Research Of Iterative CT Image Reconstruction Using Limited Projection Data

Since CT has been applied extensively in clinical exams,radiation is a big issue.Too much dose of x-ray delivered to patients may cause risk of inducing cancer.Increasing sampling rate of projection data can achieve the goal of low dose imaging,that is to say,collecting few-view projection data in one circle scanning.However,in this case,the loss of projection data does not meet the accurate reconstruction of FBP anymore.Meanwhile iterative reconstruction may show its potential and advantage.Recently,Sidky proposed accurate ART-TV iterative reconstruction.However,when the number of projection views is less or the level of noise in projection data is high,stair artifacts appear in CT image.Non-local Means filter,NLM,has been used in sparse view CT image reconstruction and was proposed in image denoising field.Huang put NLM into sparse view CT image reconstruction and got better results than ART-TV method.Based on NLM,researchers extended its application in image reconstruction.Compared to sparse view scanning and reconstruction,limited angle imaging is an effective strategy of low dose imaging.Due to the loss of projection data from other views,as the case of sparse views,FBP method leads to CT images with severe artifacts,and these artifacts distribute not evenly in the whole image.Iterative reconstruction can reconstruct better image than FPB algorithm,but still need improvement.In addition,the performance of NLM on limited angle CT reconstruction has not been seen in existing reports yet.This study focuses on low dose CT image reconstruction and the research in this paper can be concluded as following four aspects:First,a fast and high-quality FBP algorithm is proposed.In full-view CT,the projection data can be divided into two groups,one is zero data standing for going across air and the other one for going across object.Thus,for each to-be-reconstructed pixel when performing backprojection process,if the corresponding projection data at one view is zero,set the pixel as zeros and process the next pixel instead of continue processing this pixel from all other views.Unnecessary backprojecion process is avoided to speed up reconstruction,meanwhile,the pixels in air area in reconstructed image are set to zero accurately,facilitating edge extraction of object.Experimental results show that the proposed method is faster and more accurate than conventional FBP method.In limited angle CT,FBP reconstructs image with severe and local artifacts.If the outer contour of a reconstructed object is roughly axis-symmetrical and the structures surrounding object's outer contour are also roughly axis-symmetrical,the FBP image after air correction described above is used to obtain the object's outer.Then it is possible to remove most of the artifacts by using the information from some quadrants in the image.The processed image is then as the initial image for ART-TV iterative reconstruction.Experimental results show the new initial image proposed by us can help accelerate convergence.Second,a modified ART-NLM is proposed in limited angle CT reconstruction.Similar to ART-TV,conventional ART-NLM produces image with severe local artifacts,but using the information sorrounding artifacts to remove artifacts is not reasonable.To overcome this,after preliminary experiments,the artifacts are found their locations(four quadrants)in the image.If the pixel belongs to artifacts quadrant,the symmetrical quadrant's information is used to recover the pixel by using NLM.If not,the information near the pixel is used to 'filter the pixel by using NLM.This suppresses the artifacts a lot.However,if the information is not perfectly symmetrical,fake information may appear in the artifacts area.Thus after a number of iterations of ART-NLM,the ART-TV is used to solve the problem of fake information.Experiments show that the proposed ART-NLM/TV can effectively suppress artifacts and reconstruct better images compared to existing methods.Third,a modified ART-ATpV is proposed in sparse view CT reconstruction.Although TV minimization can remove noise and preserve edges well,it may lead to stair and block artifacts when the number of views is not enough or the noise level in projection data is high.This study proposes adaptive TpV used for iterative sparse view CT reconstruction.Conventional TV is the L1 norm of gradient image,and proposed algorithm is Lp norm of gradient image.A p value is determined for each pixel according to its property(on edge or in even region).When pixel is in even region,a p close to 1 is used to remove noise and suppress artifacts.When pixel is on edge region,a p close to 0 is used to preserve edges details.Experiments demonstrate the propose method on removing noise,artifacts and preserve edge details to large extent.Fourth,a modified ART-TV/SLO is proposed in sparse view CT reconstruction.Traditional ART-TV has steps of ART reconstruction,positivity constraint and TV minimization.TV minimization is implemented by steepest decent algorithm in a number of iterations.However,less or more iterations lead to suboptimal denoising performance or oversmoothing effect of structures.This study proposes a method based on combination of TV and smooth LO norm for sparse view CT reconstruction.In the new method,after operating ART and positivity constraint,and iterating TV minimization for a few iterations,SLO minimization is used to further denoise image and reduce artifacts.Sparser SLO than TV has better balance between noise removing and artifacts suppression.Data experiments show the validity of proposed method.