Diffusion Pre-Processing & Artifact Correction
Rafael Neto Henriques1
1Champalimaud Research, Champalimaud Foundation, Lisbon, Portugal

Synopsis

As any other MRI modality, diffusion MRI can be corrupted by artifacts. In this lecture, we will introduce the common artifacts that compromise diffusion MRI, discuss how these can affect different diffusion MRI estimates, and which state-of-the-art pre-processing strategies can be used to minimize their effects.

Aim

The aim of this lecture is to introduce the common artifacts that can compromise diffusion MRI (dMRI) and give an overview of state-of-the-art processing strategies to mitigate their detrimental effects.

Outcomes & objectives

Following this lecture, the audience will be able to:
  • Identify which common image artifacts compromise dMRI
  • Have an intuition on how different image artifacts can corrupt different dMRI estimates
  • Understand which pre-processing algorithm can be used to mitigate the effects of different image artifacts

Outline

In this lecture, the following image artifacts will be addressed:

1. Motion Artifacts
Diffusion MRI can be compromised by motion between or during the acquisition of dMRI image sets (i.e. volumes of images that cover a tissue of interest) which are acquired for different diffusion gradient directions and intensities.

1.1. Motion between dMRI volumes: Tissue position misalignment across different set of dMRI images are typically present due to involuntary and inevitable subject motion. Early approaches to correct image misalignment included the registration of data to the first acquired volume1-3. However, the use of these early techniques is not optimal since these do not consider the inherent contrast differences present on data acquired for different diffusion gradient directions and intensities. To overcome this issue, later approaches use predictive models in which dMRI data is registered to predicted image templates considering the expected image contrast for a given set of diffusion acquisition parameters4-7. Between-volume misalignment can also be corrected based on independent motion measurements obtained during data acquisition using imaging-based navigators or optimal tracking systems8,9. It is important to note that, after realignment of dMRI data volumes, a common step of all these motion correction procedures is to adapt the information of the applied diffusion gradient to ensure that the relative direction between diffusion gradients and subject tissue is maintained3.

1.2. Motion within-volumes: dMRI is typically acquired using an echo planar imaging (EPI) readout to enable the acquisition of a full image under a milliseconds time window. Although the use of an EPI readout may minimize in-slice motion artifacts, motion within the acquisition of a dMRI images can still induce signal dropouts in single or multiple dMRI images10. The effects of signal dropouts can be mitigated before diffusion model fitting (e.g. by replacing the signal dropout with predicted signals based on other dMRI volumes)6 or during diffusion model fitting (e.g. by using outliner identification and rejection during model fitting procedures)11-12.

2. B0 field inhomogeneities
Diffusion MRI is typically sensitive to B0 field inhomogeneities which can be partially caused by magnetic susceptibility differences between different tissues (e.g., interface between brain, air canals, bone)13. B0 field inhomogeneities introduces anatomical distortions which are independent to the applied diffusion gradient direction and intensity, and thus, these do not affect dMRI estimates such as the values of mean, axial and radial diffusivities and fractional anisotropy. Nevertheless, the correction of B0 field inhomogeneities distortions is fundamental if dMRI data is used for the virtual reconstruction of 3D tissue structures (i.e. tractography)14. Early approaches to correct for B0 field inhomogeneities distortion included the registration of dMRI to conventional structural images15. Alternatively, B0 field inhomogeneities can be more robustly corrected using susceptibility induced field maps which can be calculated from data acquired with inverted EPI phase encoding blips16.

3. Eddy Currents
The switching on and off the diffusion gradients can provoke undesired eddy currents which can induce additional anatomical distortions to dMRI images if these eddy currents overlap with imaging gradients17. Unlike the anatomical distortions from B0 field inhomogeneities, eddy-current distortions depend on the direction and intensities of applied diffusion gradients. Therefore, the correction of eddy-current distortions is fundamental to avoid misalignment across different dMRI volumes. As for the correction of motion-induced misalignment, early strategies to correct eddy-current distortions were based on registration procedures with affine transformation allowing to model image shearing, scalling and shifts which can be induced from eddy currents overlapping the image readout, phase encoding, and slice-selection imaging gradients respectively2,17. To account for different image contrast across the data acquired for different diffusion gradient directions and intensities, one can apply similar procedures than the registration methods based on predicted image templates mentioned for the correction of motion misalignment5,6. Indeed, these registration approaches are typically used to correct misalignment from both motion and eddy currents simultaneously.

