Denoising MR images using non-local means filter with combined patch and pixel similarity.

Xinyuan Zhang,Wufan Chen,Jianhua Ma,Yanqiu Feng,Yikai Xu,Wei Yang,Guirong Hou,Bingquan Lin,Xi-Nian Zuo

doi:10.1371/journal.pone.0100240

Xinyuan Zhang, Wufan Chen + Show 7 more

Open Access

https://doi.org/10.1371/journal.pone.0100240

Copy DOI

Journal: PloS one	Publication Date: Jun 16, 2014
Citations: 68	License type: CC BY 4.0

Affiliation: Southern Medical University, Nanfang Hospital

Abstract

Denoising is critical for improving visual quality and reliability of associative quantitative analysis when magnetic resonance (MR) images are acquired with low signal-to-noise ratios. The classical non-local means (NLM) filter, which averages pixels weighted by the similarity of their neighborhoods, is adapted and demonstrated to effectively reduce Rician noise without affecting edge details in MR magnitude images. However, the Rician NLM (RNLM) filter usually blurs small high-contrast particle details which might be clinically relevant information. In this paper, we investigated the reason of this particle blurring problem and proposed a novel particle-preserving RNLM filter with combined patch and pixel (RNLM-CPP) similarity. The results of experiments on both synthetic and real MR data demonstrate that the proposed RNLM-CPP filter can preserve small high-contrast particle details better than the original RNLM filter while denoising MR images.

Highlights

Magnetic resonance (MR) images are usually acquired with low signal-to-noise ratios (SNRs) especially in the implementation of high temporal resolution or high spatial resolution imaging [1,2]
We demonstrated the particle blurring of the Rician NLM (RNLM) algorithm was caused by setting self-weight as the maximum weight of non-central pixels in the search window
The evaluation results on simulated and in vivo brain MR data showed that the proposed RNLM-CPP algorithm could preserve particles better than the original RNLM algorithm

Summary

Introduction

Magnetic resonance (MR) images are usually acquired with low signal-to-noise ratios (SNRs) especially in the implementation of high temporal resolution or high spatial resolution imaging [1,2]. Noise in the MR signal is mainly produced from the thermal vibrations of ions and electrons in the receiving coil and the sample [8], resulting in intensity fluctuations of MR images and serious degradation of some clinically useful image information. The following quantitative analysis of MR images through post-processing operations is often degraded by the noise. Reducing noise in MR images is essential and critical to improve image visualization and promote reliability of associative quantitative analysis

Methods

Results

Conclusion