| Computed tomography (computed tomography, CT) is one of the most state-of-the-art medical imaging technologies, and it has been widely used for its outstanding performance in various fields of medicine. With the development of computer science and technology, electronics and information technology, the CT scanning technology has been constantly improved, experienced a translation from a single thin beam, narrow fan beam, wide fan beam rotation (rotation), wide fan beam rotation (stationary scan mode) to the recent multi-slice spiral scan mode, greatly improving the efficiency of the projection data acquisition.CT reconstruction technique is the use of X-ray tube and the detector around the body to make synchronized rotation and in each angle detector can detect the amount of the remaining X-rays passing through the human body. After photoelectric conversion and analog-to-digital conversion, digitized projection data is obtained. The particular mathematical algorithm is used to process the projection data in computer, to obtain a tomographic image of the human body. In clinical diagnosis, the doctor obtains important reference to accurately diagnose the illness by observing tomographic images, improving health care quality and safety.CT reconstruction algorithms are divided into analytic algorithm and iterative algorithm. The representative algorithm among analytic algorithms is filtered back projection (filtered back projection, FBP) reconstruction method and the idea is:first filtering the projection data, and then back-projection to the reconstructed image, after all angles of projection data are processed, the desired image is obtained. Advantage of this algorithm is simple, fast reconstruction, but the drawback is that it requires much projection data and the reconstruction image noise increases with the projection data noise increases. Iterative algorithm can be divided into algebraic and statistical iterative method. Algebraic method is based on the solution of linear equations, on behalf of the algorithm for ART (Algebraic Reconstruction Technique). Statistical iterative method takes into account the statistical properties of the projection data, on behalf of the algorithm for maximum likelihood expectation maximum method (Maximum Likelihood Expectation Maximization, ML-EM) and maximum a posteriori estimation method (Maximum a Posteriori, MAP). ML-EM algorithm assumes the projection data obeys Poisson distribution, then the imaging model is established, and its advantage is that the reconstructed image quality is superior to FBP algorithm, the disadvantage is slow convergence. MAP algorithm by further constraints on the reconstructed image data fidelity term based on the introduction of the penalty term, reconstruction of quality and the convergence of the iterative process is significantly improved, successfully applied in PET imaging and CT imaging.In order to obtain high-quality tomographic images, CT technology usually uses higher radiation dose, but high-dose X-ray irradiation on the human body may induce cancer, genetic mutations, leukemia and other diseases. Therefore, the X-ray dose aroused increasingly more attention. For example, image guided radiation therapy (IGRT) is a technology for the patient’s anatomy and the uncertainty of the location information, and three-dimensional anatomical images using imaging equipment and technology are obtained for current patients during radiotherapy, using these images to correct treatment plan, thereby increasing the conformal degree of the radiation dose to the tumor, improving the local control of the tumor and reducing normal tissue complications. However, patients usually have to accept multiple scanning, so the body has accumulated dose, which will reach dangerous levels, leading to gene mutations or other cancer inducing. Currently, there are two ways to reduce the radiation dose. The first method is to reduce the tube current parameters, but projection data noise increases, the noise of the reconstructed image is also increased, decreasing quality of a reconstructed image. The second method is to reduce the number of samples of the projection data, also known as sparse angle projection data, the noise content is small in the projection data, but the amount of data is not full, image reconstructed using FBP algorithm contains the serious streak artifacts, and the image quality is poor. Therefore, reducing the patient radiation dose, while accessing to high-quality tomographic images, is enormous challenges for the CT imaging field to be a breakthrough, and major CT manufacturers view it as a hot scientific issue.In addition, CT reconstruction speed is also a matter of concern. Acceleration technology is divided into software acceleration technology and hardware acceleration technology. In software acceleration, fast Fourier transform method (Fast Fourier transform, FFT) is sometimes used in some reconstruction methods. In hardware acceleration, one can increase the number of computer CPU to increase computing power or let host mission sent to multiple computers in parallel computing, finally host PC gets comprehensive results to process. NVIDIA company produced CUDA (Compute Unified Device Architecture)-based products in2007. CUDA can effectively use the powerful GPU parallel processing capability and large memory bandwidth for general purpose computing, which has been successfully applied to a variety of medical image processing calculations, including:image reconstruction, image registration, image segmentation, dose calculation, and