Voxel-Wise Regressor Selection for Physiological Noise Correction with RETROICOR

Mark Jenkinson¹, Rob Tijssen¹, Jonathan Brooks¹, Karla Miller¹

A method is proposed that uses the Bayesian Information Criterion to adaptively select the most suitable set of RETROICOR regressors at each voxel separately. This is shown to dramatically reduce the number of regressors needed per voxel, with advantages in reducing degrees of freedom loss and minimising interaction between stimulus-related regressors and physiological noise regressors. Tests are performed with resting FMRI data with typical TR (3s) and low TR (0.15s).