Regional Feature Preserving Based Low-Dose CT Precision Imaging

X-ray computed tomography(CT)has been established as a fundamental modality and widely used in many clinical applications because it has the capability to provide high temporal-spatial resolution and contrast resolution images for diagnostic procedures and visualization to guide therapeutic procedures.On the other hand,the concern about the radiation exposure associated with CT examinations is also growing and the excessive X-ray radiation could lead to the development of cancer,leukemia,or some other hereditary disease.Therefore,many efforts have been made to reduce the radiation dose while maintaining the image quality by optimizing CT protocols based on the as low as reasonably achievable(ALARA)principle,which is goal for low-dose CT(LdCT)precision imaging in clinical applications.One simple and effective approach to reduce the radiation dose is to lower the X-ray tube current and/or shorten exposure time during the scans.The downside of this strategy is that the photon starvation resulting from the low-mAs scan makes the image reconstructed by conventional filtered back-projection(FBP)method dramatically degraded.The associated noise and noise-induced artifacts in the image will affect the accuracy of the following diagnosis in clinics.Therefore,robust low-dose CT image reconstruction for low-mAs is one of the hot research topics in CT research.Statistical iterative reconstruction(SIR)methods,which can incorporate both the statistical properties of the measurements and the prior knowledge of the desired image into the reconstruction procedure,have shown great potential in deriving the high-quality LdCT image.For the SIR methods,the regularization term can reflect the prior knowledge of the image and plays an important role in the successful image reconstruction.However,existing statistical iterative reconstruction methods often use simple image priors,such as the sparse features,which fail to fully characterize the essential features information in CT images.In this thesis,aiming to rationally incorporate the essential features of the CT images into the regularization term in SIR methods,three main contributions for LdCT image precision imaging are as follows:(1)Aiming to learn texture information from the previous normal-dose CT(NdCT)image and leverage it for follow-up LdCT image reconstruction to preserve textures and structure features,we propose the region aware texture preserving(RATP)prior for LdCT image reconstruction.Specifically,the proposed reconstruction method first learns the texture information from those patches with similar structures in NdCT image,and the similar patches can be clustered by searching context features efficiently from the surroundings of the current patch.Then it utilizes redundant texture information from similar patches as a priori knowledge to describe specific regions in the LdCT image.The main advantage of the PATP prior is that it can properly learn the texture features from available NdCT images and adaptively characterize the region-specific structures in the LdCT image.(2)In order to reasonably introduce a priori information,and at the same time better maintain the consistency of the regional structure,we propose the convolution sparse coding(CSC)based regularization for the SIR method in this thesis.The proposed method constructs the mapping relationship between the NdCT images and LdCT images by CSC,and then through the mapping relationship,the prior image of the current LdCT image can be obtained,which will be used to guide the following LdCT image reconstruction.The proposed method can effectively weaken the influence of operations such as image segmentation and registration on the reconstruction result.Moreover,the advantage of CSC is that it directly processes the whole image instead of the overlapped patches and it can exploit the image global correlation to produce the more robust reconstruction of local structures.The global consistency constraint in CSC makes the proposed method capable of preserving the consistency of the structures in LdCT image reconstruction.(3)Cerebral perfusion CT(CPCT)can provide rapid,high-resolution,quantitative hemodynamic maps to assess and stratify perfusion in patients with acute stroke symptoms.Due to the repetitive scans,the concerns regarding high radiation dose in CPCT imaging are growing,and low-dose CPCT imaging is necessary and beneficial in clinical.One of the most challenging problems for low-dose CPCT imaging is how to improve the image quality to calculate the accurate hemodynamic maps from such noisy images.In this thesis,we propose a reweighted laplacian based CPCT image restoration method by depicting the structure sparsity inside of the images.Moreover,the maximum marginal likelihood estimation and expectation maximum(EM)algorithm are employed to learn the optimal parameters from the images,which makes the proposed algorithm more flexible for different data.These advantages together improve the restoration performance from the low-dose CPCT image obviously.The experimental results show that the image reconstruction and image restoration methods proposed in this paper are feasible and effective in improving the quality of low-dose CT images.