An Experimental Study on MRI Denoising with Existing Image Denoising Methods.

In this paper, we perform a systematical study on existing 2D denoising methods for reducing the noise in magnetic resonance imaging (MRI). We conduct experiments on six MRI images with the following denoising methods: wiener2, wavelet-based denoising, bivariate shrinkage (BivShrink), SURELET, Non-local Means (NLM), block matching and 3D filtering (BM3D), denoising convolutional neural networks (DnCNN) and weighted nuclear norm minimization (WNNM). Based on our experiments, the BM3D and the WNNM are the best two methods for MRI image denoising. Nevertheless, the WNNM is the slowest in term of CPU computational time. As a result, it is preferable to choose the BM3D for MRI denoising.