Comparison of Methods for Estimation of the Intravoxel Incoherent Motion (IVIM) Perfusion Parameters

Oscar Jalnefjord^1,2
¹Department of Medical Radiation Sciences, Institute of Clinical Sciences, Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden, ²Department of Medical Physics and Biomedical Engineering, Sahlgrenska University Hospital, Gothenburg, Sweden

Synopsis

Keywords: Contrast mechanisms: Perfusion, Image acquisition: Quantification, Contrast mechanisms: Diffusion

This lecture provides an overview of methodological aspects of the process to generate intravoxel incoherent motion (IVIM) parameter maps. We will cover both image acquisition, (diffusion encoding, readout methods, other preparation modules) and image processing (preprocessing, choice of model, methods for parameter estimation), all with an emphasis on the perfusion-related IVIM parameters.

Introduction

The intravoxel incoherent motion (IVIM) model, as suggested by Denis le Bihan and colleagues, provides a means to disentangle the effects of diffusion and perfusion on the diffusion-weighted MRI signal.¹ In its general form, it assumes that the signal originates from two compartments corresponding to the intravascular space (perfusion compartment) and the extravascular space (diffusion compartment):
$$S=S_0[(1-f)F_D+fF_P]$$
where S and S₀ are the signals with or without diffusion encoding applied, f is the signal fraction corresponding to the perfusion compartment (the perfusion fraction), and F_D and F_P describe the signal decay due to diffusion encoding in the diffusion and perfusion compartments, respectively. Assuming that all b-values are relatively small, F_D is well described by a monoexponential function. Furthermore, if the blood changes direction sufficiently during the application of diffusion encoding, the effect on the signal is similar to that of diffusion (pseudo-diffusion), thus also resulting in a monoexponential function. These assumptions give the IVIM signal model commonly used:
$$S=S_0[(1-f)e^{-bD}+fe^{-bD^*}]$$

where D is the diffusion coefficient of water in tissue, and D* is the pseudo-diffusion coefficient, describing the motion of blood through the vasculature. Although the assumption of pseudo-diffusion is not always met, this biexponential function will often describe data well for the typical IVIM experiment using monopolar gradients for diffusion encoding, but does in such cases turn more into a signal representation rather than a biophysical model, reducing the interpretability of the derived parameters.^2,3

Methods for image acquisition

The arguably most important aspect of an IVIM experiment is the design of the diffusion encoding scheme, including b-values and gradient shapes. To capture the fast signal decay in the perfusion compartment, low b-values must be utilized, but the number of unique b-values to be used depends on the context. It has been shown multiple times that repeated acquisition of a set of few key b-values provides superior precision over use of a larger number of different b-values.^4,5 Such optimization does however require prior knowledge of expected IVIM parameter values in the tissue type or organ of interest, which sometimes is not available due to limited or contradictory results in literature. Use of unconventional diffusion encoding gradient waveforms, enabling for example flow-compensation, opens the possibility for use of more specific models and may improve robustness, but can be more sensitive to factors like encoding time and puts higher demands on gradient hardware to achieve useful TEs relative to when conventional monopolar gradients are used.^2,6

As for most diffusion MRI (dMRI), EPI readouts are used in the vast majority of all IVIM scans with its well-known pros and cons. Turbo spin echo (TSE) readout provides an alternative to EPI with negligible distortions, but is associated with lower SNR and increased blurring due to the longer readout time resulting from the introduction of additional refocusing RF pulses. This might, however, be a price worth paying if distortions are a major concern.⁷ Comparing results between EPI acquisitions and TSE acquisitions, or between scanners with different gradient setups may however be difficult due to differing TE. Following the different T2 of the perfusion and diffusion compartments, which typically is not considered in the model, f (and S₀) will depend on the TE.⁸

In addition to fat suppression, which is used in dMRI in general, additional preparation modules may be of value for IVIM. When imaging the brain, the signal from cerebrospinal fluid may act as a confounder and can be suppressed by specifically designed preparation pulses.⁹

