CT Image Multi-material Decomposition

In the theory of CT image multi-material decomposition,the mass attenuation coefficient of a mixture is approximated as a linear combination of multiple material bases.The multi-material decomposition technique(MMD)decomposes the original CT image into multiple material bases by taking into account of both material spatial distribution and composition to enhance the material distinguishment and characteristic.MMD is a promising technique to achieve quantitative CT imaging.The differentiated materials can be used in clinical applications including lesion delineation,digital subtraction angiography,virtual monoenergetic synthesis,virtual non-contrast(VNC)imaging,liver fibrosis quantification,etc.The MMD technique is limited by the restricted decomposed material number,the severe decomposition noise and inferior decomposition accuracy due to the ill-posed character and the noise magnitude of the material decomposition process.To solve the above problems,this research focuses on establishing a generalized MMD model by using the sparse material constraint and piecewise constant property to achieve noise suppressing and accuracy enhancement in the decomposition.This article conducts indepth research in four aspects in the following:1.We propose an MMD method for dual-energy CT using edge-preserving regularization.The method applies penalized weighted least-square reconstruction with a negative log-likelihood term and edge-preserving regularization for each material.The triple material is enumerated in the decomposition to find the optimal material composition.The optimization transfer principle is applied to design the pixel-wise separable quadratic surrogates to monotonically decrease the objective function.In the phantom and clinical data study,the results show that the proposed method achieves 84.0% noise standard deviation(STD)suppression and 6.85% decomposition accuracy enhancement compared with the direct inversion method.2.We propose an MMD method for dual-energy CT using the block-matching technique.The edge-preserving regularization method achieves noise suppression and spatial resolution maintenance.Nevertheless,this method is challenged by the non-convex property of the objective function due to the enumerating operation.Block-matching technique is introduced into the decomposition model to realize the cluster the pixels with high similarity.The grouped pixels are pre-defined the material composition and the material decomposition process is performed on the selected material composition.The decomposition accuracy is enhanced due to the introduction of clustered pixels and material composition.The method is evaluated on the phantoms.The results show that the proposed method achieves 22.4% decomposition accuracy enhancement compared with the direct inversion method and edge-preserving method.Additionally,the decomposed cross contamination is suppressed and the diagonality of the normalized cross-correlation matrix is increased by 53.9%.3.We propose an MMD method based on single energy CT.The image domain dualenergy CT MMD method applies triple material assumption to approximate the material sparsity.In the clinical practice,some cases only need to decompose two material bases.In this part,we decrease the number of energy spectrum to one to study the material decomposition in single energy CT.Two-material assumption and volume conservation are introduced into the single energy MMD model integrated with edge-preserving regularization.The method is evaluated using phantom and clinical data.The results show that the proposed method achieves 81.51% STD suppression and 15.56% decomposition accuracy enhancement compared with the direct inversion method.In the clinical study,the virtual nonenhanced image synthesized by the proposed method achieves the root-meansquared-relative error of 2.93% compared with the contrast-free ground-truth image.4.We propose an MMD method for single energy CT using L0 norm.In the MMD process,triple material or double material assumptions are applied to approximate the material sparsity.L0 norm,which counts the number of non-zero elements in one vector,is straightforward and precise to represent material sparsity.We apply L0 norm directly on the material volume fraction as a regularization in the decomposition objective function.The introduction of L0 norm leads to the nonconvexity of the objective function since solving L0 norm is NP-hard.An accelerated primal-dual algorithm is applied to solve the problem.The method is evaluated in the phantom and clinical data.In the phantom study,the method achieves a high decomposition accuracy of 98.5%.In the clinical study,the calcification area can be clearly visualized in the virtual non-contrast image generated by the proposed method,and has a similar shape to that in the groundtruth contrast-free CT image.This research introduces the material sparsity and material piecewise constant property into the MMD process and establishes a generalized MMD model to realize accurate MMD with high image quality.Each part of the decomposition model is discussed with circumstance.In the dual-energy CT study,this research applies edgepreserving regularization and block-matching technique to represent piecewise constant property and material sparsity.The results show that the proposed method achieves 84.0% noise standard deviation suppression and 6.85% decomposition accuracy enhancement compared with the direct inversion method.In the single energy CT study,this research applies edge-preserving regularization and L0 norm to represent piecewise constant property and material sparsity.The results show that the proposed method achieves 83.3% noise standard deviation suppression and 11.3% decomposition accuracy enhancement compared with the direct inversion method.The method is initially evaluated in virtual non-enhanced imaging in the clinic.