Removal of correlated Rician noise in Magnetic Resonance Imaging

We propose a new method for Magnetic Resonance Imaging (MRI) restoration. Because MR magnitude images are corrupted by Rician distributed noise, these images suffer from a contrast-reducing signal-dependent bias. Also the noise is often assumed to be white, however a widely used acquisition technique to decrease the acquisition time gives rise to correlated noise. In this paper, we propose and motivate a two-step denoising procedure, where bias is removed from the squared magnitude image and denoising itself is then performed on the square root of this image in the wavelet domain. This denoising step takes into account noise correlation when distinguishing significant wavelet coefficients from insignificant ones. The estimated statistics of these two classes of wavelet coefficients are employed within a Bayesian estimator. The results show that the proposed technique is more efficient at removing correlated noise than existing MRI denoising techniques. The presented bias removal technique is shown to visibly improve contrast as well as to provide a large increase in PSNR.