxels in the rest of the brain as features. A correlation coefficient, obtained by cross-validation, was used to define the consistency of global functional connectivity of each target voxel. This procedure was applied to all the voxels in the brain, and a map of correlation coefficients, which measures the accuracy of predictions, over the whole brain was generated. The method was applied to two separate 10 min resting runs of four subjects. The most accurately predicted regions were mostly in the grey matter. This efficient method can detect regions with low global connectivity and also allows visualization of changes in functional connectivity between tasks.