Image Artifacts & Processing Pipelines, Part II
Jelle Veraart1,2 and Rita G. Nunes3

1Champalimaud Research, Champalimaud Centre for the Unknown, Lisbon, Portugal, 2imec-Vision Lab, Department of Physics, University of Antwerp, Antwerp, Belgium, 3ISR-Lisboa/LARSyS and Department of Bioengineering, Instituto Superior Técnico - Universidade de Lisboa, Lisbon, Portugal

Synopsis

In the second part of this topic, we focus on more advanced image processing methods. We will give an overview of image denoising methods, Gibbs ringing removal, outlier detection, frequency stabilization, effects of gradient nonlinearity, and discuss challenges of pushing for higher spatial resolution.

Denoising

Due to the signal-attenuation induced by diffusion-sensitization and T2 - relaxation resulting from the long echo time necessary to accommodate gradient pulses, the signal-to-noise ratio of the DW signals is inherently low. Thermal noise that corrupts DWI measurements propagates to the diffusion parameters of interest and, as such, hampers visual inspection and precise quantitative interpretation of the underlying diffusion process. Although attempts have been made to minimize the noise propagation by optimizing diffusion encoding settings1,2, scan time limitations put a bar on what is to gain with protocol optimization in terms of precision. Therefore, image denoising, i.e. minimizing the variance of the DW signals in a post-processing step, is essential to raise that bar e.g. 4-6.

Many denoising strategies are based on weighted averages of voxels, where the voxels (and weights) are selected by metric similarity of patches e.g. 5. Loss of spatial resolution of the image (blur) and introduction of additional partial volume effects might remove fine anatomical details and might complicate further quantitative analyses or bias the diffusion modeling. Alternatively, several approaches have adopted the idea of noise removal by means of transforming a redundant dataset into a principal component basis and preserving only the signal-carrying principal components4,6.

Gibbs Ringing

Spurious ringing near sharp edges, e.g., boundaries of tissues, appears in the reconstructed images any time the Fourier space is truncated during acquisition or compression7. MR images that are obtained via the inverse Fourier transformation (FT) of a finite k-space acquisition represent a classic case of this phenomenon, referred to as Gibbs ringing. Although ringing is a feature of all MRIs obtained via inverse Fourier transform, it becomes increasingly significant for low resolution and/or quantitative modalities such as DWI8-10. While introducing a bias of at most 9% of the intensity step in the individual “weighted” images, the Gibbs effect is amplified tremendously in the derived parametric maps, e.g., of diffusion and kurtosis tensor components, often affecting major white matter bundles such as the Corpus Callosum.

The most common strategy to suppress the Gibbs artifact has been spatial smoothing of the MR data, Alternatively, one can apply a regularized extrapolation of the k‐space beyond its measured part to avoid the sharp cut‐off. Total variation (TV) regularization has been suggested to stabilize this ill‐posed estimation problem8,9. Recently, a more targeted approach has been introduced. Gibbs ringing can be minimized by reinterpolating the image using optimized subvoxel shifts to minimize the oscillations around sharp edges10.

Outliers:

The accuracy of diffusion parameter estimators is hindered by the presence of physiological noise, i.e. a plethora of particular artifacts that stem from various sources. Spatially and temporally varying artifacts, e.g. cardiac pulsation, system instabilities, or intra-scan motion, will perturb the DW signals in such a way they often cannot be modeled or parameterized. The detection and removal of such artifacts depends on robust parameter estimatorse.g. 11-14.

Frequency Stabilization

The resonance frequency might change over time due to heating of the scan hardware. Both the application of strong DW gradient or echo-planar read-outs might induce such a temperature changes15, especially during longer scans. If resonance frequency stabilization is not performed during the scan, then artifacts such as image translation in the phase encoding direction or global signal intensity scaling can occur15,16.

