| CT(Computed Tomography)technology can detect the internal structure of the measured object without damage.CT has developed rapidly since its invention,and has been widely used in medical,industrial,security and other fields,playing a crucial role.Due to technological limitations,in practical applications,various forms of images that originally did not exist in the scanned object but appeared on the image are called artifacts.The existence of artifacts will cause some structures in the reconstructed image to be blocked or the boundary blurred,and the spatial resolution of the reconstructed image will be greatly affected,thus affecting the detection and discrimination of the measured object.At the same time,the evaluation of CT system performance index also has a certain interference.Therefore,artifact correction plays an important role in improving the quality of reconstructed images,increasing the accuracy of diagnosis,and accurately measuring the performance indexes of CT equipment.In this paper,we mainly focuses on the correction of windmill artifacts,beam hardening artifacts,and metal artifacts in CT images.According to the characteristics of artifacts,the correction algorithms are designed.The specific work is as follows:(1)For spiral CT windmill artifacts,this paper uses dual-domain filtering to reduce windmill artifacts.Firstly,CT projection data are generated by simulation and the spiral CT images with windmill artifacts are reconstructed by the FDK algorithm.Then,the reduction effect of windmill artifacts is analyzed by using bilateral filtering kernel of different radius.Finally,the kernel radius is adjusted according to the distance transformation to reduce the windmill artifacts.Compared with other methods,the method proposed in this paper performs best in terms of structural similarity.The feasibility and effectiveness of the proposed method are verified by computer simulation experiments.The proposed method can retain the details of the scanned object while reducing windmill artifacts in the reconstruction.(2)For beam hardening artifacts,this paper is based on the mathematical physics model of X-ray imaging.First,for the case that the imaging object includes water like,bone like and metal like materials at the same time,the corresponding objective function is constructed using the data consistency condition;Using non convex optimization algorithms,determine the optimal scale factors for soft tissue,bone,and metal;Obtain the optimal bone correction method that can simultaneously correct cup shape,stripe,and metal artifacts.The experimental results obtained through computer simulation indicate that the proposed method can adaptively determine the optimal scaling factor and then correct various beam hardening artifacts in the reconstructed CT images.In particular,compared with the nonlinear least squares method before variable replacement,the running time of the new CT image reconstruction algorithm is reduced by 82.36%,and the residual error is reduced by55.95%.(3)For metal artifacts,based on the previous study of traditional methods,the method of deep learning is used to reduce the artifacts.At present,most metal artifact correction methods of deep learning are based on supervised learning;In practice,due to the inability to obtain paired data,the synthetic data is usually used for supervised learning.However,the synthetic artifact image may not truly reflect the properties of metal artifacts,and the trained model often has certain limitations.In this paper,the classic unsupervised learning network Cycle GAN is adopted,and the attention mechanism module is added to the generator.The attention mechanism is used to make the network pay more attention to metal artifacts and improve the ability of artifact correction.The LSGAN and Patch GAN are introduced to improve the stability of training and the quality of generated images.The effectiveness of this method is verified by experiments. |
