Improving SNR in Susceptibility Weighted Imaging by a NLM-based denoising scheme

Abstract

The combination of magnitude and phase information inherent in Susceptibility-Weighted Imaging (SWI) greatly benefits from high-resolution MRI acquisitions. The application of a denoising filter to produce SWI images with higher signal-to-noise ratio (SNR) while preserving small structures from excessive blurring is therefore extremely desirable, but non-trivial, as the distribution of magnitude and phase noise may introduce biases during image restoration. Here we present a new dedicated noise removal algorithm based on the Non-Local Means (NLM) filter and compare its results with the original SWI and 'standard' NLM-denoised human brain images. Both the visual assessment by two expert readers and the quantitative evaluation of the contrast changes of the voxel intensities demonstrated that the images restored with the proposed algorithm fared consistently better than the other two schemes, showing that a proper handling of noise in the complex MRI dataset may lead to visible improvements of the overall SWI quality.