Investigation Of Positron Emission Tomography Image Reconstruction

Positron Emission Tomography(PET)can perform molecular imaging of the metabolic activity of the human body.By detecting the gamma photons generated by the physical annihilation reaction of the radioactive tracer injected into the human body,the body and the lesion are displayed at the molecular level.Tissue cell metabolism,function,blood flow,cell proliferation,and receptor distribution provide more physiological and pathological diagnostic information for clinical applications such as cancer,heart disease,and neurological diseases.Image reconstruction from projection data has always been a research hotspot.This paper mainly discusses the reconstruction method of positron emission tomography.PET imaging process is to inject a certain dose of radiopharmaceutical tracer into the body’s veins.With the blood circulation all over the body,these radioactive particles release positrons during the decay process.The positrons meet the surrounding tissues of the human body.The annihilation reaction of negative electrons generates two photons in reverse motion.The photons penetrate the human body and are detected by the sensitive detector of the PET system surrounding the human body,and then restore the original radionuclides of specific organs in the human body according to the measured projection data.The location information of the image can indicate the prosperity of the metabolism of the organ,so that it can be identified whether the organ has a disease or even cancer.According to the characteristics and mathematical models of PET imaging,PET image reconstruction can be viewed as post-processing denoising problem,that is,removing noise and artifact data from image data containing noise and artifacts,thereby restoring clean activity Distribute the image.Based on this,a generative adversarial network is used to construct an algorithm framework to reduce streak artifacts and improve PET image quality.The training generator directly generates a residual PET image map(Residual PET Map,RPM)for image representation,rather than directly generating a PET image.Two discriminators are used to enhance the similarity to real PET images and RPM.Secondly,in order to obtain more and more effective receptive field information in the process of image processing,we designed residual dense connections,and then used residual dense nested with pixel periodic shuffling operations(Residual Dense Connections followed with Pixel Shuffle Operations,RDPS)to encourage feature map reuse and prevent loss of image resolution.Compared with other methods,the quantitative results show that the proposed method can obtain better performance in the bias-variance trade off,and significantly improve the signalto-noise ratio of PET activity images,and provide strong support for clinical imaging diagnosis and new drug development.In practical applications,in order to obtain reliable image quality,iterative algorithms(IR)are usually required.The slow convergence rate of IR inevitably leads to an increase in reconstruction time.Recently,due to the excellent parallel computing capabilities of GPUs,deep learning(DL)has received lots of attention in the field of medical images(such as denoising,reconstruction,classification,and segmentation).For image reconstruction,the training process may produce image details that did not exist in the object,leading to the negative results of diagnosis.Considering the above two problems,the design algorithm incorporates the projection data and the image data by defining a data consistency layer(DC).Using the projection data provide more reasonable and effective constraints for the PET denoising reconstruction process,continuously correcting the image generated during the denoising process,and recovering high-quality PET images from FBP PET images with streak artifacts and high noise.The computer numerical results show that the proposed method can obtain reconstructed images with higher signal-to-noise ratio,which has certain reference value for further research on PET reconstruction methods.In summary,this paper has analyzed the data in the PET reconstruction process and designed the reconstruction algorithm based on its characteristics.Experiments prove that the proposed method can achieve excellent results.Among them,the proposal of the deep residual generation anti-network algorithm has re-examined the PET image data model from a new perspective,which provides an enlightening idea for the algorithm development;Reconstruction provides an effective and flexible model framework,and many existing reconstruction methods are slightly inferior.