| Positron Emission Tomography, PET for short, is a new developed technique in nuclear imaging filed recently, indicating the high standard of clinical diagnostic imaging. After the intravenous injection of radioisotope labeled tracer to the patient, PET can provide tracer activity images by in vivo imaging within the available FOV (Field Of View), offering the notable capability to measure changes in the metabolism and making the quantitative analysis of biochemical process possible. Compared with static PET imaging, dynamic PET imaging can provide tracer activity sequence images over time, revealing the nature of the tracer bio-distribution information inside the patient. By applying the various kinetic modeling techniques to the dynamic PET images, the functional parameters of human tissue can be estimated subsequently, such as the local blood flow, metabolic rate and material transport rant, which favors the description of the metabolic state at molecular level with abundant clinical significance. Currently, dynamic PET imaging plays a more and more important role in the diagnosis of cardio-cerebrovascular diseases, cancer staging and pathology research. As one of the myocardial perfusion imaging scanning mode, the advantages of dynamic PET imaging are particularly obvious.As a noninvasive technique to detect the defect myocardium, myocardial perfusion imaging (MPI) is easy-operated and realized which makes it the first noninvasive choice to diagnosis coronary heart disease. Through utilizing the radioisotope labeled tracer technique, MPI can provide blood perfusion information within the myocardial region in specific condition with PET. After the injection of radioactive tracer to the human body, the tracer flow with blood and is absorbed by the normal cardiac muscle cell, not by the abnormal cell, that makes the tracer intake difference within the myocardium. Doctors can determine if there existed defect myocardium and know the range and degree of the defection directly according to the reconstructed tracer activity images of the patient, which helps the diagnosis, curative effect evaluation and prognosis of the coronary heart disease. At present, two scanning modes of MPI are explored:single photon emission tomography (SPECT) and positron emission tomography (PET). However, the MP-SPECT is limited in clinic primarily due to the qualitative or semi-quantitative results provided by SPECT, which is caused by the restrictions on the mechanism and imaging detection techniques. In contrast to conventional SPECT, MP-PET measures the changes in the tracer intake within the myocardial region over time and provides the quantitative kinetic parameters of myocardial tissue after applying the kinetic model, such as coronary flow reserve (CFR) and myocardial blood flow (MBF), hence offers improved accuracy basis for the diagnosis, curative effect evaluation and prognosis of the coronary heart disease, possessing extensive application values and clinical significances.In order to improve the time resolution of dynamic sequence images, we need to obtain the continuous tracer activity images over a certain time during the dynamic san. However, this would cause shorter scan time and less detected photon counts for individual frame, which make a great impact on the quality of the individual reconstructed image ultimately. In the clinical setting of MP-PET, the image is affected by higher level noise and the quality is degenerated greatly if reconstructed using individual frame projection data directly. The images with worse quality will corrupt the accuracy of quantitative analysis later and restrict the extensive use of the MP-PET. Therefore, it is always a hot research topic in the field of PET imaging on how to obtain high quality PET images with lower photon counts under the existing hardware conditions and scanning protocols.Traditionally, standard dynamic PET images reconstruction consists of independent image reconstruction at individual frames followed by maximum-likelihood exception-maximum (MLEM) algorithm. However, the MLEM reconstructed images will be damaged by higher level noise due to the lower photon counts. At present, there are two main strategies on this problem of low counts:(1) prior information induced PET images reconstruction; (2) PET images restoration. For the first one, some relevant prior information will be introduced in the reconstruction process to improve the image quality. However, it is also complicated about how to choose the appropriate prior model. Some reconstruction strategies (e.g. 4D reconstruction) exploit the spatiotemporal correlation in dynamic PET scan to lower the noise level and improve the noise image. However they are not very appropriate for clinic due to the intensive algorithm and computation. In contrast, the easy-realized and operated PET image restoration technique is more readily applicable in clinic.The so-called image restoration technique means restoring the noise PET images reconstructed by MLEM or OSEM algorithms in image domain. Gaussian filtering is the simplest restoration method but the application is limited due to the constant filter kernel in space, which will damage the signals and organ boundaries and cause blur effect in images. Restoration by the edge-preserving filter or non-local filter relies much on the hyper parameters, but how to select the parameters