We have investigated 2D and 3D registration techniques based on either
simple rotational/translational movements or movements modeled as a f
irst or second order polynomial to correct for assumed rigid and nonri
gid body motion in fMRI. Data acquired with two 1.5T multi-slice echo
planar imaging (EPI) MRI systems were used to develop and evaluate the
registration techniques. Two sets of data were acquired with EPI. In
the first set, the subjects were asked to keep their head as stationar
y as possible while in the second set, subjects were asked to randomly
move their herd with displacements being on the order of 1cm. Images
from the second set were registered and compared to those obtained Fro
m the first set. Results indicate that 2D and 3D motion corrected imag
es show lower variance to mean ratio than uncorrected images and that
in most cases, polynomial based nonrigid body correction produces the
lowest variance to mean ratios.