Abstract: A region of interest (ROI) is delineated based on the registration by using the geometric
features of skull contour, and a new construction method of relative invariants put forward under the
framework of Clifford algebra. The method proposed regards the contour data of ROI in the skull as
the point cloud for registration, and constructs the mathematical and calculation models of the relative
geometric invariants according to the similarity of skull contour. After calculating the translation and
the rotation operator required for the registration algorithm, the registration of three-dimensional data
can be preceded directly by adopting the new similarity measure of the 3D medical image. The
registration data are from the BrainWeb database and “Retrospective image registration evaluation”
project data of the University of Vanderbilt in the United States. Experiments show that our algorithm
has high efficiency in the registration of ROI in the skull. And it can calculate the 3D position of the
tissue organ more accurately. The mean error is within 2-4 mm, and the registration accuracy is up to
sub-pixel level.

Key words: medical image registration, Clifford algebra, relative invariant, region of interest