MRI Image Denoising Using Non-Local Means (NLM) Filter for Rician Noise Suppression

Abstract

Magnetic Resonance Imaging (MRI) plays a vital role in medical diagnostics by providing high-resolution visualizations of internal organs and tissues. However, MRI images are frequently corrupted by noise, particularly Rician noise, which significantly affects diagnostic accuracy. Traditional denoising techniques such as Gaussian filters, wavelets, PDE-based methods and anisotropic diffusion often blur fine details or fail to effectively suppress Rician noise. This study explores the use of the Non-Local Means (NLM) filter, which exploits image redundancy by averaging similar patches across the entire image, thereby preserving edges and structural details while effectively suppressing noise. A hybrid framework is proposed, combining median filtering for preprocessing, histogram equalization for contrast enhancement and an unbiased NLM (UNLM) modification to address the bias effect of Rician noise. Experimental validation using both simulated and real MRI datasets (T1, T2 and PD images) shows that the proposed method significantly improves image quality. Evaluations using PSNR, RMSE and SSIM metrics confirm that the NLM filter outperforms conventional methods, especially at higher noise levels, ensuring sharper edges, improved contrast and enhanced clinical interpretability.