Ciulla C, Deek F P
New Jersey Institute of Technology, Department of Computer and Information Science, Newark, New Jersey 07102, USA.
Brain Topogr. 2001 Fall;14(1):41-56. doi: 10.1023/a:1012515822675.
An automatic technique for the registration of fMRI time series has been developed, implemented and tested. The method assumes the human brain to be a rigid body and computes a head coordinate system on the basis of three reference points that lies on the directions corresponding to two of the principal axes of the volume at the intersections with the head boundary. Such directions are found computing the eigenvectors of the symmetric inertia matrix of the image. The inertia components were extracted weighting pixels' coordinates with their intensity values. The three reference points were found in the same position, relative to the head, in both the test and the reference images. The technique has been tested using T2*-weighted Magnetic Resonance (MR) images in which known rigid body transformations have been applied. The results obtained indicate that the method offers subvoxel accuracy in correcting misalignment among time points in fMRI time series.