Methods for image processing

The combination of strong gradients for diffusion encoding, (typically) EPI readout, and low signal makes images for IVIM prone to artefacts and low SNR. Application of a set of preprocessing steps prior to estimation of IVIM parameters if often necessary to achieve sufficient quality of IVIM parameter maps. Tax et al. nicely outline the preprocessing steps commonly needed for dMRI of the brain, which at least in part is applicable to IVIM of the brain, and to some extent can serve as a guideline for IVIM in the body.¹⁰ Some preprocessing algorithms are directly applicable to the IVIM context, like Gibbs ringing removal or signal drift correction.^11,12 Preprocessing algorithms that may depend more explicitly on the acquisition protocol, like denoising, need validation for the particular IVIM application or to be developed specifically for IVIM.^13–15 Other preprocessing algorithms have been developed specifically for dMRI of the brain, for example advanced frameworks for motion, eddy current and susceptibility distortion corrections, where a protocol with many diffusion encoding directions and few b-values usually is assumed, i.e. the opposite of the typical IVIM protocol.¹⁶ When preprocessing algorithms suitable for IVIM are not available, the alternative is typically some simple generic solution with inferior performance like intensity-based image registration to correct for motion and eddy current-induced distortions. Additional development of algorithms specifically tailored for IVIM would be of value for the field.¹⁷

The particular IVIM model to be used should ideally be chosen prior to image acquisition to guide the design of the imaging protocol rather than the opposite. Regardless of which, the model used must be coherent with the imaging protocol used. If images with higher b-values are included in the analysis, non-monoexponential signal decay in the diffusion compartment must be included in the model to avoid bias propagating to the perfusion-related IVIM parameters.¹⁸ Alternatively, these higher b-values may be excluded from the analysis. Similarly, assuming that the signal from the perfusion compartment is negligible at some b-value when it is not may bias all IVIM parameters.¹⁹ However, while use of a more complicated model may reduce potential bias it does put higher demands on the data in order to achieve sufficient precision. Carefully designed imaging protocols should be used to balance these competing goals.

It was early recognized that fitting the biexponential IVIM model to data is a process sensitive to noise and it was suggested to use either a so-called segmented algorithm or Bayesian parameter estimation to improve the precision.^20,21 In the segmented algorithm, D and an extrapolated intercept term A are estimated from b-values above some threshold in a first step, followed by estimation of the remaining IVIM parameters with D and A fixed in a second step. This can be seen as a generalization of the simple algorithm suggested by le Bihan et al. in the original IVIM paper where they estimated D from two images with non-zero b-values (approx. 100 and 200 s/mm²) and then estimated f based on D and the b = 0 image.¹ Bayesian methods allow for a general inclusion of prior knowledge and assumptions, enabling for example sharing of information among voxels, thus increasing the resilience to noise.^22,23 More recently, deep learning-based methods have been developed, showing similar robustness against noise as the Bayesian methods while being more computationally efficient.^24,25

Acknowledgements

OJ is supported by grants from the Swedish state under the agreement between the Swedish government and the county councils, the ALF-agreement (ALFGBG-942664).

References

1. Le Bihan D, Breton E, Lallemand D, Aubin ML, Vignaud J, Laval-Jeantet M. Separation of Diffusion and Perfusion in Intravoxel Incoherent Motion MR Imaging. Radiology. 1988;168(2):497-505.

2. Ahlgren A, Knutsson L, Wirestam R, et al. Quantification of microcirculatory parameters by joint analysis of flow-compensated and non-flow-compensated intravoxel incoherent motion (IVIM) data. NMR Biomed. 2016;29(5):640-649. doi:10.1002/nbm.3505

3. Novikov DS, Kiselev VG, Jespersen SN. On modeling. Magn Reson Med. 2018;79(6):3172-3193. doi:10.1002/mrm.27101

4. Leporq B, Saint-Jalmes H, Rabrait C, et al. Optimization of Intra-voxel Incoherent Motion Imaging at 3.0 Tesla for Fast Liver Examination. Journal of Magnetic Resonance Imaging. 2015;41:1209-1217. doi:10.1002/jmri.24693

5. Jalnefjord O, Montelius M, Starck G, Ljungberg M. Optimization of b-value schemes for estimation of the diffusion coefficient and the perfusion fraction with segmented intravoxel incoherent motion model fitting. Magn Reson Med. 2019;82(4):1541-1552.

