Denoising Magnetic Resonance Images Using Collaborative Non-Local Means

Geng Chen,Pei Zhang,Yafeng Wu,Dinggang Shen,Pew-Thian Yap

DOI: https://doi.org/10.1016/j.neucom.2015.11.031

IF: 6

2016-01-01

Neurocomputing

Abstract:Noise artifacts in magnetic resonance (MR) images increase the complexity of image processing workflows and decrease the reliability of inferences drawn from the images. It is thus often desirable to remove such artifacts beforehand for more robust and effective quantitative analysis. It is important to preserve the integrity of relevant image information while removing noise in MR images. A variety of approaches have been developed for this purpose, and the non-local means (NLM) filter has been shown to be able to achieve state-of-the-art denoising performance. For effective denoising, NLM relies heavily on the existence of repeating structural patterns, which however might not always be present within a single image. This is especially true when one considers the fact that the human brain is complex and contains a lot of unique structures. In this paper we propose to leverage the repeating structures from multiple images to collaboratively denoise an image. The underlying assumption is that it is more likely to find repeating structures from multiple scans than from a single scan. Specifically, to denoise a target image, multiple images, which may be acquired from different subjects, are spatially aligned to the target image, and an NLM-like block matching is performed on these aligned images with the target image as the reference. This will significantly increase the number of matching structures and thus boost the denoising performance. Experiments on both synthetic and real data show that the proposed approach, collaborative non-local means (CNLM), outperforms the classic NLM and yields results with markedly improved structural details.