Rician noise removal in MR images using an adaptive trilateral filter

Noise removal in MR images has been challenging due to the unpleasant noise with a Rician distribution, which is signal dependent. Denoising of MR images is of importance for subsequent diagnoses and analyses, such as tissue classification, segmentation, and registration. We propose a post-acquisition denoising algorithm to adequately and adaptively remove the random fluctuations and bias introduced by Rician noise. The proposed filter consists of geometric, radiometric, and median-metric components that replaces the intensity value with an weighted average between neighboring pixels associated with an entropy function. In addition, a parameter automation mechanism is proposed to reduce the burden of laborious interventions through a fuzzy membership function, which adaptively responses to local intensity difference. Quantitatively and qualitatively experimental results indicate that this new filter outperformed several existing methods in providing greater noise reduction and clearer structure boundaries in a variety of MR images.