The Research On Statistical Iterative Reconstruction Algorithm For Computed Tomography Based On Image Domain

As computer tomography(CT) technology is widely used in medical diagnosis and treatment, people pay more and more attention to the influence of radiation dose on human body. High doses would bring certain harm to people’s health, therefore low dose CT has gained much currency in recent years. Low dose data can obtain by reducing the ray tube voltage or current. However, such method results in heavy noise in projection data, and lowers the density resolution because of quantum noise, so that the reconstructed image ends in serious degradation, affecting medical diagnosis and treatment. So the research on radiation dose reduction while reconstructing images with high signal to noise ratio and high quality attracts more and more attention. Many methods about the low-dose CT reconstruction have been proposed, and could be mainly divided into the following three types: one is the image post-processing, and one is declining noise existing in projection data. As the FBP algorithm is simple and easy to implement, FBP projection reconstruction is often used after the noise reduction of the projection data. The third method is reconstructed image de-noising directly in image domain. After the noise suppression in projection domain, even the little speckle noise, when mapped to the image domain, will lead to bar artifact. Given all these concerns,this paper reduces the noise directly in image domain. The main work is as follows:1. firstly, expound the physical and mathematical principles of the CT imaging, then introduce three classical image reconstruction algorithms, and finally show the noise model used in this paper.2. In view of serious degradation of the low-dose CT reconstruction images, there are two kinds of improved methods based on the anisotropic diffusion: the first one is to improve the diffusion function with Patch Similarity, which could decline the noise while maintaining the image edges and details. Considering the insufficient noise reduction of traditional anisotropic diffusion algorithm, another one is to introduce variable exponential and similarity function into traditional anisotropic diffusion algorithm to construct the modified diffusion function, thus to achieve better image quality. For variable exponential can reach a goodbalance between heat transfer and PM mode, while similarity function can detect the edge and details instead of gradient method. Thus our method will improve the image quality.3. A new low-dose CT reconstruction algorithm based on wavelet shrinkage and fourth order anisotropic diffusion has been proposed. Because wavelet transform has a good time-frequency local features, and fourth order partial differential is highly sensitive to the noise, and less noise would contribute to better noise reduction performance, so the fourth order partial differential can avoid the ladder effect of the second order PDE. In the paper,wavelet shrinkage and anisotropic diffusion are combined to decompose the image rebuilt by MLEM reconstruction algorithm with discrete stationary wavelet in each iteration. Then its high frequency part in wavelet domain is processed with wavelet shrinkage, while low frequency part with fourth-order anisotropic diffusion which can effectively suppress the noise. So image with high quality can be obtained.4. Total variation Median prior reconstruction algorithm based on wavelet and nonlocal method is put forward to solve the problems of stepladder edge and over-smoothness in reconstructed image by Maximum A Posterior which only provides local prior information.Firstly, on the basis of the median prior MP algorithm, the TV method, which has an excellent de-noising performance, is introduced to revise the objective function to form MP reconstruction algorithm based on TV. Then considering the advantages of wavelet shrinkage and nonlocal method, the algorithm in this paper is proposed, which, in each iteration of the MP reconstruction algorithm based on TV, adopts the wavelet shrinkage and nonlocal noise reduction to process the image after wavelet transform in the wavelet domain, thus improving the image quality.