Simultaneous Estimation Of Image Parameters For Dynamic PET

Positron emission tomography(PET)has been one of the outstanding representatives of the current function molecular imaging technology.Through an injection of labeled radio tracer,PET makes it possible to image the concentration in living tissue and reveal the abnormal metabolism in the molecular level,and therefore provide an effective way for early diagnosis.Compared with static PET,which can only provide the activity distribution within a certain period of time,dynamic PET can be responsible for a much more complicate research task due to its ability to monitor the change of activity distribution and a suitable tracer kinetic modelling process,thus making PET imaging techniques significantly valuable in science research and clinical diagnosis.In this paper,we mainly focused on the activity reconstruction and kinetic analysis in dynamic PET and the contribution is listed as follows.1.Because of the low signal-to-noise ratio(SNR)at each time measuring,failing to incorporate other information into the reconstruction framework may lead to a blurred result of the activity distributions which would further increase the noise in the indirect reconstruction scheme.In this paper,we present a joint estimation framework to reconstruct the temporal sequences of dynamic PET images and the associated parametric images of tracer kinetics.This algorithm,which combines the statistical data measurement and tracer kinetic models,integrates a sparsity regularized dictionary into a total variational(TV)minimization based algorithm for simultaneous reconstruction of the activity distribution and parametric map directly from measurement emission sinograms.2.Usually in dynamic reconstruction problems,image segmentation,activity reconstruction and estimation for kinetic parameters are considered as three independent problems and the accuracy of segmentation is closely connected to the image quality from previous steps.In this paper,we then introduced a joint reconstruction framework that can achieve image segmentation,reconstruction and estimation for tracer kinetics at the same time.During one iteration process,each sub-problem is updated based on the current expectation of other variables and achieve a simultaneous results for all the problems.When we decided to convolve reconstruction and segmentation into a joint optimization problem,it is because that the raw sinogram data can therefore be well modeled by Poisson statistics while segmentation also enforce homogeneity for each region.3.Inspired by the achievements of deep neural network in monaural speech separation,we tried to extract the time-activity curve(TAC)from dual tracer PET imaging and separate the mixed signal into two TACs and therefore fulfill the task for dual tracer PET reconstruction.