is still an open problem. CT/MRI anatomically guided PET images restoration is always limited by the registration between the PET images and anatomical images. A better registration requires tiny movements in the process of the two scans which is hard to implement. Another common-used restoration strategy, HYPR, utilize the high SNR of the composite image to improve the low SNR of single reconstructed image. The anatomical location of dynamic PET sequence images must be same when applying HYPR. All these approaches mentioned above are mainly used for 2D images filtering, which mean the PET images are denoised frame by frame. Actually, they are not very appropriate for dynamic PET images restoration because the correlation between the dynamic sequence images is out of consideration. Alternatively, there are many image restoration techniques considering the use of sparse property of PET images, such as restoration based on the total variation regularization and wavelet transform. Compared with the restoration techniques frame by frame, techniques exploit the sparse properties perform better obviously.So far, there is still less research on myocardial perfusion dynamic PET images restoration. We usually adopt conventional restoration models to restore myocardial perfusion dynamic PET images, without a special restoration model. However, conventional models are not very suitable due to the specific scanned region in the MP-PET scan. Aim at solving this problem, our work focus on a restoration model special for myocardial perfusion dynamic PET images considering the properties the dynamic sequence images owned. Discussion on analysis and evaluation of the images are also included in this work:(1) Evaluation of the restored dynamic PET images from the proposed model; (2) Estimation and evaluation of the kinetic parameters after applying the kinetic model. We utilized polar map which is more comprehensive for myocardial perfusion images to make valid evaluation in addition. As a quantitative analysis method for myocardial perfusion tomography images, polar map, also known as bull's eye plot, can evaluate the viability and defection of myocardium quantitatively with improved accuracy. Therefore, we will utilize polar map to evaluate the restored images and parametric images in order to develop a more comprehensive and intuitive evaluation. In summary, the main works of this paper are:(1) Considering the properties of myocardial perfusion dynamic PET images, we develop a new restoration model based on low rank plus sparse decomposition. The restoration models designed for 2D images are not suitable for dynamic PET sequence images due to the more complicated noise distribution causing by different tracer activity in different frame. In this work, we consider the restoration of the whole dynamic PET sequence images as the restoration of a matrix. Each row of the matrix is corresponding to the time activity curve of single pixel, each column of the matrix is corresponding to a single reconstructed image. This matrix can be decomposed into two matrixes based on the low rank plus sparse decomposition:The first one represents the regions with tracer activity changing relatively slowly over time, taken as the background component (e.g. muscle and lung regions). The second one represents the regions with tracer activity changing rapidly over time, taken as dynamic component (e.g. myocardium and blood pool). Compared with the background component, the change of tracer activity is obvious in dynamic component. Dynamic sequence images always stand for highly correlated information among frames, which means the image matrix is low rank. And some regions of the dynamic sequence images are sparse due to the blood perfusion information. That is to say, the background and dynamic component are low rank and sparse respectively. For that reason, we introduce the low rank plus sparse decomposition into the restoration model, below detailed research is included:(1) Construction and optimization of the myocardial perfusion dynamic PET images restoration model based on low rank plus sparse decomposition. (2) Evaluation of the proposed model. The results demonstrate better performances with decreased noise and clearly visible anatomical structures of the restored images from our proposed model.(2) Estimating the kinetic parameters and then evaluating using polar map for validating the restored results from the proposed model. As the abundant usefulness of kinetic parameters in clinical diagnosis, we design realistic Rb-82 MP PET simulation experiment and simulate the time activity curves of different tissues (e.g. myocardium and lung) based on the known blood input function and kinetic parameters (e.g. K1 and k2). Using the simulated data, we extract the time activity curves of ROIs (e.g. myocardium) from the restored images to estimate the kinetic parameters. We adopt the conventional estimation method, weighted least square (WLS) to estimate the K1 and k2 respectively. And the estimated parameters are evaluated using polar map in order to validate the accuracy of the estimation and prove the reliability of the proposed restored model. All the results demonstrate the satisfied estimation of kinetic parameters based on our proposed restoration model, which can provide more accurate quantitative judgments for clinical diagnosis. |
PDF Full Text Request