Head motion and its correction

Abstract

Abstract The aim of fMRI data analysis is, in simple terms, to identify small, spatially localized changes in image intensity that accompany the performance of some experimental task. This is normally accomplished by collecting a series of images covering part or all of the brain at intervals of a few seconds and analysing the resulting time-series obtained at each voxel. The experimentally-associated intensity changes are frequently only a few per cent of the mean value at that spatial location and are embedded in a noisy signal contaminated by electronic and physiological effects. Head motion is one further and critically important confound. Very small movements of the head, on a scale of less than a millimetre, can be a major source of error in fMRI analysis if not identified and treated correctly-a problem drawn to the attention of workers in this area in 1994 by Hajnal et al. This problem can be particularly acute in areas lying close to a high-contrast boundary in the brain image, where movement-related changes in image intensity may be large. In analysing the image intensity at each voxel over a time series of brain volumes, we make the assumption that we are sampling an identical region of the brain at every point in an experiment.