Quantitative outcome of registration methods for correcting cardiac drift in cardiac PET/CT imaging

Myocardial perfusion studies suffer from artifacts caused by misalignment of the transmission and emission data due to the influences of voluntary and involuntary patient motion. Regardless of 68Ge or respiratory-averaged CT based attenuation correction and good patient cooperation, approximately 21% of perfusion studies exhibit artifacts arising from misalignment that cannot be corrected by manipulating the attenuation acquisition protocol. This misalignment, termed cardiac drift, is caused by slow-moving abdominal cavity contents that reposition the heart in the thorax and appear as myocardial uptake overlying the left CT lung in fused PET/CT images. This study evaluates three postimaging registration techniques to correct PET/CT misalignment by altering the transmission map to match myo-cardial uptake. Simulated misalignment studies were performed with a cardiac torso phantom filled with [18F]FDG at 10:1 myocardium/background. An air-filled saline bag affixed to the medial left lung surface served as a distensible lung. An initial CT acquisition was followed by successive PET acquisitions consisting of small displacements of the cardiac insert into the left lung. Phantom transmission scans were aligned to the myocardial uptake in the emission scans by applying 1) full rigid-body translations and rotations, 2) rigid-body restricted to medial / lateral and superior / inferior translation, or 3) an emission-driven method that adds myocardial tissue to the transmission scan. These methods were also applied to 10 low-likelihood coronary artery disease (CAD) patients showing signs of cardiac drift. Full rigid-body registration showed significant over-correction (p < 0.004) of activity concentrations in the artifact areas of the phantom data due the relocation of highly attenuating structures (i.e., spine). Inaccurate regional activity distributions were also observed as streaks extending from the spine and these results were replicated in the patient population. There was no significant difference between the true phantom activity concentration after correction with the emission-driven method. Misalignment corrected with the rigid-body registration results in an increase in activity concentration but fails to accurately recover the true concentration. These data suggest that a nonlinear image registration approach such as an emission-driven method results in a more uniform activity distribution throughout the myocardium, and is more appropriate for addressing the cardiac drift misalignment problem.