Rician Noise Removal by Non-Local Means Filtering for Low Signal-to-Noise Ratio MRI: Applications to DT-MRI

Abstract

Diffusion-Weighted MRI (DW-MRI) is subject to random noise yielding measures that are different from their real values, and thus biasing the subsequently estimated tensors. The Non-Local Means (NLMeans) filter has recently been proposed to denoise MRI with high signal-to-noise ratio (SNR). This filter has been shown to allow the best restoration of image intensities for the estimation of diffusion tensors (DT) compared to state-of-the-art methods. However, for DW-MR images with high b-values (and thus low SNR), the noise, which is strictly Rician-distributed, can no longer be approximated as additive white Gaussian, as implicitly assumed in the classical formulation of the NLMeans. High b-values are typically used in high angular resolution diffusion imaging (HARDI) or q-space imaging (QSI), for which an optimal restoration is critical. In this paper, we propose to adapt the NLMeans filter to Rician noise corrupted data. Validation is performed on synthetic data and on real data for both conventional MR images and DT images. Our adaptation outperforms the original NLMeans filter in terms of peak-signal-to-noise ratio (PSNR) for DW-MRI.

MRI noise estimation and denoising using non-local PCA

“…The asymmetry of the Rician distribution results in a non-constant intensity bias as it depends on the local SNR. To reduce such bias, some authors have proposed to remove the bias in the squared magnitude image (Wiest-Daesslé et al, 2008, Manjón et al, 2008.…”

“…In MRI, early works using the NLM method are from Coupe et al (2008) and Manjón et al (2008). The bibliography related to this method is quite extensive (Tristan-Vega et al, 20012;Coupe et al, 2012;Manjón et al, 2009Manjón et al, , 2010Manjón et al, , 2012Wiest-Daesslé et al, 2008;He and Greenshields, 2009;Rajan et al 2012Rajan et al , 2014.…”

Efficient and robust computation of PDF features from diffusion MR signal

“…at high q values, and undesirably leads to overestimated diffusion measures [36]. Several studies has tackled this issue [39,40,41,42,43], but these methods are restricted to HARDI acquisitions. To the extend of our knowledge, no appropriate methods have been proposed for the robust estimation of dMRI signal from a HYDI-like dataset so far.…”

Design and Construction of a Realistic DWI Phantom for Filtering Performance Assessment

“…To palliate the effect of noise, a number of techniques may be used, including regularization after DT estimation [7], regularized estimation of the DT [8] and DWI restoration before DT estimation. In this last category, a number of techniques have been proposed: the Conventional Approach [9] (CA), based on the properties of the second order moment of Rician data; Maximum Likelihood (ML) estimation [10]; anisotropic diffusion [11]; wavelets [12]; total variation [13]; Unbiased Non Local Means (UNLM) [14,15]; Linear Minimum Mean Squared Error (LMMSE) filtering [16,17]; multichannel, Rician-corrected Wiener filtering [18], and others.…”

“…They include visual assessment [14,15,16,17] and indirect measures based on the properties of the DTI volumes recovered from filtered DWI [16,18]; in this case, the computation of related parameters such as the smoothness in the fiber tracts estimated from DTI [18] make it difficult to evaluate the filtering performance, since the final result depends on a number of factors others than the filtering itself. On the other hand, direct evaluation on the filtered DWI is a difficult task; in [14] 12 acquisitions of the same patient are available, so a leave-one-out strategy is used to filter one of the volumes each time and compare the result to a noise-free image obtained from the remaining 11 volumes.…”