PET Image Reconstruction From Low Dose Sinogram Based On Teacher-student Network

Positron emission tomography is the most advanced non-invasive imaging technology in the field of nuclear medicine.It can image the metabolism of organs and tissues in human body by injecting radioactive tracer drugs into human body.The injection of radioactive tracer drugs has two main shortcomings,on the one hand,because the radioactive tracer drugs have a certain degree of radioactivity,the patients and staff produce nuclear radiation;on the other hand,because of the high cost of drugs,it brings economic burden to patients.In order to reduce the risk and cost of PET imaging system,it is necessary to study low-dose PET image reconstruction algorithm.Low dose PET image reconstruction algorithm is a challenging problem,most of the previous research work focused on the pre-processing of sinogram domain,post-processing of image domain,adding prior constraints to the reconstruction algorithm and so on.However,these solutions separate the sinogram domain from the image domain and ignore the relationship between the two domains,which leads to the low quality of the final reconstruction.In this paper,the teacher student deep learning network is used to link the projection domain denoising with PET image reconstruction task,and an end-to-end model of PET image reconstruction from low-dose projection domain data is realized.For the denoising algorithm in projection domain,considering the particularity of data in projection domain,this paper proposes a bilateral pyramid convolution denoising algorithm in projection domain based on spatial attention mechanism.In the process of image reconstruction,we combine the projection domain denoising and image domain reconstruction through the teacher student network,and realizes the mutual supervision and promotion of the two modules.Finally,this paper uses simulation data and clinical patient data to verify the proposed algorithm,and achieves ideal results.It is hoped that the work of this paper can provide a new solution for low-dose PET image reconstruction.