Highlights
- •
MRI resolution is bounded by the level of thermal noise
- •
Mitigating noise typically requires time-consuming scans or expensive hardware upgrades
- •
We leverage redundancy in multi-channel-complex-valued data to reduce noise
- •
Noise reduction improves downstream analyses
The bigger picture
Summary
Graphical abstract
Keywords
Introduction
- Budinger T.F.
- Bird M.D.
- Frydman L.
- Long J.R.
- Mareci T.H.
- Rooney W.D.
- Rosen B.
- Schenck J.F.
- Schepkin V.D.
- Sherry A.D.
- et al.
Results
Efficacy on high-resolution in vivo human brain data
DWIs
Microstructure
- Huynh K.M.
- Xu T.
- Wu Y.
- Chen G.
- Thung K.-H.
- Wu H.
- Lin W.
- Shen D.
- Yap P.-T.
Probing brain micro-architecture by orientation distribution invariant identification of diffusion compartments.
Axonal orientations and tractography
Efficacy on in silico data
MCC versus magnitude denoising
Noise floor reduction
Axonal orientations and tractography
Discussion
Limitations of the study
Experimental procedures
Resource availability
Lead contact
Materials availability
Data and code availability
Problem formulation
Low-rank matrix recovery
or covariance matrix :
where and are the unitary matrices containing the left and right singular vectors of and the elements of diagonal matrix are the singular values . The elements of diagonal matrix are the eigenvalues , , of .
MP-PCA
where with , and is the number of signal components. The threshold P can be estimated simultaneously with σ based on the procedure described by Veraart et al.
where
where is a diagonal matrix with elements , .
OS-SVD
- •
OS-SVD does not just zero out singular values like MP-PCA but instead manipulates all singular values to mitigate noise contamination. This is especially important when M is small (e.g., due to limited channels, volumes, or block size) because limited singular values are available for accurate MP-PCA.
- •
OS-SVD is proven to be optimal with respect to a cost function.32
and
where with noise level σ and . The noise-free matrix is estimated as
with diagonal matrix containing elements .
Truncated SVD (TSVD)
Hard thresholding (Hard)
where the threshold is calculated as
Soft thresholding (Soft)
- (1)
Removing singular values below a threshold and retaining the rest, with the threshold depending only on the matrix size and noise level (TSVD and Hard),
- (2)
Removing singular values below a threshold and retaining the rest, with the threshold depending on the matrix size and the singular values (MP-PCA), or
- (3)
Altering all singular values (Soft, Fro, Op, and Nuc).
Noise estimation
where is the median empirical singular value of and is the median of the MP distribution determined by solving for μ in
where . Empirical results (Figure S5) indicate that estimator (Equation 16) yields greater accuracy. We therefore use this estimator for all MCC denoising methods, except for MCC MP-PCA, which concurrently estimates the noise level and removes noise based on MP-PCA with MCC data.
Determining the matrix size
Automated pipeline for effective noise removal
Channel decorrelation
where is the noise covariance matrix computed from channel-specific noise acquired without excitation, is the diagonal matrix of eigenvalues of , and is a unitary rotation matrix. can be calculated based on 1 k-space line sample with no RF excitation or channel signals of non-brain voxels.
Phase unwinding
Noise mapping and removal
where is the denoised signal of x in block j, J is the total number of blocks containing x, and is the weighting factor. To reduce Gibbs ringing artifacts, a block with a lower rank is assigned a greater weight
where is the rank of .
Phase rewinding (optional)
- Chang W.-T.
- Huynh K.M.
- Yap P.-T.
- Lin W.
Image reconstruction
- Chang W.-T.
- Huynh K.M.
- Yap P.-T.
- Lin W.
Evaluation approaches
In vivo data processing and evaluation
Structure preservation and SNR
Microstructure model fitting, axonal orientation estimation, and tractography
- Huynh K.M.
- Xu T.
- Wu Y.
- Chen G.
- Thung K.-H.
- Wu H.
- Lin W.
- Shen D.
- Yap P.-T.
Probing brain micro-architecture by orientation distribution invariant identification of diffusion compartments.
In silico data simulation and evaluation
where is the noise-free signal, is the channel sensitivity map, and and are complex noise added to channel c. The background phase was simulated using a bidimensional sinusoid along the x direction and y direction, with random shift along the z direction mimicking the smooth intra-slice and abrupt interslice transitions.
