Motion-based, multi-modality image registration for cardiac imaging

A novel image registration method is presented for multi-modality, gated cardiac imaging. The motion of the myocardium is registered instead of attributes obtained from image intensities, which may be drastically different. Optical flow methods are used to estimate a set of 3D vector fields for both modalities. The 3D vector fields are assumed to be similar and are rigidly aligned by minimizing a sum-of-squares error objective function. Evaluation of the motion-based (MB) method was performed using simulated cardiac SPECT and CT images of a 4D thorax phantom for registration errors of 1 to 3 cm translation, with and without rotation. The MB method was compared to a mutual information (MI) based method. The MB method was able to register the images with an accuracy of 1-5 mm for an anatomical point in the left ventricle. The MI method required a common background distribution within the two modalities for accurate registration.