Gradient Nonlinearity

In practice, magnetic field gradients deviate from their nominal magnitude and orientation. This gradient nonlinearities increase with the distance from the isocenter and lead to geometric distortions as well as biases in DWI quantification17. Indeed, the nonlinearities alter the strength and direction of diffusion-sensitizing gradients from their intended values, leading to spatially dependent errors in the b‐value. If the b-values and DW directions are not corrected retrospectively using scanner-specific gradient calibration settings, as specified by the manufacturer, systematic errors in the computation of diffusion metrics, such as the apparent diffusion coefficient, can exceed 10–20% over a clinically relevant field‐of‐view18 and becomes increasingly important on modern high performing scanner such as the Siemens Connectom Scanner19.

Challenges of Pushing for Higher Spatial-Resolution

As explained in the first lecture, single-shot Echo Planar Imaging is most frequently used due to dMRI’s extreme sensitivity to motion. Covering k-space in one single-shot minimizes motion effects, but it does limit the achievable spatial resolution: as transverse relaxation is occurring throughout the readout window, blurring is introduced in the image reducing its effective spatial resolution while progressively decreasing the available signal. An alternative approach is to cover k-space over multiple-shots and to deal with inconsistencies introduced in the data as a result of different types of motion occurring during the corresponding diffusion preparation modules20.

Early strategies involved the acquisition of a navigator which would sample the same central region of k-space for each shot21. This region corresponded to a low-resolution but full field-of-view image which would enable to measure and correct for the motion-induced and typically smoothly varying spatial phase patterns. Recently, an alternative approach has been introduced to estimate this information which does not require the acquisition of navigators22. Provided the number of coil channels is sufficiently higher than the number of k-space shots, it is possible to reconstruct a non-aliased image (albeit with low signal-to-noise ratio) from each data segment using parallel imaging. A full reconstruction can then be performed, incorporating the estimated phase information. Extensions to this method have since been proposed, allowing for 3D motion-correction and reconstruction23. Recently, reconstruction methods based on low-rank regularization have also been introduced24.

Acknowledgements

JV is a Postdoctoral Fellow of the Research Foundation - Flanders (FWO; grant number 12S1615N). RGN has received funding from the POR Lisboa 2020 program (grant number LISBOA-01-0145-FEDER-029686).

References

[1] Jones, D.K., 2003. Determining and visualizing uncertainty in estimates of fiber orientation from diffusion tensor MRI. Magnetic Resonance in Medicine, 49(1), pp. 7-12.

[2] Poot, D.H., Den Dekker, A.J., Achten, E., Verhoye, M. and Sijbers, J., 2010. Optimal experimental design for diffusion kurtosis imaging. IEEE transactions on medical imaging, 29(3), pp.819-829.

[3] Rudin, L.I., Osher, S. and Fatemi, E., 1992. Nonlinear total variation based noise removal algorithms. Physica D: nonlinear phenomena, 60(1-4), pp.259-268.

[4] Hotelling, H., 1933. Analysis of a complex of statistical variables into principal components. Journal of educational psychology, 24(6), p.417.

[5] Manjón, J.V., Carbonell-Caballero, J., Lull, J.J., García-Martí, G., Martí-Bonmatí, L. and Robles, M., 2008. MRI denoising using non-local means. Medical image analysis, 12(4), pp.514-523.

[6] Veraart, J., Novikov, D.S., Christiaens, D., Ades-Aron, B., Sijbers, J. and Fieremans, E., 2016. Denoising of diffusion MRI using random matrix theory. NeuroImage, 142, pp.394-406.

[7] Gibbs, J.W., 1898. Fourier's series. Nature, 59(1522), p.200.

[8] Veraart, J., Fieremans, E., Jelescu, I.O., Knoll, F. and Novikov, D.S., 2016. Gibbs ringing in diffusion MRI. Magnetic resonance in medicine, 76(1), pp.301-314.

[9] Perrone, D., Aelterman, J., Pižurica, A., Jeurissen, B., Philips, W. and Leemans, A., 2015. The effect of Gibbs ringing artifacts on measures derived from diffusion MRI. NeuroImage, 120, pp.441-455.

[10] Kellner, E., Dhital, B., Kiselev, V.G. and Reisert, M., 2016. Gibbs‐ringing artifact removal based on local subvoxel‐shifts. Magnetic resonance in medicine, 76(5), pp.1574-1581.

