Use of independent component analysis to define regions of interest for fMRI studies

Introduction Regions of interest (ROIs) are frequently used in fMRI studies in order to limit an analysis to an anatomically or functionally defined area, or to investigate interactions between groups of voxels. Whether functional ROIs are defined on the basis of independent data, or non-independently based on contrasting a subset of conditions within the experiment, investigators face a number of arbitrary choices concerning the statistical threshold to employ and the method for delineating ROI boundaries. (Poldrack, 2006). We propose a method for defining ROIs using independent component analysis (ICA). This method avoids many of the shortcomings of general linear model (GLM) based ROI definition, and is robust and easy to implement using FSL (http://www.fmrib.ox.ac.uk/fsl). As a demonstration, we apply this method to define ROIs in the cortex and cerebellum that respond selectively to aurally paced movements of the lips, hands, and feet.