Noise mapping, background phase estimation, noise floor reduction, and denoising accuracy
where and are, respectively, the denoised and ground-truth images at voxel x and volume v. We also computed the PSNR
where MSE is the mean squared error between the denoised and noise-free images. PSNR was calculated for each b value separately.
Fiber orientations, fODF estimation, and tractography
with 0 corresponding to fully ICs or NCs and 1 corresponding to fully VCs.
Acknowledgments
Author contributions
Declaration of interests
Supplemental information
-
Document S1. Figures S1–S7 and Tables S1 and S2
References
-
The Rician distribution of noisy MRI data.Magn. Reson. Med. 1995; 34: 910-914
-
Toward 20 T magnetic resonance for human brain studies: opportunities for discovery and neuroscience rationale.Magnetic Resonance Materials in Physics, Biology and Medicine. 2016; 29: 617-639
-
Ultimate intrinsic signal-to-noise ratio in MRI.Magn. Reson. Med. 1998; 39: 462-473
-
7T: Physics, safety, and potential clinical applications.J. Magn. Reson. Imag. 2017; 46: 1573-1589
-
Fast submillimeter diffusion MRI using gSlider-SMS and SNR-enhancing joint reconstruction.Magn. Reson. Med. 2020; 84: 762-776
-
Diffusion MRI noise mapping using random matrix theory.Magn. Reson. Med. 2016; 76: 1582-1593
-
Diffusion weighted image denoising using overcomplete local PCA.PLoS One. 2013; 8e73021
-
Denoise magnitude diffusion magnetic resonance images via variance-stabilizing transformation and optimal singular-value manipulation.Neuroimage. 2020; 215116852
-
Complex diffusion-weighted image estimation via matrix recovery under general noise models.Neuroimage. 2019; 200: 391-404
-
Noise reduction with distribution corrected (NORDIC) PCA in dMRI with complex-valued parameter-free locally low-rank processing.Neuroimage. 2021; 226117539
-
Denoising of diffusion MRI using random matrix theory.Neuroimage. 2016; 142: 394-406
-
System, Method and Computer-Accessible Medium for Facilitating Noise Removal in Magnetic Resonance Imaging.(US Patent App. 17/109,050)2021
-
Diffusional kurtosis imaging: The quantification of non-Gaussian water diffusion by means of magnetic resonance imaging.Magn. Reson. Med. 2005; 53: 1432-1440
-
Probing tissue microarchitecture of the baby brain via spherical mean spectrum imaging.IEEE Trans. Med. Imag. 2020; 39: 3607-3618
-
NODDI: practical in vivo neurite orientation dispersion and density imaging of the human brain.Neuroimage. 2012; 61: 1000-1016
-
Probing brain micro-architecture by orientation distribution invariant identification of diffusion compartments.in: Medical Image Computing and Computer-Assisted Intervention (MICCA). Springer, 2019: 547-555
-
Model-based analysis of multishell diffusion MR data for tractography: How to get over fitting problems.Magn. Reson. Med. 2012; 68: 1846-1855
-
Investigating the prevalence of complex fiber configurations in white matter tissue with diffusion magnetic resonance imaging.Hum. Brain Mapp. 2013; 34: 2747-2766
-
Direct estimation of the fiber orientation density function from diffusion-weighted MRI data using spherical deconvolution.Neuroimage. 2004; 23: 1176-1185
-
Improved probabilistic streamlines tractography by 2nd order integration over fibre orientation distributions.in: International Society for Magnetic Resonance in Medicine. ISMRM, 2010
-
Achieving sub-mm clinical diffusion MRI resolution by removing noise during reconstruction using random matrix theory.in: International Society for Magnetic Resonance in Medicine (ISMRM). 27. 2019
-
MRI below the noise floor.in: International Society for Magnetic Resonance in Medicine (ISMRM). 28. 2020
-
Noise mapping and removal in complex-valued multi-channel MRI via optimal shrinkage of singular values.in: Medical Image Computing and Computer-Assisted Intervention (MICCAI). Springer, 2021: 191-200
-
