The Research On Techniques Of3D PET Image Reconstruction

Positron Emission Tomography (PET) acts as one of state-of-the-art technology of nuclear medicine. It provides tissue functional imaging with the aid of radioisotope, and can measure quantitative changes over time in the bio-distribution of radiopharmaceuticals throughout a target structure or the organs of interest. Physiological and/or biochemical parameters are then derived with the additional use of tracer kinetic modeling techniques. Positron emission tomography, however, is an ill-posed inverse problem because the observed projection data are contaminated by noise due to low count rate and physical effects. This problem of low counts is further accentuated with increased temporal sampling in a certain time period.Though traditional filter back projection (FBP) method has the advantage of less compution cost, it often results in noisy images of low quality. Better expressing system models of physical effects and modeling the statistical character of the measurement data, the maximum-likelihood expectation-maximization (ML-EM) approach outperforms the FBP method with regard to image quality and becomes the standard reconstruction algorithm instead of FBP for clinical PET. However, MLEM will produce inceasing noise with the increasing iteration, thus results in non-covergence iteration. Recently, Bayesian methods or MAP (Maximum A Posteriori) methods solve this prolem by incorporating image prior information and have been proved theoretically correct and practically effective compared to other methods. As to the problem of reconstructed image quality, Bayesian reconstruction can greatly improve reconstruction by incorporating prior information compared MLEM reconstruction. We also find that relied on the local neighborhood information of image-self, Bayesian methods can only contribute limit local prior information to reconstruction.Our researches are carried out under Major State Basic Research Development Program (No:2010CB732503). Our work on PET reconstruction algorithm is based on how to further improve PET image reconstruction quantitatively, especially for dynamic PET reconstruction quantitatively. We have done following work on PET reconstruction algorithms:1. We proposed an anatomically adaptive nonlocal prior. To deal with the parameter optimization problem in nonlocal prior MAP PET reconstruction, we proposed an anatomically adaptive nonlocal prior, which incorporate the regional information of anatomical image. The parameter of prior is adaptively estimated by using the regional information of anatomical image in the iteration. In experiment, we compare the proposed prior model with the nonlocal prior using fixed parameter. The simulation results show that the anatomically adaptive nonlocal prior can improve reconstructed image quantitatively and preserve the edge.2. We proposed generalized entropy and MR prior induced MAP reconstruction. Based on the development of incorporating the MR anatomical information into PET imaging, we proposed generalized entropy and MR prior induced MAP reconstruction. This approach did not require explicit segmentation or boundary extraction, and construct a prior based on the mutal information or joint entropy of the grey character of MR image and PET image for MAP PET reconstruction. The mutal information and joint entropy is based on generalized entropy. Simulation results show that generalized entropy and MR prior induced MAP reconstruction can effectively incorpating the MR information and improve the reconstructed image quantitatively.3. We proposed a "3.5D" dynamic PET reconstruction incorporating the kinetics-based cluster to enhance the image reconstruction and estimate the parmetric image. To deal with the high noise and low resolution in conventional3D dynamic PET reconstruction, we clustered the preliminary reconstructed dynamic images, defined neighborhoods of voxels with similar kinetics, and then introduced the prior into the MAP reconstruction of individual frame. The approach is labeled "3.5D" image reconstruction, because on the one side it is related to the emerging field of spatiotemporal4D PET reconstruction while on the other hand, the final reconstruction step includes straightforward application of maximum a posteriori (MAP) reconstruction to the original individual dynamic frames without the need for advanced transforms, temporal basis functions, or kinetic models as pursued in the4D reconstruction framework. The proposed3.5D dynamic reconstruction algorithm resulted in quantitatively enhanced parametric images, as demonstrated in extensive11C-raclopride PET simulation as well as an HRRT patient study.