论文部分内容阅读
Purpose:Sparse-view computed tomography(CT)reconstruction without significantly degrading image quality is an important issue in medical imaging domain.Non-local means(NLM)based reconstruction methods from sparse projections have been studied for years,but often suffer from over-smoothness on edge information.To overcome this limitation,we present an adaptive NLM(ANLM)based iterative-correction algebraic reconstruction technique(ART)algorithm,named as ART-ANLM.Methods:The ART-ANLM algorithm contains two major steps:1)Utilize ART reconstruction to satisfy the consistency of the measurement and the positivity of the reconstructed image; 2)Update iterative image using ANLM filtering.For ANLM,a novel similarity metric that is rotationally invariant between any two patches is proposed.Any patch with similar structure but different orientation to the reference patch by proper rotation wins a relatively large weight to avoid over-smoothed image.Moreover,one parameter h in ANLM which controls the decay of the weights would change adaptively with the iteration number increases,and vary with the pixel position,while the parameter h in NLM is changeless during the whole iteration process.This is another way to avoid over-smoothness.The method is validated on Shepp-Logan phantom and real head phantom data.Results:In our experiments,the searching neighborhood size is set 11×11 and the iteration number is set 100.For the simulated case,the ART-ANLM produces higher SNR(31.67dB>22.98dB)and lower MAE(0.0005<0.0021)reconstructed image than ART-NLM.The experimental results have demonstrated that the proposed method could significantly improve the reconstructed image quality and achieve a better compromise between artifacts or noise removing and image edge preserving.Conclusions:An adaptive NLM based reconstruction method for sparse-view CT is proposed.Compared to the conventional ART-NLM method,the SNR and MAE from ART-ANLM increases 38%and decreases 76%,respectively.