Alignment of dynamic cardiac PET images for correction of motion

A technique was developed to align images from dynamic cardiac N-13 ammonia positron emission tomography (PET) scans. Inaccurate alignment of images from these scans stems from breathing, motion of the heart within the thorax, and overall motion of the patient (particularly during pharmacologic stress). The images from these scans represent changing distributions, from high blood pool concentrations to predominantly myocardial uptake, and are generally noisy. In this technique, templates are defined from a long, late frame (5-10 min). These templates include a myocardial template, which matches late frames, a blood pool template, which matches frames in which the left ventricle blood pool is dominant, and a modified blood pool template, which reduces the effect of activity in the right ventricle. The correlation function is used as the maximized parameter over shifts between each three-dimensional image frame and the appropriate template. The technique was tested on phantom, human, and animal data with myocardial defects. Phantom data showed the technique to be reliable to within one voxel (1.7/spl times/1.7/spl times/4.2 mm/sup 3/), and motion which was apparent in human data was reduced significantly. Blood flow values measured from corrected data showed two traits: higher values, due to better matching regions of interest to images, and better fits, due to smoother time-activity curves.