Dynamic PET Reconstruction Based On Rebinned Images Guidance

Positron emission tomography(PET)is a powerful molecular imaging technology that can monitor biochemical reactions in organisms by injecting radiotracers into organisms,and plays an important role in clinical diagnosis and new drug development.During dynamic scanning,PET can obtain multiple consecutive time frame data to reconstruct multiple images.In order to monitor the rapid changes in the distribution of tracers in the body,the scanning time of each time frame is usually relatively short,resulting in each time frame.The counts of time frames are all low and the quality of the reconstructed images is poor.Images reconstructed from low-count PET projection data usually contain a lot of noise.Since the inverse problem is ill-posed,improving the quality of PET reconstructed images has been facing many challenges.Currently,there are two ways to improve the quality of PET images: one is by changing the algorithm of image reconstruction,and the other is by post-processing the reconstructed image.The research done in this paper is mainly to study and improve the reconstruction algorithm of PET images.The work carried out is as follows:1.We propose dynamic PET image reconstruction based on a maximum likelihood expectation maximization algorithm and guided kernel filtering.The method uses the combined image reconstructed from the composite frame data as the prior image and then obtains the kernel function needed to construct the guided kernel filter.We estimated the characteristic coefficients of the PET projection data using a maximum likelihood expectation maximization algorithm,using a kernel function to filter the estimated coefficients obtained at each iteration.In order to verify the feasibility of the algorithm,we carried out simulation experiments,and the comparison algorithms are the maximum likelihood expectation maximization algorithm and the kernel method respectively.Experimental results show that the proposed algorithm can reduce the noise of reconstructed images and obtain better tumor reconstruction.2.We propose a dynamic PET imaging method based on dual texture features.The method also uses the composite frame data to reconstruct the combined image and its texture feature image as the prior image and then obtains the prior feature.The acquired prior features are formed into a new kernel function,and then the application of texture features in PET reconstruction is realized with the help of the framework of the kernel method.In order to verify the feasibility of the algorithm,we conducted simulation experiments and clinical experiments.The comparison algorithms are the maximum likelihood expectation maximization algorithm,the kernel method and the new kernel method proposed by Gao recently.The experimental results show that the proposed algorithm can significantly improve the contrast recovery coefficient of the reconstructed image tumor region,and can improve the signal-to-noise ratio of the reconstructed image.