Research On PET Image Reconstruction Algorithm Based On Under-sampling

Positron emission tomography(PET)imaging technique can display in vivo metabolism of organs and tissues.It is the most advanced nuclear medicine imaging technique,and has the characteristics of noninvasive diagnosis,so it is more and more popular among people.However,the PET imaging system is expensive and brings a certain economic burden to doctors and patients.To reduce the cost of PET scanning systems,image reconstruction algorithms for under-sampling data have been extensively studied.This paper mainly studies the application of maximum likelihood expectation maximization(MLEM)algorithm in PET undersampling data,and it is improved.The main tasks are as follows in this paper:Firstly,the problems of the MLEM algorithm for the reconstruction of PET unsampled data are studied.Tests using Emission activity brain phantom show that the MLEM algorithm results in increased noise once iteration reaches a certain point and produces incorrect estimation for under-sampling data.Secondly,total variation(TV)minimization regularization was nested in the MLEM method to suppress noise,which is called “MLEM-TV” algorithm.Simulation experiments were performed using Emission activity brain phantom and Shepp-Logan phantom and compared with MLEM algorithm.The experimental results show that the MLEM-TV algorithm can suppress the noise.However,the gradient operator of the TV cannot distinguish true structures and noise in the image;that is,the image reconstructed by the MLEM-TV algorithm may lose some fine features.Then,the feature extraction(FR)operation was applied to the MLEM-TV algorithm to recover the missing image structure information in the TV minimization operation,which is called "MLEM-TV-FR" algorithm.Simulation experiments were performed using Emission activity brain phantom and Shepp-Logan phantom,and physical experiments were also performed using NEMA phantom and mouse,and compared with the MLEM algorithm and the MLEM-TV algorithm.The experimental results verify that the MLEM-TV-FR algorithm performs better than MLEM and MLEM-TV algorithms in preserving the fine structure and suppressing artifacts and noise.In addition,this paper assesses the effect of parameter patch(size of image block)of FR operation on MLEM-TV-FR algorithm.The result shows that this algorithm works best when patch value is 7×7 to 15×15.Due to computational efficiency,in this experiment,the value of patch is 7×7.Finally,the MLEM algorithm,MLEM-TV algorithm and MLEM-TV-FR algorithm are quantitatively evaluated using the Peak Signal-to-Noise Ratio(PSNR)metric and the Structural Similarity Index(SSIM).Experimental results show that the MLEM-TV-FR algorithm can reconstruct high quality images for under-sampling data in PET imaging.