[11] Chang, L.C., Jones, D.K. and Pierpaoli, C., 2005. RESTORE: robust estimation of tensors by outlier rejection. Magnetic Resonance in Medicine, 53(5), pp.1088-1095.

[12] Tax, C.M., Otte, W.M., Viergever, M.A., Dijkhuizen, R.M. and Leemans, A., 2015. REKINDLE: robust extraction of kurtosis INDices with linear estimation. Magnetic Resonance in Medicine, 73(2), pp.794-808.

[13] Collier, Q., Veraart, J., Jeurissen, B., den Dekker, A.J. and Sijbers, J., 2015. Iterative reweighted linear least squares for accurate, fast, and robust estimation of diffusion magnetic resonance parameters. Magnetic resonance in medicine, 73(6), pp.2174-2184.

[14] Chang, L.C., Walker, L. and Pierpaoli, C., 2012. Informed RESTORE: a method for robust estimation of diffusion tensor from low redundancy datasets in the presence of physiological noise artifacts. Magnetic resonance in medicine, 68(5), pp.1654-1663.

[15] Foerster, B.U., Tomasi, D. and Caparelli, E.C., 2005. Magnetic field shift due to mechanical vibration in functional magnetic resonance imaging. Magnetic Resonance in Medicine, 54(5), pp.1261-1267.

[16] Vos, S.B., Tax, C.M., Luijten, P.R., Ourselin, S., Leemans, A. and Froeling, M., 2017. The importance of correcting for signal drift in diffusion MRI. Magnetic resonance in medicine, 77(1), pp.285-299.

[17] Bammer, R., Markl, M., Barnett, A., Acar, B., Alley, M.T., Pelc, N.J., Glover, G.H. and Moseley, M.E., 2003. Analysis and generalized correction of the effect of spatial gradient field distortions in diffusion‐weighted imaging. Magnetic Resonance in Medicine, 50(3), pp.560-569.

[18] Malyarenko, D., Galbán, C.J., Londy, F.J., Meyer, C.R., Johnson, T.D., Rehemtulla, A., Ross, B.D. and Chenevert, T.L., 2013. Multi‐system repeatability and reproducibility of apparent diffusion coefficient measurement using an ice‐water phantom. Journal of Magnetic Resonance Imaging, 37(5), pp.1238-1246.

[19] Sotiropoulos, S.N., Jbabdi, S., Xu, J., Andersson, J.L., Moeller, S., Auerbach, E.J., Glasser, M.F., Hernandez, M., Sapiro, G., Jenkinson, M. and Feinberg, D.A., 2013. Advances in diffusion MRI acquisition and processing in the Human Connectome Project. Neuroimage, 80, pp.125-143.

[20] Anderson, A. and Gore, J., 1994. Analysis and correction of motion artifacts in diffusion weighted imaging. Magnetic resonance in medicine, 32(3), pp.379-87

[21] Butts, K., Pauly, J., de Crespigny, A., Moseley, M., 1997. Isotropic diffusion-weighted and spiral-navigated interleaved EPI for routine imaging of acute stroke. Magnetic resonance in medicine, 38(5), pp.741-9.

[22] Chen, N., Guidon, A., Chang, H. and Song, A., 2013. A robust multi-shot scan strategy for high-resolution diffusion weighted MRI enabled by multiplexed sensitivity-encoding (MUSE). NeuroImage, 72, pp.41-47.

[23] Chang, H.C., Hui, E.S., Chiu, P.W., Liu, X. and Chen, N.K., 2018. Phase correction for three-dimensional (3D) diffusion-weighted interleaved EPI using 3D multiplexed sensitivity encoding and reconstruction (3D-MUSER). Magnetic resonance in medicine, 79(5), pp.2702-2712.

[24] Mani, M., Jacob, M., Kelley, D., Magnotta, V., 2017. Multi-shot sensitivity-encoded diffusion data recovery using structured low-rank matrix completion (MUSSELS). Magnetic resonance in medicine, 78(2), pp.494-507.

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)