4. Thermal noise
Diffusion MRI is intrinsically susceptible to random signal fluctuations of receiver coils (i.e. thermal noise). Suppression of thermal noise effects may be fundamental for dMRI since this modality relies on the information of attenuated diffusion-weighted signals. Thermal noise in typically magnitude reconstructed MRI images is described by a non-central chi-distribution18-19, and thus it introduces signal biases at low SNR, particularly for signals acquired with high intensity diffusion gradient intensities20-21. Given its non-Gaussian characteristics, both signal fluctuations and biases from thermal noise can compromise the precision and accuracy of diffusion MRI estimates.

4.1. Thermal noise signal fluctuations: Signal fluctuation due to thermal noise can be propagated and even amplified on different diffusion MRI estimates (e.g. overestimation of diffusion fractional anisotropy due to diffusion tensor eigenvalue repulsion effects20, underestimation of diffusional kurtosis for tissues with expected low diffusivities22). Several strategies to suppress the effects of thermal noise signal fluctuations have been proposed on the last years, including standard image-domain smoothing and filtering strategies23-24, edge-preserving anisotropic filters25, wavelet transformations26-27, total variation minimization28, non-local mean filters29. However, denoising strategies based on Principal component analysis (PCA) are becoming popular for dMRI since these exploits the information redundancy of signals across the multiple diffusion MRI measurements30-31. These strategies assume that only a few principal components contain significant information for dMRI, and thus noise fluctuations can be suppressed by removing principal components that are mostly related to noise. The criteria for the classification of signal and noise components can be performed using prior noise variance estimates30, random matrix theory31-32 or the combination of both33. In addition to the PCA-based approaches, emerging denoising strategies based on machine learning are showing promising results34-35. For instance, self-supervising machine learning denoising can potentially overcome some of the assumptions of the nature of thermal noise considered by current PCA-based denoising strategies35.

4.2. Thermal noise signal biases: Noise signal biases are higher when the signal-to-noise ratio is lower, and, therefore, it specially affects signals from regions in which diffusion gradients are parallel to highly anisotropic structures and signals acquired for high diffusion gradient intensities. Consequently, if not considered, noise biases can impact the accuracy of diffusion anisotropic estimates and non-Gaussian diffusion quantities21,36. Strategies to suppress the effects of noise biases includes: 1) the incorporation of the noncentral-Chi distributions directly on the fitting procedures of different diffusion models using different probabilistic frameworks36,37; and 2) correction of biases on dMRI signals before model fitting by using moment inversion algorithms38. Alternatively, in case that MRI complex data is available, noise signal biases can be avoided by reconstructing real-valued dMRI images where noise is expected to be Gaussian39.

5. Gibbs artefacts
As any other MRI modality, dMRI images are reconstructed from a finite sampled k-space. If not enough k-space high-frequencies are sampled, diffusion MRI images will be corrupted by spatial oscillations known as Gibbs Artifacts40. Gibbs Artefacts can have opposite effects on data acquired along different gradient directions and intensities due the different contrasts provided. Consequently, Gibbs ringing can introduce systematic biases on dMRI estimates41,42. One way to correct for Gibbs artifacts is to apply filters on diffusion-weighted images. However, the use of filters introduces blurring - impacting the effective image resolution. More advanced strategies to mitigate the effects of Gibbs Rings includes strategies to extrapolate k-space high frequencies42 and procedures to interpolate MRI signals at the oscillations zero-crossing using subvoxel-shifts34,43-45.

