The evaluation of SPECTMR registration error from the internal landmark matching technique

In using the internal landmark matching technique for registering brain volumes from single photon emission computed tomography (SPECT) and magnetic resonance (MR) imaging, the resulting errors are different from similar registrations of positron emission tomography (PET) and MR volumes because of the anisotropic nature of SPECT's tomographic resolution. SPECT/MR registration errors are investigated with point simulation to allow a controlled study of the dependence of the translation and rotation errors on different factors. These factors include: the number of point pairs, m, the error in point pair homology (characterised by the FWHM of a 3-D Gaussian error envelope), and the configuration of the point sets (as they depend on their radius from their centroid). The results of this study are used to interpret the results from studies based on a 3-D brain phantom fitted with external fiducials to provide the basis for true registrations.It is found that the same error distributions may not be expected across ordinates. Based on the imaging capabilities of the sensors used in this study, (a) maximum translation errors or $ pm$2.1 mm in the anterior-posterior orientation are found to be a consequence of SPECT's anisotropic resolution and the brain's major axis in that orientation, and (b) rotation errors of about $ pm$2.1$ sp circ$ in the coronal plane are larger than in the transaxial because of the comparative dimensions in those planes