Technical Aspect of Fat Quantification
Timothy Bray1,2
1Centre for Medical Imaging, University College London, London, United Kingdom, 2Department of Imaging, University College London Hospital, London, United Kingdom

Synopsis

Keywords: Contrast mechanisms: Fat

This talk will give an overview of methods for quantifying fat with MRI. MRI-based methods have emerged as valuable tools for assessing the fat content of tissue in a wide variety of organs and disease states, and can also provide fat-corrected measurements of other tissue characteristics such as relaxation times and diffusion coefficients. The basic principles of quantifying fat with MRI will be discussed, and methods for eliminating bias in fat quantification will be explained. Opportunities for future development including the inclusion of fat quantification within multiparametric acquisitions and the use of deep learning in fat quantification will be considered.

1. Why is fat important in MRI?

The majority of magnetic resonance imaging (MRI) methods detect signal from protons, which reside in either water or fat molecules. To detect and interrogate pathology, we need to determine which of these species the MRI signal has come from. If we can separate the fat and signals, we can accurately determine whether observed changes in spin density, relaxation times or diffusion characteristics (amongst others) affect water, fat or both, and therefore correctly interpret the significance of these changes [1]. If we can separate these signals, we can also quantify the proportion of fat in the tissue, as well as obtaining ‘corrected’ measurements of other tissue properties, such as relaxation times or diffusion coefficients. Changes in fat content are central to pathological processes in a wide variety of disease processes and organs, including hepatic steatosis [2–4] as well as disorders of the pancreas, muscle and bone marrow [5–10], meaning that MRI measures of fat content have become increasingly important in clinical imaging and research.

2. How can we measure tissue fat content?

Fat protons have several physical properties that differ from those of water protons, including relaxation times (particularly T1) and resonant frequency. These differences can be utilized for separating fat and water signals and measuring fat content. Most current methods for fat quantification rely on the difference in the resonant frequency of fat and water protons. This property can be exploited through various methods including chemical shift selective fat suppression (‘fat saturation’) [11], water- or fat-selective imaging [12,13], magnetic resonance spectroscopic imaging (MRSI) [14], magnetic resonance spectroscopy (MRS) [15] and chemical shift-encoded MRI (CSE-MRI) [16–25]. Of these methods, CSE-MRI has become increasingly popular and is now widely used as a clinical imaging tool.

3. How does chemical shift-encoded MRI work?

CSE-MRI methods differ in terms of input data, assumptions about the tissue and the algorithm for parameter estimation [16–25] . However, these methods have three common elements:

(i) Measurement of the signal at multiple echo times
Images are acquired with multiple echo times in order to manipulate the phase of the MRI signal. These echo times can include those at which the water and fat signals are completely ‘in phase’ and ‘out of phase’, where fat and water signals constructively and destructively interfere, respectively. However, signals may also be acquired at intermediate times when the signals are not fully in or out of phase. Varying numbers of echo times may be acquired (the number of echo times is commonly referred to as the number of ‘points’). The specific echo times that are chosen impact on the effectiveness of the subsequent fat-water separation. CSE-MRI can be implemented with a variety of MRI sequences, including both gradient echo and spin echo-based sequences. Multi-echo gradient echo sequences are commonly used for fat quantification in clinical practice.

(ii) A signal model
A mathematical model of the signal behavior is specified. In the simplest case, this model includes water and a single-peak model of fat, where triglycerides are assumed to contain protons with only a single frequency shift (in practice, this is an oversimplification). This model is typically extended to include a multipeak model (to more accurately reflect the composition of fat observed in vivo) and to include relaxation effects due to decay (again to more accurately reflect the in vivo situation).

(iii) An algorithm for estimating the desired parameter values
The algorithm is used to find the parameters of the signal model. This algorithm can take a variety of forms, ranging from simple analytic expressions to maximum likelihood estimation (MLE)-based methods, which find the parameter values that result in the closest ‘fit’ (lowest error / highest likelihood) with the signal measurements.

4. What is fat-water ambiguity, and how can this be resolved?

One of the key difficulties with CSE-MRI is that the signals from water-dominant (low fat fraction) voxels and fat-dominant (high fat fraction) voxels can be difficult to distinguish. In particular, the magnitude of the signal acquired from these two types of voxels can be almost identical. Therefore, some method for distinguishing the fat-dominant and water-dominant voxels is needed.

