Automatic self-alignment and registration for PET/CT reconstruction by a cross-correlation maximization method

The images obtained from a PET/CT system need to be registered and fused together to provide the doctor a better view of the tumor's anatomical locations. Ideally, the PET and CT are aligned in the manufacturing/setup stages. However, patient loading, bed deflection and subject movement can cause misalignment in fusion registration. The misalignment also introduces scattering and attenuation correction errors to the PET images resulting in quantitative errors and artifacts. In this study, we propose an automatic self-alignment method for PET/CT fusion registration by deriving the PET and CT alignment-transformation matrix from the real object/patient data based on cross-correlation maximization principle. The new method can reduce the requirement for the mechanical precision between the PET and CT; eliminate the use of specially designed calibration phantoms; the patient-table deflection effect is already included in the self-alignment process thus will be corrected automatically. An iterative procedure based on 2D cross-correlation maximization on the three projections is developed. Several modification techniques for the projections are studied to improve the accuracy and convergence of the iterations. Real mouse images from the MuPET/CT system are tested and good alignment results are achieved.