Collateral filtering of magnetic resonance images

Denoising of magnetic resonance (MR) images is of importance for clinical diagnosis and computerized analysis, such as tissue classification, segmentation, and registration. It is well known that the noise in MR magnitude images obeys a Rician distribution, which is signal-dependent. As a consequence, separating signal from noise in those images is particularly difficult. We propose a post-acquisition denoising method called collateral filtering to adaptively remove the random fluctuations and bias introduced by Rician noise. It replaces the intensity value on each pixel with an average value weighted by the geometric, radiometric, and medianmetric components between neighboring pixels associated with an entropy function. The experimental results indicate that the collateral filter outperformed several existing methods in providing greater noise reduction and clearer structure boundaries both quantitatively and qualitatively.