Using the phase of the MRI signal enables the signals from water-dominant and fat-dominant tissues to be distinguished; however, the phase is invariably ‘contaminated’ because inhomogeneity in the main magnetic field, B0, also affects the phase of the signal. This means that algorithms which only consider the information in a single voxel (so-called ‘voxel-independent’ methods) often fail. To address this, a number of image-based methods have been developed to simultaneously solve the fat-water separation problem across the entire image; assuming that the variation in the B0 field is smooth can enable correct separation of fat and water [20,21,26].

An alternative approach is to discard the phase and only use magnitude data. With magnitude data, fat-water ambiguity can be resolved on the basis that the complex fat spectrum introduces a subtle ‘fat-fat interference’ in fat-dominant tissues [27], which is not present in water-dominant tissues; comparing the likelihood of fat-dominant and water-dominant tissues allows the correct solution to be obtained [24,25].

5. How can we make measures of tissue fat content ‘quantitative’?

In order to make measures of fat content accurate, and to facilitate comparison across imaging platforms and between sites, it is necessary to measure the true concentration of water and fat in the tissue. However, fat fraction measurements can be confounded by other properties of the tissue and acquisition; these confounding factors include T1 and T2* relaxation, the multipeak nature of the fat spectrum, noise, phase errors and temperature. If we can remove these confounding factors, we can define the proton density fat fraction (PDFF) as the fat signal divided by the combined signal obtained from both water and fat.

Specific modifications can be made to the CSE-MRI method to correct for each confounding factor and enable PDFF measurement:

  • T1 bias - Use a small flip angle [28]
  • T2* bias - Incorporate T2* decay term into the signal model [29]
  • Multipeak fat spectrum - Incorporate multipeak fat into the signal model [30]
  • Noise bias - The denominator in the fat fraction equation is calculated by first summing the complex water and fat signals before taking the magnitude [28]
  • Phase errors - Use phase error correction schemes [31] or magnitude fitting [24,25,32]
  • Temperature - Adjust chemical shift in the signal model depending on temperature [33]

6. How can fat quantification be accelerated?

Acceleration techniques including compressed sensing and parallel imaging can be used to accelerate PDFF measurements for the purposes of trials [34–37], although clinical imaging may require more modest acceleration factors to maintain image fidelity. Techniques such as MRF and multitasking offer to accelerate the acquisition of multiparametric datasets, as discussed in Section 6.

7. Can we extend CSE-MRI to enable measurement of other tissue properties?

Depending on the acquisition, it is often straightforward to extend CSE-MRI to measure other properties. For example, with a multi-echo gradient echo acquisition, we typically already include R2* in the signal model. The resulting R2* estimate can be regarded as ‘fat-corrected’ and can be a useful parameter in its own right, for example for the quantification of iron [38] or bone mineral density [39].

The acquisition and signal models can be further extended. For example, CSE-MRI can be incorporated into Carr-Purcell-Meiboom-Gill acquisitions to enable measurement of T2 and fat fraction [8]. Similarly, multiecho acquisitions can be repeated with varying flip angles [40] or with presaturation pulses [41] to enable fat-corrected T1 measurement.

Magnetic resonance fingerprinting (MRF) has the potential to increase the number of contrasts available within a clinically-feasible timeframe. For example, MRF has been used to provide simultaneous T1, T2, T2* and FF maps in the liver [42], heart [43] and skeletal muscle [44]. MRI multitasking can enable acquisition of multiple parameters whilst also accounting for respiratory motion, which can be of value in organs such as the liver where such motion is problematic [45].

8. What is the role of deep learning in fat quantification?

Deep learning has been investigated as a method for reducing computational cost, increasing robustness to diversity and errors in source data, reducing dependence on initialization and thus reducing artefacts [46–49]. The methods published so far have used convolutional neural networks (CNNs) which take whole image data as input and make use of spatial relationships in the data [46–49], in a fashion analogous to the earlier image-based optimization techniques.

Acknowledgements

TJPB is supported by the NIHR UCLH Biomedical Research Centre.

References

1. Bley TA, Wieben O, Francois CJ, Brittain JH, Reeder SB. Fat and water magnetic resonance imaging. Journal of Magnetic Resonance Imaging. 2010;31(1):4–18.

2. Middleton MS, Van Natta ML, Heba ER, et al. Diagnostic accuracy of magnetic resonance imaging hepatic proton density fat fraction in pediatric nonalcoholic fatty liver disease. Hepatology. 2018;67(3):858–72.

3. Middleton MS, Heba ER, Hooker CA, et al. Agreement Between Magnetic Resonance Imaging Proton Density Fat Fraction Measurements and Pathologist-Assigned Steatosis Grades of Liver Biopsies From Adults With Nonalcoholic Steatohepatitis. Gastroenterology. 2017;153(3):753–61.

