Principal Components Analysis Reveals the Correlation Structure of Resting-State fMRI Data

Hongjian He¹, Thomas T. Liu²

We use principal components analysis to generate low-dimensional approximations of resting-state fMRI correlation maps, where the components are ranked by their contribution to the original correlation map. Applying this approach to connectivity maps with a seed region in the posterior cingulate cortex, we find that the first ranked component map represents correlation with the global signal, while the second component shows the anti-correlated relation between the default mode network and task positive network. Our results support the general validity of global signal regression and the existence of anti-correlated resting-state networks.