Principal axes and surface fitting methods for three-dimensional image registration

We evaluated the effect of the image acquisition parameters on the accuracy of the principal axes and surface-fitting techniques for three-dimensional image registration. Using two types of phantom objects, MR brain image and a mathematically defined ellipsoid, we simulated pairs of scans with known acquisition parameters, including longitudinal coverage, magnitude of mis-registration, number of sections and section thickness. Both methods are sensitive to the systematic deformation of contours. The principal axes method is also sensitive to incomplete scan coverage and to the x-axis and y-axis misangulation. Both methods are insensitive to the number of sections, section thickness and the number of points per section. Surface fitting performed well without user supervision. There is no need for routine inclusion of the scaling factors as search parameters. The results confirm the feasibility of three-dimensional multimodality registration of brain scans with accuracy 1-2 mm, with surface fitting being the method of choice.