PET Myocardial Polar Map Quantitative Analysis And Image Reconstruction

Positron emission tomography (PET) represents the most cutting-edge technology in nuclear medicine. It is a non-invasive nuclear medicine imaging technology. It makes use of radionuclides to mark some compounds as tracers. Then inject the tracers into human body directly and collect the rays from different angles of body. Then computer is used to complete the image reconstruction for the data collected from the body. Through the distribution of the radionuclide in the human body, the metabolism level and function of various organs in the body can be dynamically, noninvasively and quantitatively evaluated. That makes PET plays an important role in the diagnosis and prognosis of cardiovascular diseases, tumor and diseases of the nervous system. PET can reflect the myocardial metabolism and function of the human body on a molecular level with high specificity and sensitivity. We can get physiological or biochemical parameters by using the dynamic model.Coronary heart disease (CHD) is a major cause of death in developed countries since the middle of twentieth Century. It is a progressive disease. Its course is generally beginning in the childhood and showing clinical symptoms in adulthood. In our country, the incidence and mortality of coronary heart disease is on the rise. Therefore, early diagnosis and prognosis of coronary heart disease is very important. PET myocardial perfusion and metabolic imaging has incomparable advantages in diagnosis of heart diseases compared with other diagnostic equipment. Nowadays, 18F-FDG (FDG) PET myocardial imaging is considered the gold standard for examination of myocardial viability. The most commonly used quantitative analysis method of PET myocardial perfusion and metabolism is polar map. It rearranges the myocardium circumferential profile image from apex to base in the mode of concentric circles from the inside to the outside. So the whole left ventricular myocardial will be projected into a bull’s eye diagram. In fact, it is a reconstruction of continuous cardiac short axis images. Computer and medical image processing technology has been widely used in the field of nuclear medicine imaging overseas, some processing PET myocardial metabolism and perfusion images polar map quantitative analysis software have been used in clinical, such as one-tissue compartment model PMOD and two-tissue compartment model QPET. But, there is little such software in China.This article made a piece of polar map quantitative analysis software and compares the effect of hemisphere cylinder myocardial model and ellipse model. Our polar map quantitative analysis software was generated as follows:(1) to obtain the initial short axis images. The human body cross section tomographic images was reconstructed according to the direction of left ventricular long axis and so we could obtained short axis slice images, horizontal long axis images and vertical long axis slices images. Where, the tomographic images vertical to the long axis of heart were short axis images; (2) the largest radioactive count value sampling. Maximum radioactive counting sampling method was the key of myocardial radionuclide imaging quantitative analysis. The automatic correction of the left ventricular long axis, myocardial model fitting and the bull’s-eye drawing should use the maximum radioactive count sampling. This study used half sphere cylinder model and the ellipsoid model to generate the polar map; (3) left ventricular long axis orientation correction. The left ventricular long axis was not parallel to the body long axis. So we should determine and correct the left ventricular long firstly. We designed an automatic correction method for left ventricular long axis adjustment; (4) left ventricular myocardial model fitting. We chose different objective functions for the two kinds of left ventricular model and used the Gauss-Newton optimization method to fit model parameters; (5)generate polar map. In the polar coordinates the entire cardiac short axis slice images are projected onto a plane with 16 ring from inside to outside are apical cap, apex, left ventricular cavity and basal of left ventricular.,1 to 4 rings are the apical cap,5 to 8 rings are the apex,9 to 12 rings are heart cavity,13 to 16 ring to the basal of left ventricular. Each ring sampled 36 largest radioactive count value points and the entire myocardium is divided into 576(16 X 36) segments. Then, the 576 segments were merged as 17 segments according to the American Heart Association. We made a bull’s-eye quantitative analysis for a set of real 18F-FDG PET myocardial metabolic images and compared the 17 segments bull’s-eye and 576 segments bull’s-eye of the two kinds of myocardial models respectively In addition, to further compare the difference between the two myocardial model, the Bland-Altman analysis were taken for 576 segments of different model sampling by using the coefficient of variation to assess the consistency of the two models The 95% confidence interval was (-12.7,18.9); mean value was 2.5; and the coefficient of variation was 3.76%. The result showed there was no statistical difference between the two model sampling. It illustrated that our polar map quantitative analysis method can be both applied to the two kinds of myocardial models. In addition, we also simulated a set of a dynamic PET 82Rb myocardial perfusion images according to the one-tissue compartment model simulation method. And we also made a polar map analysis for them. The results showed that our polar