Study For PET Parametric Imaging Algorithm Based On Patlak Model

Positron Emission Tomography(PET)is a functional imaging method in clinical medicine.Based on the principle that different tissues of organisms have different intakes of radioactive tracers,it realizes visual imaging of biological processes in vivo,and records the changes of intake and distribution of radiopharmaceuticals in different structures or regions of interest with time.Therefore,PET has been widely used in oncology,myocardial perfusion imaging and brain imaging.However,due to the low sensitivity,high physical noise and low dose of radioactive tracer,there is still room for improvement in the spatial resolution and reconstructed image quality of PET system.Among the existing PET reconstruction algorithms,the Filtered Back Projection(FBP)algorithm,which was originally applied to Computed Tomography(CT),has a fast imaging speed,but this algorithm model ignores the influence of noise on imaging.However,the Maximum-Likelihood Expectation-Maximization(MLEM)algorithm adds the statistical characteristics of data into the mathematical model,and the imaging quality has been greatly improved.Therefore,this algorithm has been widely studied and applied in scientific research and clinic.However,MLEM algorithm is affected by statistical noise,and too many iterations will reduce the image quality instead of increasing.In order to improve the Signal-to-noise Ratio,SNR)of images,some scholars have proposed two algorithms: Bayesian or Maximum A Posterior,MAP)algorithm,and Kernel method.Both algorithms build a system model containing prior or posterior images on the original basis,and regularize the reconstruction process by using the high signal-to-noise ratio of prior/posterior images to improve the final image quality.These algorithms can be applied to PET parametric imaging.Parameter imaging is essentially a process of fitting dynamic imaging and dynamic model.The linear irreversible Patlak model,as one of the most commonly used dynamic models,can represent the blood transmission rate and other parameters in the organism,and effectively extract the physiological information contained in the dynamic images.Therefore,based on this model,the following two researches related to PET parametric imaging algorithms have been carried out in this paper: Firstly,Guided Kernel Means(GKM)is proposed to be applied to nuclear medicine image processing,and based on the prior information of dynamic PET images,the speed and accuracy of parametric imaging have been greatly improved.The quantitative analysis of the experiment shows that compared with the original algorithm,the proposed algorithm has great advantages in improving the quantization level of parametric images.Secondly,a PET parameter imaging algorithm based on prior texture information and intensity information of images is proposed.In the new imaging algorithm,the texture information in frequency domain and the intensity information in spatial domain of anatomical images are combined as a new prior,and parametric image reconstruction is carried out based on Kernel algorithm.Experiments show that the texture information of prior images can effectively improve the richness of prior information,and thus play a good role in improving the quality of PET parametric imaging.