6. Wetscherek A, Stieltjes B, Laun FB. Flow-compensated intravoxel incoherent motion diffusion imaging. Magn Reson Med. 2015;74(2):410-419. doi:10.1002/mrm.25410

7. Hilbert F, Wech T, Neubauer H, Veldhoen S, Bley TA, Köstler H. Comparison of image quality and quantitative parameters in intravoxel incoherent motion imaging at 3-T based on turbo spin-echo and echoplanar imaging in patients with oral cancer. Z Med Phys. 2017;27(3):193-201. doi:10.1016/j.zemedi.2016.12.001

8. Jerome NP, D’Arcy JA, Feiweier T, et al. Extended T2-IVIM model for correction of TE dependence of pseudo-diffusion volume fraction in clinical diffusion-weighted magnetic resonance imaging. Phys Med Biol. 2016;61(24):667-680. doi:10.1088/1361-6560/61/24/N667

9. Federau C, O’Brien K. Increased brain perfusion contrast with T2-prepared intravoxel incoherent motion (T2prep IVIM) MRI. NMR Biomed. 2015;28(1):9-16. doi:10.1002/nbm.3223

10. Tax CMW, Bastiani M, Veraart J, Garyfallidis E, Okan Irfanoglu M. What’s new and what’s next in diffusion MRI preprocessing. Neuroimage. 2022;249(April 2021):118830. doi:10.1016/j.neuroimage.2021.118830

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

12. Vos SB, Tax CMW, 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. doi:10.1002/mrm.26124

13. Huang H. Reliable estimation of brain intravoxel incoherent motion parameters using denoised diffusion‐weighted MRI. NMR Biomed. January 2020. doi:10.1002/nbm.4249

14. Reischauer C, Gutzeit A. Image denoising substantially improves accuracy and precision of intravoxel incoherent motion parameter estimates. PLoS One. 2017;12(4). doi:10.1371/journal.pone.0175106

15. Gurney-Champion OJ, Collins DJ, Wetscherek A, et al. Principal component analysis for fast and model-free denoising of multi b-value diffusion-weighted MR images. Phys Med Biol. 2019;64(10):105015. doi:10.1088/1361-6560/ab1786

16. Andersson JLR, 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

17. Kurugol S, Freiman M, Afacan O, et al. Motion-robust parameter estimation in abdominal diffusion-weighted MRI by simultaneous image registration and model estimation. Med Image Anal. 2017;39:124-132. doi:10.1016/j.media.2017.04.006

18. Le Bihan D. What can we see with IVIM MRI? Neuroimage. 2019;187:56-67. doi:10.1016/J.NEUROIMAGE.2017.12.062

19. Wurnig MC, Donati OF, Ulbrich E, et al. Systematic analysis of the intravoxel incoherent motion threshold separating perfusion and diffusion effects: Proposal of a standardized algorithm. Magn Reson Med. 2015;74(5):1414-1422. doi:10.1002/mrm.25506

20. Pekar J, Moonen CTW, van Zijl PCM. On the precision of diffusion / perfusion imaging by gradient sensitization. Magn Reson Med. 1992;23:122-129.

21. Neil JJ, Bretthorst GL. On the use of Bayesian probability theory for analysis of exponential decay data: an example taken from intravoxel incoherent motion experiments. Magn Reson Med. 1993;29:642-647.

22. Orton MR, Collins DJ, Koh DM, Leach MO. Improved Intravoxel Incoherent Motion Analysis of Diffusion Weighted Imaging by Data Driven Bayesian Modeling. Magn Reson Med. 2014;71:411-420. doi:10.1002/mrm.24649

23. Freiman M, Perez-Rossello JM, Callahan MJ, et al. Reliable estimation of incoherent motion parametric maps from diffusion-weighted MRI using fusion bootstrap moves. Med Image Anal. 2013;17(3):325-336. doi:10.1021/nl061786n.Core-Shell

24. Barbieri S, Gurney‐Champion OJ, Klaassen R, Thoeny HC. Deep learning how to fit an intravoxel incoherent motion model to diffusion‐weighted MRI. Magn Reson Med. 2020;83:312-321. doi:10.1002/mrm.27910

25. Kaandorp MPT, Barbieri S, Klaassen R, et al. Improved unsupervised physics-informed deep learning for intravoxel incoherent motion modeling and evaluation in pancreatic cancer patients. Magn Reson Med. 2021;86(4):2250-2265. doi:10.1002/mrm.28852

Proc. Intl. Soc. Mag. Reson. Med. 32 (2024)