Multi-modal rigid and non-rigid registration for attenuation correction in cardiac SPECT/CT using emission scatter to CT conversion

Patient body or respiratory motion between emission imaging and CT can cause misalignment of position of heart in the emission and attenuation maps, causing significant artifacts in emission reconstruction upon attenuation correction. In this work, we investigate multi-modal registration of cardiac SPECT and CT datasets, and then use the registered CT maps for attenuation correction (AC). The challenges for multi-modal registration are many: missing structural data in SPECT, nonrigid motion between organs, noise etc. Our hypothesis is that some of the missing structural information in the photo-peak window slices is available if we use the scatter window reconstructed data because of the tissue density dependence of scatter at 140 keV. We investigate this by converting the emission scatter-window slices into "CT like"-images and then using rigid and non-rigid registration methods to align them to actual CT. The non-rigid registration method used is the diffeomorphic demons algorithm with sum-squared difference as the metric. To evaluate the method, MRI-derived volunteer-specific XCAT datasets were obtained for volunteers undergoing large motions between successive MRI imaging scans. These were used to create Monte Carlo simulated SPECT and CT studies. The motions investigated were a bend, a twist, and an axial slide between emission and CT. AC using the matched CT was employed to gauge the success of registration. We observed that using registered attenuation maps eliminated or reduced the AC artifacts due the motions, where present. More motion cases for a larger population of volunteers and polar map quantification of cardiac update need to be analyzed before conclusions can be drawn from these studies.