Dynamic PET Images Denoising Based On Spectral Graph Wavelet Transform

Positron emission tomography(PET)is a nuclear medicine imaging technique,which can detect molecular-level activity in the tissue by injecting radioactive tracers.PET is widely used in early warning and diagnosis of major diseases such as cardiovascular and tumor.PET image has the characteristics of low resolution and low signal to noise ratio,which bring a relatively big obstacle to the subsequent processing of PET image.Therefore,PET images denoising has important clinical significance.In this paper,three different dynamic PET images denoising algorithms are proposed by applying Graph Wavelet Transform(GWT)in the two cases of no prior information and known MRI(Magnetic Resonance Imaging)prior information.The main research works are as follows:1.A GWT-based denoising algorithm for dynamic PET images with composite image is proposed.Firstly,we use the PET composite image to construct the graph adjacency matrix.Then,we perform multi-scale graph wavelet decomposition on the noisy PET image.Finally,the denoised PET image is obtained by inversing GWT on the low-frequency graph wavelet coefficients.The simulation results show that proposed approach has better performance in objective evaluation indicators and visual effect.2.In the case of known prior information of MRI,this paper proposes a GWT-based denoising algorithm for PET images by incorporating MRI knowledge.Firstly,we perform image fusion of MR image and PET composite image through hard threshold.Then,the graph adjacency matrix is constructed based on the fusion image.Finally,we apply GWT to the dynamic PET images for denoising.Experimental results indicate that the proposed algorithm further improves the denoising performance.3.The image fusion method based on hard threshold only considers the local features of the lesions in the PET composite image.In order to preserve the three features of lesion,edge and structure at the same time,we propose a graph wavelet denoising algorithm based on multi-feature fusion.This method combines the PET composite image,the PET composite image gradient and the MR image to form a three-channel image with equal weights,and uses the three-channel image to construct the graph adjacency matrix.The simulation results demonstrate that the proposed algorithm has obvious denoising advantages in the homogeneous area of the PET image.