6. Signal drifts
During the acquisition of dMRI data along different gradient directions and intensities, undesired changes on signal intensity can occur due to scanner instabilities (e.g. due to rapid switching of EPI and diffusion gradients, magnetic fields drifts)46. If not corrected, these gradual signal changes (i.e. signal drifts) can introduce systematic artifacts on dMRI estimates46. To inspect and correct from signal drifts, recent studies had proposed the interleaved b = 0 acquisitions. Signal drifts can be corrected based on b = 0 images assuming linear or quadratic signal drift models46,47.

Final remarks

At the end of this lecture, we will discuss the order that the different pre-processing algorithms can be applied and indicate some future vistas on the development of novel pre-processing techniques that can be fundamental to further improve the reliability of dMRI.

Acknowledgements

No acknowledgement found.

References

1 Maes F, Collignon A, Vandermeulen D, Marchal G, Suetens P. Multimodality image registration by maximization of mutual information. IEEE Trans Med Imag 1997, 16: 187– 198.

2 Jenkinson M, Smith SA. global optimisation method for robust affine registration of brain images. Med Image Anal 2001 5 (2), 143–156. doi: 10.1016/s1361-8415(01)00036-6

3 Leemans A, Jones DK. The b-matrix must be rotated when correcting for subject motion in DTI data. Magn Reson Med 2009, 61 (6), 1336–1349. doi: 10.1002/mrm.21890

4 Ben-Amitay S, Jones DK, Assaf Y. Motion correction and registration of high b-value diffusion weighted images. Magn Reson Med 2012, 67 (6), 1694–1702. doi: 10.1002/mrm.23186

5 Nilsson M, Szczepankiewicz F, van Westen D, Hansson O. Extrapolation-based references improve motion and eddy-Current correction of high B-Value DWI data: application in Parkinson’s disease dementia. PLoS ONE 2015, 10 (11), e0141825. doi: 10.1371/journal.pone.0141825

6 Andersson JL, Sotiropoulos SN. An integrated approach to correction for off-resonance effects and subject movement in diffusion MR imaging. Neuroimage 2016, 125, 1063–1078. doi: 10.1016/j.neuroimage.2015.10.019

7 Henriques RN, CamCAN, Correia MM. Reducing inter and intravolume instabilities on diffusion-weighted data for ageing studies. Proceedings of 22nd Annual Meeting of the OHBM 2016, Geneva, Switzerland

8 Alhamud A, Tisdall MD, Hess AT, Hasan KM, Meintjes EM, van der Kouwe AJ. Volumetric navigators for real-time motion correction in diffusion tensor imaging. Magn Reson Med 2012, 68 (4), 1097–1108. doi: 10.1002/mrm.23314

9 Zaitsev M, Dold C, Sakas G, Hennig J, Speck O. Magnetic resonance imaging of freely moving objects: prospective real-time motion correction using an external optical motion tracking system. Neuroimage 2006, 31 (3), 1038–1050. doi: 10.1016/J.NEUROIMAGE.2006.01.039

10 Tournier JD, Mori S, Leemans A. Diffusion Tensor Imaging and Beyond. Magn. Reson. Med. 2011, 65(6): 1532–1556. doi:10.1002/mrm.22924

11 Chang L-C, Walker L, Pierpaoli C. RESTORE: a method for robust estimation of diffusion tensor from low redundancy datasets in the presence of physiological noise artifacts. Magn Reson Med 2012, 68(5), 1654–1663. doi: 10.1002/mrm.24173

12 Tax CM, Otte WM, Viergever MA, Dijkhuizen RM, Leemans A. REKINDLE: Robust extraction of kurtosis INDices with linear estimation. Magn Reson Med 2015, 73 (2), 794–808. doi: 10.1002/mrm.25165

13 Jezzard P, Balaban RS. Correction for geometric distortion in echo planar images from B0 field variations. Magn Reson Med 1995, 34, 65–73

14 Irfanoglu MO, Walker L, Sarlls J, Marenco S, Pierpaoli C. Effects of image distortions originating from susceptibility variations and concomitant fields on diffusion MRI tractography results. Neuroimage 2012, 61(1):275-88. doi: 10.1016/j.neuroimage.2012.02.054.