map analysis method can show the position and information of myocardial ischemia correctly, which proofed that our polar map quantitative analysis is reliable.The polar map quantitative analysis method has great value in PET myocardial blood flow(MBF) quantitative analysis. But the PET dynamic data will be subdivided into several short frames in dynamic reconstruction. It will bring in noise, which affecting the absolute myocardial blood flow quantitative, so application of dynamic PET perfusion imaging was limited in the clinical. Traditionally, standard dynamic PET image was reconstructed by individual frame independently and then in the voxel or region of interest (ROI) level, the dynamics model is applied to generate the time activity curves (TAC). Independent image reconstruction was mainly relied on statistical image reconstruction methods, such as the maximum likelihood expectation method. However, when in low count situations, using direct maximum likelihood expectation method, the variance was high and will increase with the sampling interval increasing. Some reconstruction strategy by improving the reconstruction accuracy of dynamic frame to estimate the kinetic parameters had been put out. For example,4D reconstruction improved the accuracy of parameter estimation by using dynamic PET scan time and space. But 4D reconstruction method increased the algorithm complexity and needed further optimization. There are many methods considering to use sparse characteristics of PET image. For example, the total variation regularization has been used in the PET reconstruction whether in the image space or in the measure space. Multidimensional wavelet denoising method can restore the biological signal fidelity hidden in the dynamic PET image, which madding quantification of myocardial perfusion accurate. In 2008, Su and Shoghi proposed the wavelet denoising technique, which is less sensitive to noise and can provide a more accurate estimation of the parameters in the high noise levels. Recently, some researchers put forward using low rank matrix properties to restore image, this study was first noted in dynamic magnetic resonance imaging. Scholar Lingala and Liang combined it with the sparse prior for magnetic resonance image restoration and reconstruction respectively. There were highly correlated redundant information between consecutive images in dynamic PET myocardial perfusion imaging, which makes the image matrix has low rank property. In addition, some parts of the image contained the perfusion information, which were sparse. Therefore, the matrix contained the background part L and the dynamic part S in fact. The background component stands for the highly correlated information among frames or the regions (e.g. muscle and lung regions) with tracer activity changing slowly over time, and can be assumed to be low rank property. The dynamic component corresponds to the innovation introduced in each frame or the regions (e.g. myocardium and blood pool) with tracer activity changing rapidly over time. So the PET dynamic image has both low rank and sparse properties.Therefore, in this paper, we propose a reconstruction method combination with low rank and sparse penalty terms in dynamic PET myocardial perfusion imaging. This paper uses an improved Split Bregman method to solve optimal cost function. For our simulation, we used five patients scans with healthy myocardial at the Johns Hopkins PET Center, from which K1, k2 rate constant and Fa were estimated for multiple regions (e.g. myocardium, lung et al) for each study and subsequently averaged. The arterial blood tracer concentration Ca(t) was measured from the mean value of five patient blood samples. Through the K1,K2, Fα and arterial blood radioactive tracer concentration Ca(t), we generated a set of dynamic PET myocardial perfusion image. And use real 82Rb PET myocardial perfusion image to optimize the regularization parameters. The proposed reconstruction model resulted in substantial visual as well as quantitative accuracy improvements in terms of noise versus bias performance. The quality of images reconstructed from the proposed reconstruction algorithm was in particular compared with that reconstruction using MLEM, single sparse penalty or low-rank penalty. In order to compare the performance of different reconstruction algorithms, we computed the mean square error (MSE) for each frame of dynamic image. The results showed that the MLEM approach results in largest MSE value for all the frames. The low rank penalty can achieve substantially less MSE than the sparse penalty. Our reconstruction method results in smallest MSE value for all the frames. To provide an intuitively direct visual impression of the reconstruction algorithm across the entire images, we give the reconstructed images using different algorithms with optimized regularization parameters. It is clearly seen that the MLEM algorithm and sparse penalty (S) approach significantly increased the level of noise, while low rank penalty (L) and our reconstruction method significantly reduces the noise level. Our method reduces the noise level efficiently while maintains feature information perfectly. In addition, in order to evaluate myocardial defect, we show the 17-segment polar maps of frame 10 obtained from different reconstruction algorithms to compare different reconstruction algorithms. We can find that the polar map created from L&S model has much similar grey level performance with the reference true polar map, and can accurately reflect myocardial ischemia position and information.