MRI Guided Myocardial Perfusion PET Image Reconstruction Jing Tang, Member, IEEE, Xinhui Wang, Nicolas A. Karakatsanis, Member, IEEE, Lijun Lu, and Arman Rahmim, Senior Member, IEEE Abstract–Integrated whole-body PET/MRI provides opportunities to fully take advantage of simultaneously acquired anatomical and functional information. The purpose of this study is to incorporate the MR measured anatomical information in myocardial perfusion (MP) PET image reconstruction and to quantitatively evaluate the reconstructed images. Using the 4D XCAT phantom, we simulated cardiac-gated MP PET data with and without a perfusion defect and the corresponding MR images. Noisy PET sinograms were generated with count levels comparable to patient Rb-82 MP PET measurement. MR images were simulated using the SIMRI simulator, with the MR sequence specified to be 3D T1-weighted as used in a clinical PET/MRI protocol. For each cardiac gate, we applied the closed- form maximum a posteriori (MAP) PET image reconstruction taking the joint-entropy (JE) between intensity of the PET and MR images as the prior. To quantitatively evaluate the reconstructed images, we used the tradeoff between bias and noise on the left ventricle polar map. The contrast recovery ratio was also calculated to quantify the ability to classify the polar maps with and without the MP defect. On the whole polar map and its segments, the activity values estimated from the JE MAP algorithm showed significantly improved noise versus bias tradeoff compared to those from the conventional maximum likelihood algorithm. The JE MAP algorithm also resulted in improved noise versus contrast recovery for the MP defect. To conclude, we demonstrated quantitatively improved performance of the anato-functional JE MAP reconstruction on MP PET imaging with realistic simulation. The reconstruction technique will have promising potential applications especially in the emerging integrated cardiac PET/MR imaging. I. INTRODUCTION HE arrival of integrated whole‐body PET/MRI provides new opportunities and challenges to fully take advantage of the simultaneous acquisition of anatomical and functional information [1]. MR image based attenuation correction is critical in quantitative PET image reconstruction and has been broadly studied [2, 3]. Works have also been carried out on motion compensated PET imaging using motion estimated from simultaneously taken MR images [4, 5]. Before the availability of integrated PET/MRI, techniques on using anatomical information to assist PET image reconstruction have been developed [6]. However, very few studies have Manuscript received November 15, 2013. This work was supported in part by the U.S. National Science Foundation under grant ECCS-1228091. J. Tang (e-mail: jtang@oakland.edu) and X. Wang are with the Department of Electrical and Computer Engineering, Oakland University, Rochester, MI 48309 USA. N. A. Karakatsanis and A. Rahmim are with the Department of Radiology, The Johns Hopkins University, Baltimore, MD 21287 USA. L. Lu is with the School of Biomedical Engineering, Southern Medical University, Guangzhou, Guangdong, China. been performed on cardiac imaging. This is most likely due to the complicating issues of cardiac motion and of the difficulties in registering the PET and MR images, prior to the advent of the simultaneous acquisition capability. We developed a Bayesian image reconstruction technique incorporating the joint-entropy (JE) of PET and MR image features as the regularization constraint [7]. The technique was applied and evaluated with simulated and patient brain data. In this study, we propose to use MR information for myocardial perfusion (MP) PET image reconstruction. Using realistically simulated cardiac-gated PET data and MR images, our goal is to evaluate the JE incorporated image reconstruction method in MP activity estimation. The proposed technique is expected to contribute to the growing clinical applications of cardiac PET/MR imaging [8]. II. MATERIALS AND METHODS In this study, we simulated both the cardiac-gated MP PET imaging data with and without a perfusion defect and the corresponding MR images. For each given cardiac gate, we performed maximum a posteriori (MAP) reconstruction using the JE between the PET and MR images as the prior. To quantitatively evaluate the reconstructed MP images, polar maps were created for the left ventricle (LV) myocardial activities. The tradeoffs between noise versus bias on the polar map segments and the tradeoff between noise and lesion contrast recovery were used to compare the images reconstructed from the JE MAP method and the conventional maximum likelihood (ML) method. A. PET Data Simulation Using the 4D XCAT phantom [9], we simulated two cardiac gated MP imaging datasets, one with normal perfusion and the other with regionally reduced perfusion. The perfusion defect was a transmural defect spanning 40º over the anterior-lateral region and 1.5 cm over the long-axis direction. Its activity was 30% less than the normal activity. The time activity curves (TACs) of the blood pool, myocardium, and other organs were extracted from Rb-82 PET images of 5 patients with normal cardiac function [10]. They were smoothed and averaged to acquire a set of TACs representing the typical Rb-82 biodistribution. Analytical simulations were performed to simulate noise-free PET data of 8 cardiac gates. The noise-free sinograms were scaled to 5-min cumulated activity after pre- scan delay of ~30 sec to avoid high blood pool activity. For both normal and abnormal perfusion cases, Poisson noise realizations were implemented. T