15 (61), 275–288.15 Tao R, Fletcher PT, Gerber S, Whitaker RT. A variational image-based approach to the correction of susceptibility artifacts in the alignment of diffu- sion weighted and structural MRI. Information Processing in Medical Imaging 2009, 21, 651–663

16 Andersson JLR, Skare S, Ashburner J. How to correct susceptibility distortions in spin-echo echo-planar images: application to diffusion tensor imaging. NeuroImage 2003, 20(2):870-888

17 Jezzard P, Barnett AS, Pierpaoli C. Characterization of and correction for eddy current artifacts in echo planar diffusion imaging. Magn Reson Med 1998, 39(5), 801–812. doi: 10.1002/mrm.1910390518

18 Aja-Fernández S, Tristán-Vega A. Influence of Noise Correlation in Multiple-Coil Statistical Models With Sum of Squares Reconstruction. Magn Reson Med 2012, 67(2), 580-585. doi: 10.1002/mrm.23020

19 Aja-Fernández S, Tristán-Vega A, Hoge WS. Statistical noise analysis in grappa using a parametrized non-central chi approximation model. Magn Reson Med 2011, 65(4), 1195–1206. doi: 10.1002/mrm.22701

20 Pierpaoli C, Basser PJ. Toward a quantitative assessment of diffusion anisotropy. Magn. Reson. Med. 1996, 36(6), 893-906. doi:10.1016/j.cam.2007.10.012

21 Jones DK, Basser PJ. “Squashing peanuts and smashing pumpkins”: How noise distorts diffusion-weighted MR data. Magn. Reson. Med. 2004, 52(5), 979–993. doi:10.1002/mrm.20283

22 Henriques RN, Jespersen SN, Jones D, Veraart J. Towards more robust and reproducible Diffusion Kurtosis Imaging. Magn. Reson. Med. 2021, 86(3):1600-1613. doi: 10.1002/mrm.28730

23 Jones DK, Cercigani M. Twenty-five pitfalls in the analysis of diffusion MRI data. NMR Biomed. 2010, 23:803–820 doi: 10.1002/NBM.1543

24 Jones DK, Symms MR, Cercignani M, Howard RJ. The effect of filter size on VBM analyses of DT-MRI data. Neuroimage 2005, 26:546–554 doi: 10.1016/J.NEUROIMAGE.2005.02.013

25 Gerig G, Kbler O, Kikinis R, Jolesz FA. Nonlinear Anisotropic Filtering of MRI Data. IEEE Trans. Med. Imaging 1992;11:221–232 doi: 10.1109/42.141646

26 Nowak RD. Wavelet-based Rician noise removal for magnetic resonance imaging. IEEE Trans. Image Process. 1999, 8:1408–1419 doi: 10.1109/83.791966

27 Pižurica A, Philips W, Lemahieu I, Acheroy M. A versatile wavelet domain noise filtration technique for medical imaging. IEEE Trans. Med. Imaging 2003;22:323–331 doi: 10.1109/TMI.2003.809588

28 Knoll F, Bredies K, Pock T, Stollberger R. Second order total generalized variation (TGV) for MRI. Magn. Reson. Med. 2011;65:480–491 doi: 10.1002/MRM.22595

29 Coupé P, Yger P, Prima S, Hellier P, Kervrann C, Barillot C. An optimized blockwise nonlocal means denoising filter for 3-D magnetic resonance images. IEEE Trans. Med. Imaging 2008;27:425 doi: 10.1109/TMI.2007.906087

30 Manjón JV, Coupé P, Concha L, Buades A, Collins DL, Robles M. Diffusion Weighted Image Denoising Using Overcomplete Local PCA Gong G, editor. PLoS One 2013, 8, e73021 doi: 10.1371/journal.pone.0073021

31 Veraart J, Novikov DS, Christiaens D, Ades-aron B, Sijbers J, Fieremans E. Denoising of diffusion MRI using random matrix theory. Neuroimage 2016;142:394–406 doi: 10.1016/j.neuroimage.2016.08.016

