The Role of Registration and Spatial Normalisation in Detecting Activations in Functional Imaging

onsider the head as a rigid object. In three dimensions, six parameters are needed to define a rigid body transformation (three translations and three rotations). The first step in the correction is image registration, which involves determining the parameter values that optimize some criteria for matching each image with a reference image. In order to be effective, the accuracy of the registration needs to be within a fraction of a voxel. Following the registration, the images are transformed by resampling according to the determined parameters, generally using a form of sinc interpolation. In the purest form of sinc interpolation, the values of every voxel in the image are needed to resample a single voxel. This is not feasible in practice, so a windowed sinc function (using only a few nearest neighbours) is generally used as an approximation. Motion correction is especially important for paradigms where the subject may move in the scanner in a way that is correlated to the