Respiratory motion corrected dynamic cardiac PET imaging

Objectives Quantification of myocardial blood flow (MBF) with dynamic cardiac PET imaging can assist in the clinical diagnosis of coronary artery disease. However, respiratory motion causes blurry artifacts in the reconstructed images, affecting analysis of the tracer kinetics and hence estimation of the MBF. The goal of this study is to introduce respiratory motion correction in dynamic cardiac PET image reconstruction and to evaluate its effect on the estimation of the K1 values compared to the conventional reconstruction method with no motion correction.

Methods The respiratory motion correction in dynamic cardiac PET imaging starts from estimating motion vector fields that match the target respiratory gates to the reference end-expiration phase. We summed up the data in each respiratory gates regardless of the dynamic framing and performed reconstruction of the respiratory-gated images for the motion estimation. With the estimated motion vector fields, we applied the 4D motion compensated image reconstruction algorithm to each dynamic frame reaching at the motion-corrected image sequence. To evaluate the effect of the proposed method on dynamic image frame reconstruction and kinetic parameter estimation, we simulated a dynamic Rb-82 cardiac PET imaging dataset of 2 minutes acquisition with ongoing respiratory motion using the XCAT phantom. A set of organ time activity curves representing the typical bio-distribution of Rb-82 in a normal patient was extracted from clinical measurement. The fitted parametric images together with the plasma input function were used as the ground truth for the simulation. We simulated data at the clinical count level with 10 noise realizations, carrying 5 respiratory gates each of 1 sec and 8 of 15 sec dynamic frames, for the geometry of the GE discovery RX PET/CT scanner. Using the one-tissue compartment model, we estimated the K1 values on the left ventricle polar map from the dynamic image sequences reconstructed with and without the proposed respiratory motion correction. To quantitatively evaluate the reconstructed dynamic image frames, the regional mean and standard deviation as well as the tradeoff between noise and bias were measured on the left ventricular polar map and its segments. To evaluate the estimated kinetic parameters, the regional mean and the standard deviation of the K1 values were calculated on the left ventricle parametric polar maps from all the noise realizations.

Results The blurry artifacts in the reconstructed dynamic images were visually reduced with the proposed respiratory motion correction technique. The noise versus bias tradeoff of each frame was clearly improved by the motion compensation. With less variation on and among the segments, the polar maps of the dynamic frames from the motion correction method was more uniform than those without motion correction. The estimated K1 values from the ten segments on the polar map with and without motion correction were 1.31+/-0.17 and 1.84+/-0.68, respectively. The former has less bias (simulated K1=1.48) and reduced variation compared to the latter.

Conclusions We developed a respiratory motion compensated image reconstruction method for dynamic cardiac PET imaging. Using realistic simulation, we demonstrated that incorporation of respiratory motion correction within the reconstruction of dynamic frames improved the accuracy and uniformity of reconstructed polar map sequences. Moreover, the estimation of kinetic parameters was also shown to be significantly improved, especially for the regions mostly affected by respiratory motion.