PET Image Denoising Based On Non-local Low Rank Matrix Approximation

Positron emission tomography(PET)is an effective molecular imaging method that allows non-invasive quantitative observation of physiological and biochemical changes in living organisms.PET image is reconstructed via sinogram which is often affected by various external disturbances and the total counts during the reconstruction,so the reconstructed image has low signal-to-noise ratio and low spatial resolution.In order to remove the noises and improve the spatial resolution of PET images,a denoising scheme based on nonlocal low rank matrix approximation and its improved method are proposed.The main research of this paper are as follows:1.An improved PET image denoising scheme based on low rank matrix approximation is proposed which exploits the advantage of the high redundancy of PET images.The proposed method first performs pre-filtering on the noisy image,selects similar image blocks and builds a block matrix;performs low-rank matrix approximation to remove the noise in the block matrix;then restores the denoised matrix to a complete image;after iterative operation,the final denoised image is obtained.Experimental results show that the proposed method has a good denoising effect on PET images.2.The above denoising method can effectively denoise the PET image,but the method has the problems of residual noise and over-smoothing.To further improve the result,a combination of non-local low-rank approximation denoising method and guided image filtering are proposed.From the perspective of whether there is a prior information: when there is no prior information,the last few frames of denoised PET images with greater contrast are used as the guide images,and the characteristics of more detailed information of multi-images are used to improve the denoising result;when there is a prior information,the pre-processed MRI is used as the guide image to improve the denoising result.Experimental results show that both of proposed methods can effectively reduce the residual noise and enhance the texture details of the image.