A novel approach for global noise reduction in resting-state fMRI: APPLECOR

Noise in fMRI recordings creates uncertainty when mapping functional networks in the brain. Non-neural physiological processes can introduce correlated noise across much of the brain, altering the apparent strength and extent of intrinsic networks. In this work, a new data-driven noise correction, termed "APPLECOR" (for Affine Parameterization of Physiological Large-scale Error Correction), is introduced. APPLECOR models spatially-common physiological noise as the linear combination of an additive term and a mean-dependent multiplicative term, and then estimates and removes these components. APPLECOR is shown to achieve greater consistency of the default mode network across time and across subjects than was achieved using global mean regression, respiratory volume and heart rate correction (RVHRCOR (Chang et al., 2009)), or no correction. Combining APPLECOR with RVHRCOR regressors attained greater consistency than either correction alone. Use of the proposed noise-reduction approach may help to better identify and delineate the structure of resting state networks.