Development and evaluation of two 4D image reconstruction methods with dual respiratory and cardiac motion compensation for gated cardiac PET

Abstract

The goal of this study is to develop and evaluate two 4D statistical iterative image reconstruction (SIIR) methods with respiratory motion (RM) and cardiac motion (CM) compensation for improving the detection of motion defects in 4D gated cardiac PET. The first 4D SIIR method, motion correction after reconstruction (MCAR) was developed previously, where the dual RM&CM compensation was achieved by applying the estimated RM&CM transformation to the RM&CM gated reconstructed images after the 4D image reconstruction, respectively. A new motion correction during reconstruction (MCDR) method was developed, where the RM&CM compensation was applied at each iterative step by modeling the estimated RM&CM within the projection and backprojection matrix during the 4D SIIR. In both approaches, the RM and CM were first estimated independently from only the 4D respiratory amplitude-gated and cardiac time-gated reconstructed images, respectively. In both methods, the RM&CM compensated image was transformed back into each cardiac gate to the acquired gated images, and the non-gated attenuation map was transformed using the estimated RM to reduce mis-registration artifacts. The performance of the two methods were evaluated using Monte Carlo simulated gated MP PET images of the 4D XCAT phantom with known true RM&CM at different noise levels, and sample clinical 4D cardiac gated PET studies. The simulation study showed the MCDR provides sharper 4D gated MP PET images with ∼12% reduction in mean-square-error when compared with those from the MCAR at large number of iterations approaching convergence. The results from using patient data showed the MCDR method provided consistently superior contras-to-noise ratios. We conclude that although the MCDR method is slower, it outperforms the MCAR method in 4D gated cardiac PET image reconstructions. Our findings are supported by simulation and sample patient studies. Human trials using this method are pending.