Sparse-view Helical CT Reconstruction And Metal Artifact Correction

Computed Tomography(CT)is widely used in medical clinical screening and diagnosis for the non-invasive inspection and high spatio-temporal resolution However,medical CT examinations still suffer from high radiation doses and metal artifacts.At present,there are two ways to reduce the radiation dose of CT scan.The first one is to minimize the milliamp-seconds(mAs)by reducing the tube current or shortening the scan time.However,due to insufficient detection of low-mAs,direct analytical reconstruction can easily lead to obvious noise and artifacts in the reconstructed image.The seconds is to use sparse-view scanning to reduce the number of exposures for one rotation of the X-ray tube.However,the number of projection in the sparse-view scan does not satisfied with the Nyquist sampling theorem.Which can easily cause stripe artifacts in the analytical reconstruction.In addition,during the CT scans of containing metallic materials,the high attenuation of the metallic materials and the multispectral characteristics of the X-rays degrade the projection data is the main reason of generating metallic artifacts.Directly analytical reconstruction of metal corrupted projection data can lead to bright and dark streak artifacts in the reconstructed image.If the above problems are not processed through technology,noise and artifacts in the image will inevitably affect the clinical diagnosis.Therefore,the author proposes a sparse-view CT iterative reconstruction method based on TTGV-POCS and a metal artifact correction method based on deep learning.(1)To solve the problem of sparse-view helical CT reconstruction,most of the iterative reconstruction algorithms are based on the single-layer rebining processing.The projection data is reformed into 2D projection data,then the single-layer image is reconstructed,which doesn’t take the volume data characteristics for consideration.In this paper we present an iterative reconstruction model based on the TTGV-POCS algorithm.We consider 3D volume data as a tensor,and use the high-order derivative properties of Tensor Total Generalized Variation(TTGV)to characterize the sparseness and correlation of 3D volume data in the adjacent layer,also the projections onto convex sets(POCS)iterative reconstruction framework is incorporated into the sparse-view helical CT iterative reconstruction.To evaluate the effectiveness of the proposed TTGV-POCS model,we used clinical data and XCAT digital phantom simulation data for the comparative experimental.From quantitative and qualitative analysis,it can be seen that TTGV-POCS can effectively suppress artifacts and noise introduced by sparse-view scans,and maintain the edge details.(2)To solve the problem of metal artifacts,we propose a deep learning-based metal artifact removal network model(VVBP-SinoCNN).The network model takes VVBP-tensor data as the input of convolutional neural network to improve network feature extraction capabilities by taking advantage of the similar structure and sparse tensor of view-by-view back-projection tensor(VVBP-tensor).In order to suppress the streak artifacts,we use the sinogram data and VVBP-tensor data as the input of the network at the same time,and connect the two sub-networks by filtering and the view-by-view back-projection layer.Experiments show that the VVBP-SinoCNN method is superior to NMAR,CNNMAR and other contrast methods in terms of metal artifact removal and image detail preservation,and also improves the clinical diagnostic value of CT images.