Research On Projection Recovery And Insufficient Data Reconstruction For Low-dose CT

With the widely application of CT technology in the fields of medical, concerns is growingover the radiation dose. How to obtain a satisfied reconstruction while reducing the radiationdose has become the goal of domestic and overseas researchers. There are two ways to reduceX-ray dose at present, one cost-effective way is to acquire CT images with lower mAsprotocol. However, the projection data will be contaminated with excessive noise, resulting ina degraded image. Therefore, suppressing noise in the low-dose CT projection data is a crucialissue for the low-dose CT application. Another way is to reduce the total number of X-raypaths. However, projection data obtained by this method is not complete and unable to meetthe requirements of accurate CT reconstruction. Therefore, the reconstruction study of suchincomplete projection also has important significance for low-dose CT.This dissertation investigated the noise problem and sparse angle reconstruction in low-dose X-ray CT. The main contributions are outlined as follows:1．Since there are high correlation between the nearby views and detector units oftomographic projection, the nearby pixels in the Sinogram have a certain degree of similarity.So a new anisotropic diffusion filter based on fuzzy entropy was proposed to remove noises inCT projection data, taking the fuzziness between neighborhood pixels into consideration. Theproposed algorithm could adaptively adjust the diffusion degree of each pixel in the CTSinogram, avoiding excessive smoothing or excessive maintaining caused by the singlegradient information. In addition, new diffusion model could accurately determine diffusiondegree of pixels in the diffusion process, accelerating the diffusion rate and then savingprocessing time.2．A hybrid fuzzy-median filter which combined with the fact that isolated points inprojection data could be removed by a median filter was proposed. In this algorithm the traditional anisotropic diffusion function was replaced by the fuzzy membership function, andthen in diffusion process the intermediate image was updated by the median filter. Theexperimental results showed that the reconstructed image obtained by this new algorithm hada good effect in both noise suppression and strip artifacts removal.3．Considering the noise characteristics of low-dose CT projection data, an improvedpenalty weighted least squares algorithm, namely a statistical iterative algorithm based onenergy minimization was proposed. The algorithm worked with the gradient energy prior asthe penalty term instead of the Gibbs prior, and then the objective function was restricted bythe gradient energy. In the iterative process, a median filter also was used to update the middleimage to accelerate noise suppression. Then the Euler-Lagrange equation was used totransform the minimization problem into a diffusion issue in solving objective function, andas a result, the amount of calculation was reduced.4．About the sparse angle reconstruction issue, a median prior constrained compressedsensing reconstruction method for sparse angle low-dose CT was proposed after studying theTV-based compressed sensing. The median prior was constructed by introducing an auxiliaryvector, and in fact the composing process was actually another sparse transformation (hereinreferred to as "Median Gradient Transform"). This transformation further constrained thereconstructed image and this new algorithm overcame the shortcomings of blocky artifactsand unclear details in the TV-based compressed sensing, and could obtain better qualityreconstructed image.