32 Bazin P-L, Alkemade A, van der Zwaag W, Caan M, Mulder M, Forstmann BU. Denoising High-Field Multi-Dimensional MRI With Local Complex PCA. Front. Neurosci. 2019;0:1066 doi: 10.3389/FNINS.2019.01066

33 Henriques RN, Jespersen SN, Shemesh N. Hybrid PCA denoising - Improving PCA denoising in the presence of spatial correlations. Proceedings of Joint Annual Meeting ISMRM-ESMRMB 2022, London, UK

34 Muckley MJ, Ades-Aron B, Papaioannou A, Lemberskiy G, Solomon E Lui YW, Sodickson DK, Fieremans E, Novikov DS, Knoll F. Training a neural network for gibbs and noise removal in diffusion mri. Magn Reson Med 2021, 85 (1), 413–428 .

35 Fadnavis S, Batson J, Garyfallidis, E. “Patch2self: Denoising diffusion mri with self-supervised learning,” in Advances in Neural Information Processing Systems 2020, 33, 16293–16303

36 Veraart J, Van Hecke W, Sijbers J. Constrained maximum likelihood estimation of the diffusion kurtosis tensor using a rician noise model. Magn Reson Med 2011, 66 (3), 678–686.

37 Jbabdi S, Sotiropoulos SN, Savio AM, Graña M, Behrens TEJ. Model-based analysis of multishell diffusion mr data for tractography: how to get over fitting problems. Magn Reson Med 2012, 68 (6), 1846–1855. doi: 10.1002/mrm.24204.

38 Koay CG, Basser PJ. Analytically exact correction scheme for signal extraction fromnoisy magnitude MR signals. J Magn Reson 2006, 179, 317–322. doi: 10.1016/j.jmr.2006.01.016

39 Cordero-Grande L, Christiaens D, Hutter J, Price AN, Hajnal JV. Complex diffusion-weighted image estimation via matrix recovery under general noise models. Neuroimage 2019, 200, 391–404

40 Gibbs JW. Fourier’S series. Nature 1898, 59 (1522), 200–200.

41 Perrone D, Aelterman J, Pižurica A, Jeurissen B., Philips W, Leemans A. The effect of gibbs ringing artifacts on measures derived from diffusion MRI. Neuroimage 2015, 120, 441–455. doi: 10.1016/j.neuroimage.2015.06.068

42 Veraart J, Fieremans E, Jelescu IO, Knoll F, Novikov DS. Gibbs ringing in diffusion mri. Magn Reson Med 2016, 76 (1), 301–314

43 Kellner E, Dhital B, Kiselev VG, Reisert M. Gibbs-ringing artifact re- moval based on local subvoxel-shifts. Magn Reson Med 2016, 76 (5), 1574–1581. doi: 10.1002/mrm.26054

44 Zhang Q, Ruan G, Yang W, Liu Y, Zhao K, Feng Q, Chen W, Wu EX, Feng Y. MRI gibbs-ringing artifact reduction by means of machine learning using convolutional neural networks. Magn Reson Med 2019, 82 (6), 2133–2145

45 Lee H-H, Novilov DS, Fieremans E. Removal of partial Fourier-indiced Gibbs (RPG) ringing artifacts in MRI. Magn Reson Med 2021, 86(5), 2733-2750. doi: 10.1002/mrm.28830.

46 Vos SB, Tax CM, Luijten PR, Ourselin S, Leemans A, Froeling M. The importance of correcting for signal drift in diffusion MRI. Magn Reson Med 2017, 77 (1), 285–299 .

47 Hansen CB, Nath V, Hainline AE, Schilling KG, Parvathaneni P, Bayrak RG, Blaber JA, Irfanoglu O, Pierpaoli C, Anderson AW et al. Characterization and correlation of signal drift in diffusion weighted MRI. Magn Reson Imaging 2019, 57, 133–142 .

Proc. Intl. Soc. Mag. Reson. Med. 30 (2022)