achieved a good performance and dozens of times acceleration. With CUDA-based GPU products further upgrading, a variety of medical computing processing speed will be further enhanced and make more remarkable contribution to the field of human medicine. Part of this paper is to use CUDA to accelerate the proposed iterative algorithm for CT reconstruction, and satisfactory reconstruction efficiency was obtained.CT reconstruction algorithm research work, under the premise of reducing the radiation dose and improve image quality and imaging speed, is iterative image reconstruction algorithm study for sparse angle projection data, having made initial progress and results.As the basis for the research content, the paper first introduces the composition of the CT hardware, theoretical basis of CT imaging, fan beam FBP and cone-beam FDK reconstruction algorithm, the ART algorithm and the ML-EM algorithm. Experiments results showed the shortcomings of FBP and FDK algorithm and advantages of the iterative algorithm using sparse angle projection data for image reconstruction. According to the existing various iterations imaging algorithm, three different new iterative algorithms were improved and achieved.First, propose a bilateral filtering based iterative m reconstruction algorithm modified ART using few view projection data. When the projection data is sparse, ART algorithm is solving an underdetermined equation, but the solution is not unique, needing to force the image further constraints, such as the filtering process. C.Tomasi and R.Manduchi first proposed bilateral filtering algorithm for image denoising. Bilateral filtering algorithm can not only effectively reduce the image noise but also keep the details of the edges of the image, while taking advantage of the similarity of the spatial proximity of the neighborhood of pixels and brightness information. In each iteration, the proposed algorithm first reconstructed image using ART algorithm to meet the projection data consistency and a non-negative constraints on the pixel values in the reconstructed image, then bilateral filtering method is performed on the image to filtering correction. The process re-enters the next iteration until meeting the iteration termination condition. Shepp-Logan digital phantom reconstruction and a real head phantom projection data reconstruction is conducted to verify the feasibility of the method, the results show that the algorithm can reconstruct the image with higher signal-to-noise ratio, better able to maintain the edges of the image information.Second, propose and achieve an algorithm by minimizing the total variation penalized a divergence reconstruction from sparse views data. Total variation is also a good method for image denoising. Hence it can be used in image reconstruction model. First building a cost function consisting of the a divergence term and the total variation term in the imaging model, and then using auxiliary function to solve the model. In order to speed up the reconstruction, ordered subset (Ordered Subsets, OS) strategy is used. In OS approaches, the projections are grouped into different subsets. Only one subset of projections at one time is used for updating of the image estimate, and this update together with a different subset of projections is then employed for performing the next update. Simulation data and real data reconstruction experiments show that, the total variation penalized a divergence reconstruction from sparse views data achieved very good performance in the suppression of image noise and streak artifacts, greatly improving the image quality.Third, develop a density priori guided cone beam CT sparse angle3D image reconstruction technique, and use CUDA to accelerate reconstruction. Normal CT images, distribution of some homogeneous regions in the image is large (such as air, muscle, bone, lung, etc.), and the difference between the values of the density of each region of homogeneity is smaller. Therefore, knowing the density of the structure of some of the parts as priori to scan the patient, will play a very important role in the quality of the image reconstruction, particular suitable to patients needing multi-scan imaging, such as image-guided radiotherapy. First of all, the objective function is defined by the data consistency term and density of a priori information item posed; then splitting algorithm is used for solving this objective function; Finally, using CUDA programming to implement sparse angle cone-beam CT image reconstruction of good quality. NL-Means method, due to its superior denoising edge-preserving properties, is used only3-5times in the process iterative method, taking into account the time-consuming calculation. Experimental results show that using the iterative reconstruction algorithm for CUDA programming, under the premise of lowering the dose, not only obtain a satisfactory image quality but also reconstruction speed is improved significantly.CT image reconstruction is a very complex subject involving many fields of physics, mathematics, computer science and other disciplines knowledge. A lot of factors affect the quality of the CT image reconstruction, such as CT geometry calibration, the detector element response consistency, reconstruction algorithm selection, and these aspects are affected and constrained by each other. The work done in this paper is just some preliminary exploration in the field of CT image reconstruction, though has made some preliminary research results, but still need to do further and more in-depth study. |
PDF Full Text Request