Application Research On Several Problems In Iterative CT Reconstruction Technique

CT technology is able to accurately and visually display the structure information of objects in the internal organization, and nowadays has become an integral part in the field of diagnostic radiology. Especially in recent years, along with the rapid development of MDCT and multi-source CT technology, its clinical application has been extended to heart, lung dynamic checks and CT angiography, interventional treatment guide, and other fields, marking CT entered a new stage of rapid development. But at the same time, the incomplete projection data in CT imaging is inevitable, such as obstructions or shortening scan time to protect the patients’health. So the research on reconstructing CT images that meet the requirements of clinical diagnostic in the case of incomplete projection data is significant both in terms of theory and clinical applications.When the projection data is incomplete or noisy, the CT images reconstructed by classical analytic methods are incomplete or degraded by severe artifacts. Then, another type of CT image reconstruction method:iterative algorithm, would be adopted. It has great advantages of fewer projection required, lower SNR requirements and high reconstruction quality, and then it is ideally suited for CT image reconstruction in the case of incomplete projection data. Therefore, iterative CT reconstruction method is a better solution in the case of incomplete projection data.However, the iterative algorithms, such as ART and SART, take a much longer computing time, as is a fatal limitation of these methods. So it is difficult in practical applications where bear a relative high requirements of real-time. Due to this, in recent years, in order to shorten the reconstruction time, many scholars engaged in a large number of in-depth studies from the hardware structure, software acceleration and parallel computing. On the basis of full understanding of ART and SART, taking advantage of the emerging GPU heterogeneous computing model, a novel implementation of ART/SART based on the platform of CUDA is proposed in this study. The experimental results show that there are no differences between the images reconstructed by the new methods and those by serial implementation, but the reconstruction time is greatly decreased, as more applicable to clinical application.On the other hand, CT image reconstruction is essentially an inverse problem on finite number of projection data which is less than the number of unknown pixels. Such inverse process is an ill-posed problem, and then has an ill-posed solution. As the linear least squares method (LLS) can obtain the solution of underdetermined equations with certain error conditions, LLS-based optimal solution for image reconstruction has gradually become a research focus. In this study, on consideration of real-time and accuracy, LLS is improved with iterative method, and we propose new methods for CT image reconstruction. ISTA/FISTA are introduced, and a novel class of CT image reconstruction method:ITVA/FITVA is proposed. The simulation results show that these methods are able to reconstruct high-quality CT images in the case of under-sampling and limited-angle projection with a rapid convergence rate, which bear a wonderful application performance and great potential for development.