Rician Noise Removal Based On Adaptive Trainable Nonlinear Diffusion Model

Rician noise reduction is an essential issue in magnetic resonance imaging(M-RI).Recently,learning based methods have achieved great success for image restora-tion problems,which provide fast inference and good performance.One limitation of these methods,however,is that the training procedure is noise-level dependent,i.e.the trained models are usually bound to a specific noise level and lack the ability to adapt to different noise levels,for different levels of noise,different models need to be trained.In this paper,we mainly study the model which can automatically adapt to different levels of Rician noise.Based on the denoising model of the field of experts(FoE)prior image and the trainable nonlinear reaction diffusion method(TNRD),a noise adaptive function is introduced to describe the noise level,so that it can automatically adapt to different noise levels and ensure that the model can effectively deal with Rician noise of different noise levels.We call this model the noise adaptive trainable nonlinear reaction diffusion model(A-TNRD).The realization of the model is to learn the parameters which are suitable for different levels noise reduction through the supervised training process,so as to achieve the purpose of noise removal.The proposed method is applied to T1-,T2-,and PD-weighted MRI dataset.The experimental results show that the proposed method can achieve superior performance compared with other methods in terms of both the peak signal to noise ratio and the structure similarity index.In addition,in order to show the wide application of the A-TNRD method,it can also be applied to deal with denoising problems with other noise distribution,such as Gaussian noise and Laplace noise,which can also achieve good denoising results.