Methods For Improving CT Reconstruction Quality With Incomplete Projection Data

Computed tomography(CT)is an advanced imaging technology that obtains the cross-sectional information of the scanned object by measuring the projection data in different views,which is widely used in medical diagnosis,industrial detection and so on.The traditional analytic reconstruction algorithm requires the completeness of projection data to achieve satisfied reconstruction quality.However,due to the restriction of the detected object and environment,the sampling conditions are usually not ideal,thus the projection data can only be sampled with sparse view or within a limited angluar scope in practical applications,resulting in incomplete projection data.In addition,irregular workpiece with the uneven thickness can be met in industrial applications.Restricted by the detector's dynamic response range,some regions of the detector can be underexposed or overexposed.Therefore,projection data in those regions can not reflect the real attenuation of the X-ray after penetrating the workpiece,leading to the imcompleteness of projection data.The structural information of the reconstructed image is incomplete and the artifacts are serious.All of the above cases belong to typical reconstruction problems with incomplete projection data,which are hot spots and one of the most difficult subjects in the field of CT reconstruction.Thus,research on them has great theoretical significance and practical application value.In recent years,the compressed sensing theory shows that sparse prior can effectively improve the image reconstruction quality in the case of sparse view sampling and limited-angluar-scope sampling.How to design an efficient regularization term is the key to improve the quality of reconstructed images in the above two cases.For the reconstruction problem of irregular workpiece,the lack mechanism of projection data becomes more complex.How to fuse the projection information of the multi-energy and complete the projection information is the current research focus.In this dissertation,the sparse view,limited angle CT reconstruction and the reconstruction problem of workpiece with irregular shape are studied intensively,and the corresponding reconstruction schemes are proposed.The main achievements of this dissertation are summarized as follows:(1)A reconstruction algorithm using Shearlet-based nonlocal regularzaiton and total variation regularization is proposed for sparse view CT reconstruction.Based on Shearlet,this algorithm constructs a nonlocal sparse regularization term based on image block-matching and block-grouping,which can effectively use the non-local self-similarity prior information to restore the details of a CT image.As Shearlet transform belongs to multi-scale directional sensitivity transform,compared with traditional wavelet transform,it can better preserve the anisotropic features of the image.Furthermore,the proposed algorithm combines the proposed nonlocal sparse regularization term with the TV regularization term to better preserve the edges of the reconstructed image.In addition,the algorithm uses the Moreau-envelope-enhanced L1 norm,which can effectively overcome the shortcomings of the traditional L1 norm in sparse view CT reconstruction.Experimental results show that the proposed algorithm can not only suppress artifacts and noise effectively,but also protect the fine structures,leading to improving the overall quality of the reconstructed images.(2)Concerning the CT reconstruction problem with sampling projection data in a limited angluar scope,a L0 norm minimization constrained reconstruction algorithm using adaptive block-matching 3D sparse transform regularization is proposed.In this algorithm,orthogonal dictionary learning technology is introduced to sparsely encode the nonlocal similar image blocks to construct adaptive block-matching 3D sparse transform(ABM3D).The advantage of ABM3D is that it can effectively capture the local structural varations of the image attributed to the adaptability of the dictionary to the image structure,and recover the structurual patterns of the image by using the nonlocal self-similarity information.In consideration of the characteristics of limited angle CT reconstruction problem,we also design an alternative optimization algorithm based on the POCS algorithm to effectively solve the proposed model.In addition,the procedure of image block-matching is repeated in the iterative reconstruction process to improve the accuracy of image block-matching,which promotes the reconstruction accuracy.Simulation and practical data studies show that the proposed method is superior to the competitive methods in both visual and quantitative evaluation.(3)Aiming at suppressing the artifacts and noise in spectral CT reconstruction with voltage switching mode and photon counting detector mode in the condition of sparse view sampling,we extend the traditional total nuclear variation(TNV)model,and propose a nonlocal total nuclear variation(NLTNV)regularized spectral CT reconstruction algorithm.The proposed algorithm uses the low rank property of the nonlocal gradient vector of the spectral CT images to construct a more robust structural similarity measure.The NLTNV regularization can model three kinds of prior information about the spectral CT image,namely structural similarity across channels,gradient sparsity and nonlocal self-similarity.This method can not only effectively couple the structural information of the image of each channel,but also can effectively utilize the nonlocal self-similar prior information of the image in the spatial domain to improve the ability of suppressing artifacts and noise.As only one regularization parameter is needed to select in the reconstruction model,the complexity of selecting which is reduced.In addition,the reconstruction model is convex,which guarantees the stability and convergence of the reconstruction algorithm.(4)In order to solve the irregular metal workpiece reconstruction problem of single material with high density and large thickness variation,a reconstruction scheme based on dual-energy projection data fusion is proposed.By adjusting the voltage and current of the X-ray source,the projection data of both high and low energy spectrum are collected at each sampling view.As the beam hardening problem caused by the wide energy spectrum of X-ray source has a serious impact on the reconstructed images of the metal workpiece,the projection estimation model of narrow energy bin with a wide energy spectrum is established by combining the measurement data of each energy spectrum.Thus,the nearly monochromatic-energy projection of a narrow energy bin is obtained through optimization.Then,the projection fusion algorithm based on scale transform and residual smoothing model is proposed to fuse the projection of narrow energy bin of both high and low energy spectrum,so as to obtain more complete projection information.Finally,the conventional reconstruction method is utilized to reconstruct the sectional images.Compared with the traditional CT reconstruction method with fixed energy scanning mode,the proposed scheme can effectively extend the dynamic range of CT imaging system without requiring to change the hardware of CT equipment,which can improve the structural integrity of reconstructed images and suppress beam hardening artifacts effectively.