Accuracy of surface fit registration for PET and MR brain images using full and incomplete brain surfaces.

Abstract

The accuracy of a surface-fitting image registration technique has been investigated for matching [18F]fluorodeoxyglucose (FDG) and [15O]H2O PET brain images with MR images. Use of partial-brain surfaces (a single hemisphere or a limited number of slices) was investigated to simulate cases in which severe brain defects or limited axial field of view would preclude using the entire brain surface. Three FDG and three H2O scans were performed on five volunteers, in addition to volume MR studies. Fiducial markers were placed on the subjects' scalps to provide references for registration accuracy. The registration procedure was applied to each PET-MR set, using the surfaces defined by locating the brain edge in multiple slices for each set. The surfaces fit well, with only 1% scaling necessary for the best fit. Errors in fiducial marker positions between MRI and transformed PET were < 2 mm in the transverse directions and < 4.5 mm in the axial direction. Fits based on the partial surfaces worked well and gave results very similar to the full-brain fits. The surface-fitting technique is accurate for FDG and H2O PET studies, even when part of the brain surface cannot be used.