A new truncation strategy for the higher-order singular value decomposition.SIAM J. Sci. Comput. 2012; 34: A1027-A1052
-
Tensor denoising of multidimensional MRI data.Magn. Reson. Med. 2023; 89: 1160-1172
-
Denoising of fMRI volumes using local low rank methods.in: International Symposium on Biomedical Imaging. ISBI, 2023: 1-5
-
Ventilation defect percentage comparison between spiral with post-acquisition denoising and Cartesian acquisition for 19F lung magnetic resonance imaging.J. Cyst. Fibros. 2022; 21: S259
-
Feasibility of free-breathing 19F MRI image acquisition to characterize ventilation defects in CF and healthy volunteers at wash-in.Magn. Reson. Med. 2023; 90: 79-89
-
Image denoising by sparse 3-D transform-domain collaborative filtering.IEEE Trans. Image Process. 2007; 16: 2080-2095
-
SNR-enhanced diffusion MRI with structure-preserving low-rank denoising in reproducing kernel Hilbert spaces.Magn. Reson. Med. 2021; 86: 1614-1632
-
Optimal Shrinkage Denoising Breaks the Noise Floor in High-Resolution Diffusion MRI.Data at OSF, 2024https://doi.org/10.17605/OSF.IO/F384H
-
Optimal shrinkage of singular values.IEEE Trans. Inf. Theor. 2017; 63: 2137-2152
-
The approximation of one matrix by another of lower rank.Psychometrika. 1936; 1: 211-218
-
Matrix estimation by universal singular value thresholding.Ann. Stat. 2015; 43: 177-214
-
A singular value thresholding algorithm for matrix completion.SIAM J. Optim. 2010; 20: 1956-1982
-
Minimax risk of matrix denoising by singular value thresholding.Ann. Stat. 2014; 42: 2413-2440
-
De-noising by soft-thresholding.IEEE Trans. Inf. Theor. 1995; 41: 613-627
-
SENSE: sensitivity encoding for fast MRI.Magn. Reson. Med. 1999; 42: 952-962
-
Real diffusion-weighted MRI enabling true signal averaging and increased diffusion contrast.Neuroimage. 2015; 122: 373-384
-
Noise floor removal via phase correction of complex diffusion-weighted images: Influence on DTI and q-space metrics.in: Medical Image Computing and Computer-Assisted Intervention (MICCAI). Springer, 2016: 21-34
-
Real valued diffusion-weighted imaging using decorrelated phase filtering.Magn. Reson. Med. 2017; 77: 559-570
-
Navigator-free submillimeter diffusion imaging using multishot-encoded simultaneous multi-slice (MUSIUM).arXiv. 2020; (Preprint at)https://doi.org/10.48550/arXiv.2012.00664
-
Influence of multichannel combination, parallel imaging and other reconstruction techniques on MRI noise characteristics.Magn. Reson. Imaging. 2008; 26: 754-762
-
Generalized autocalibrating partially parallel acquisitions (GRAPPA).Magn. Reson. Med. 2002; 47: 1202-1210
-
Dipy, a library for the analysis of diffusion MRI data.Front. Neuroinf. 2014; 8: 8-17
-
Multi-tissue constrained spherical deconvolution for improved analysis of multi-shell diffusion MRI data.Neuroimage. 2014; 103: 411-426
-
How to correct susceptibility distortions in spin-echo echo-planar images: application to diffusion tensor imaging.Neuroimage. 2003; 20: 870-888
-
a flexible software library to simulate diffusion MR phantoms.in: International Society for Magnetic Resonance in Medicine. ISMRM, 2014
-
Noise estimation and removal in MR imaging: The variance-stabilization approach.in: IEEE International Symposium on Biomedical Imaging (ISBI). IEEE, 2011: 1809-1814
-
MRtrix: diffusion tractography in crossing fiber regions.Int. J. Imag. Syst. Technol. 2012; 22: 53-66
-
Tractometer: towards validation of tractography pipelines.Med. Image Anal. 2013; 17: 844-857
-
Parallel transport tractography.IEEE Trans. Med. Imag. 2021; 40: 635-647
Article info
Publication history
Identification
Copyright
User license
Creative Commons Attribution (CC BY 4.0) |Permitted
- Read, print & download
- Redistribute or republish the final article
- Text & data mine
- Translate the article
- Reuse portions or extracts from the article in other works
- Sell or re-use for commercial purposes
Elsevier's open access license policy