Noise reduction in diffusion MRI using non-local self-similar information in jointx−qspace

Abstract

Diffusion MRI affords valuable insights into white matter microstructures, but suffers from low signal-to-noise ratio (SNR), especially at high diffusion weighting (i.e., b-value). To avoid time-intensive repeated acquisition, post-processing algorithms are often used to reduce noise. Among existing methods, non-local means (NLM) has been shown to be particularly effective. However, most NLM algorithms for diffusion MRI focus on patch matching in the spatial domain (i.e., x-space) and disregard the fact that the data live in a combined 6D space covering both spatial domain and diffusion wavevector domain (i.e., q-space). This drawback leads to inaccurate patch matching in curved white matter structures and hence the in-ability to effectively use recurrent information for noise reduction. The goal of this paper is to overcome this limitation by extending NLM to the joint x – q space. Specifically, we define for each point in the x – q space a spherical patch from which we extract rotation-invariant features for patch matching. The ability to perform patch matching across q-samples allows patches from differentially orientated structures to be used for effective noise removal. Extensive experiments on synthetic, repeated-acquisition, and HCP data demonstrate that our method outperforms state-of-the-art methods, both qualitatively and quantitatively.

Multi-channel framelet denoising of diffusion-weighted images

“…For instance, collaborative NLM [35] extended the search volume to a number of co-denoising images to enrich the similar information used in noise reduction. Chen et al [36, 37] proposed to improve NLM by considering the similar information in both spatial domain and diffusion wavevector domain. This idea was further employed to improve atlas building [38] and resolution enhancement [39].…”

Denoising Diffusion MRI via Graph Total Variance in Spatioangular Domain

“…Both ANLM and MP-PCA are implemented by DIPY [33] (3) XQ nonlocal means (XQ-NLM): XQ-NLM [17] denoises the signal via weighted averaging of selfsimilar information, which is defined in the spatialangular domain. The parameters are set to the default values as suggested in [17] We transform the Rician signal to its Gaussian-distributed counterpart using Özarslan et al's method [34]. For a fair comparison, we estimate the nonstationary noise by MP-PCA for ANLM, XQ-NLM, and GTV.…”

“…Recently, Chen et al have proposed two novel NLM-based denoise methodologies in joint x − q space. Before NLM denoising, for each point in x − q space, they (1) defined a spherical patch from which they extracted the rotation-invariant features patch matching [ 17 ] and (2) performed graph framelet transforms to extract robust rotation-invariant features after encoding the q -space sampling domain using a graph [ 18 ]. Graphs have the ability to well modeling the data with irregular and complex structures [ 19 ].…”

The sensitivity of diffusion MRI to microstructural properties and experimental factors

“…In recent years, several methods have been proposed to mitigate the unfavorable effects of the non-stationary Rician and nc- χ noise. These include well-celebrated the non-local means framework that evaluates the similarity in terms of non-local patches ( Manjón et al, 2010 , Manjón et al, 2013 , Manjón et al, 2015 , Bouhrara et al, 2016 , Sudeep et al, 2018 , Pieciak et al, 2018 ), the random matrix theory approach which exploits the Marchenko–Pastur law of the eigenvalues of noise ( Veraart et al, 2016c ) and a group of algorithms that uses joint information from spatial and q -space domains to significantly improve previous results ( St-Jean et al, 2016 , Chen et al, 2019b , Chen et al, 2019a ). Notice here that any aggregation-based algorithm introduces a systematic bias to aggregated signal that should be corrected prior to a quantitative interpretation.…”