Research On Projection Domain And Post-processing Algorithms For Low-dose CT

As a non-invasive means of observing inside the human body,Computerized Tomography(CT)has been widely used in clinical practices and medical diagnosis.However,the threat to human health caused by X-ray exposure has attracted more and more attention,reducing the dose of X-ray has become an urgent need.The quality of reconstructed images decreases with the decrease of dose,which affects the accuracy of clinical diagnosis.How to reduce the X-ray radiation dose while ensuring the reconstructed image is free from noise and keeping the structure and details clear,has always been a research difficulty and hot spot in the field of low-dose CT imaging.In this article,the algorithms based on the projection data and post-processing are deeply researched,the main works are summarized as follows:(1)A projection data restoration algorithm based on modified anisotropic diffusion weighted prior(MADWP)was proposed.Considering that the intuitionistic fuzzy entropy was able to distinguish the smooth and the edge area,the method combined it with the traditional anisotropic diffusion coefficient to form a novel anisotropic diffusion coefficient.This coefficient adaptively smooths the sinogram with local variance,enlarges the smoothness in the flat region to remove noise,and weakens the smoothness in the edge region to maintain the structural information.Then,it was integrated into the Maximum A Posteriori(MAP)optimization estimation model based on Huber prior to realize the processing of projection data.Finally,the final reconstructed LDCT image was obtained by performing the filter back projection(FBP).The visual analysis and objective evaluation indexes both demonstrated that the proposed method can obviously improve the image quality.(2)A difference-based morphological component analysis(DMCA)LDCT image post-processing method was proposed.The algorithm transformed the artifacts and noise suppression in low-dose CT images into image decomposition.Firstly,the stationary wavelet transform(SWT)was applied to get the high-frequency part of the LDCT image,so the artifacts suppression could be processed on multiple scales and multiple directions.The dictionary used in the traditional morphological component analysis(MCA)is fixed,and the selection of the dictionary is dependent on experience.How to select the appropriate dictionary is a challenging problem.To solve this problem,the online dictionary learning is used in this method to learn dictionary from the high frequency sub-bands.Based on the difference between the artifact atom and the tissue structural atom,a difference operator was developed to partition and reorganize the dictionary.Then the spare coding was applied to obtain the noise removal high frequency images,which were combined with low-frequency to restore the image.Finally,the total variation(TV)filter was performed to further remove the residual noise and artifacts in the image.In the experiment,the simulation phantom and real image were used to verify the effectiveness of the proposed algorithm.Both the visual effect and the objective evaluation indexes proved that the proposed method was not only effective in suppressing noise and artifacts,but also better than other algorithms in maintaining the structure and details.