Research Of PET Image Reconstruction Iterative Algorithm

Positron emission tomography(PET)has attracted more and more attention because of its high sensitivity,good specificity,good whole-body imaging and safety.It is also the successor of X-ray computed tomography(X-CT)and magnetic resonance imaging(MRI)imaging technology needs high technology and high cost,which is also the most advanced and important imaging technology in the medical field.However,due to the limitation of the hardware and software of pet sampling system and the interaction of gamma photons in the process of in transmission,the pet projection data collected often contains a lot of noise,which will affect the quality of PET image reconstruction.It is known that when the reconstructed image contains noise or the quality of the reconstructed PE T image is not high,it will affect the diagnosis of the follow-up doctors.Therefore,how to reconstruct the high-quality image through the design of reconstruction algorithm in the clinical application of PET imaging technology is also an extremely impor tant and valuable step.At present,PET image reconstruction algorithm has some main problems: it cannot accurately identify the edge of the image,over rely on the degree of agreement between anatomical structure information and functional information,and cannot better balance the image smoothness and image detail information.This paper studies and improves the algorithm for the above problems.The main work is as follows:(1)The PET image reconstruction algorithm based on the patch-based regulation(patch)is implemented.In PET image reconstruction,the traditional non second penalty is to calculate the image roughness by using the intensity difference between adjacent pix els.Although the reconstructed image can retain the edge information,the edge of the image cannot be correctly determined only by calculating the intensity difference between adjacent pixels when the image contains noise.Using neighborhood block instead of a single pixel to calculate the image roughness,the edge of the image can be accurately recognized even if there is noise interference,and the degree of compliance between anatomical structure information and functional information will not be excess ively dependent.(2)A PET image reconstruction algorithm based on neighborhood block priori and Gaussian filter(patch-filter)is proposed.The image reconstructed by patch algorithm still contains noise and artifact.Referring to the experience of clinical treatment of this kind of problem,the image reconstructed by neighborhood block prior algorithm is processed by post gaussian filter to further optimize the image reconstructed by patch algorithm.(3)A PET image reconstruction algorithm based on neighborhood block priori and dictionary learning(patch-DL)is proposed.Although the patch-filter algorithm is proved to be better than the patch algorithm by experiments,the reconstructed image still contains noise and artifacts.If the noise is excessively reduced,the reconstructed image will lose some details.Therefore,this paper proposes a patch-DL method,which can balance the noise in the reconstructed image and the small structure information in the reconstructed image as much as possible,that is,th e reconstructed image will not be too smooth due to excessive noise reduction,and the detail information will not be retained less because of the noise,and the reconstructed image will not be poor quality sound judgment.In this paper,the performance of the PET image algorithm proposed in this paper is analyzed based on the simulated 2D brain PET data.Firstly,the quality of the reconstructed image is qualitatively analyzed,and then according to different evaluation indexes such as mean absolute error(MAE),correlation coefficient(CORR)and root mean square error(RMSE).The experimental results show that the patch-filter algorithm and the patch-DL algorithm proposed in this paper are effective.Compared with the existing PET image reconstruction algo rithm,the patch-DL algorithm can not only suppress the artifacts and noise,but also retain the small structure information in the image.