4. Noureddin, M, Lam, J, Peterson, MR, Middleton, M, Hamilton G, Le T. Utility of magnetic resonance imaging versus histology for quantifying changes in liver fat in nonalcoholic fatty liver disease trials. Hepatology. 2013;58(6):1930–40.

5. Yoon JH, Lee JM, Lee KB, et al. Pancreatic Steatosis and Fibrosis: Quantitative Assessment with Preoperative Multiparametric MR Imaging. Radiology. 2016;279(1):140–50.

6. Kühn J-P, Berthold F, Mayerle J, et al. Pancreatic steatosis demonstrated at MR imaging in the general population: Clinical relevance. Radiology. 2015;276(1):129–36.

7. Morrow JM, Sinclair CDJ, Fischmann A, et al. MRI biomarker assessment of neuromuscular disease progression: A prospective observational cohort study. Lancet Neurol. 2015;15(1):65–77.

8. Janiczek RL, Gambarota G, Sinclair CDJ, et al. Simultaneous T 2 and lipid quantitation using IDEAL-CPMG. Magn Reson Med. 2011;66(5):1293–302.

9. Bray TJP, Bainbridge A, Punwani S, Ioannou Y, Hall-Craggs MA. Simultaneous Quantification of Bone Edema/Adiposity and Structure in Inflamed Bone Using Chemical Shift-Encoded MRI in Spondyloarthritis. Magn Reson Med. 2018;79(2):1031–42.

10. Latifoltojar A, Hall-Craggs M, Bainbridge A, et al. Whole-body MRI quantitative biomarkers are associated significantly with treatment response in patients with newly diagnosed symptomatic multiple myeloma following bortezomib induction. Eur Radiol. 2017;27(12):5325–36.

11. Haase A, Frahm J, Hänicke W, Matthaei D. 1H NMR chemical shift selective (CHESS) imaging. Physics Med Biol. 1985;30(4):341–4.

12. Schick F, Machann J, Brechtel K, et al. MRI of muscular fat. Magn Reson Med. 2002;47(4):720–7.

13. Schick F. Simultaneous highly selective MR water and fat imaging using a simple new type of spectral-spatial excitation. Magn Reson Med. 1998;40(2):194–202.

14. Vidya Shankar R, Chang JC, Hu HH, Kodibagkar VD. Fast data acquisition techniques in magnetic resonance spectroscopic imaging. Vol. 32, NMR in Biomedicine. John Wiley and Sons Ltd; 2019.

15. Hamilton G, Middleton MS, Bydder M, et al. Effect of PRESS and STEAM sequences on magnetic resonance spectroscopic liver fat quantification. Journal of Magnetic Resonance Imaging. 2009 Jul;30(1):145–52.

16. Dixon WT. Simple proton spectroscopic imaging. Radiology. 1984;153(1):189–94.

17. Glover GH, Schneider E. Three-point Dixon technique for true water/fat decomposition with B0 inhomogeneity correction. Magnetic resonance in medicine : official journal of the Society of Magnetic Resonance in Medicine / Society of Magnetic Resonance in Medicine. 1991;18(2):371–83.

18. Reeder SB, Wen Z, Yu H, et al. Multicoil Dixon Chemical Species Separation with an Iterative Least-Squares Estimation Method. Magn Reson Med. 2004;51(1):35–45.

19. Reeder SB, Pineda AR, Wen Z, et al. Iterative decomposition of water and fat with echo asymmetry and least-squares estimation (IDEAL): Application with fast spin-echo imaging. Magn Reson Med. 2005;54(3):636–44.

20. Yu H, Reeder SB, Shimakawa A, Brittain JH, Pelc NJ. Field map estimation with a region growing scheme for iterative 3-point water-fat decomposition. Magn Reson Med. 2005;54(4):1032–9.

21. Hernando D, Kellman P, Haldar JP, Liang ZP. Robust water/fat separation in the presence of large field inhomogeneities using a graph cut algorithm. Magn Reson Med. 2010;63(1):79–90.

22. Berglund J, Johansson L, Ahlström H, Kullberg J. Three-point Dixon method enables whole-body water and fat imaging of obese subjects. Magn Reson Med. 2010;63(6):1659–68.

23. Bydder M, Yokoo T, Hamilton G, et al. Relaxation effects in the quantification of fat using gradient echo imaging. Magn Reson Imaging. 2008;26(3):347–59.

24. Triay Bagur A, Hutton C, Irving B, et al. Magnitude-intrinsic water–fat ambiguity can be resolved with multipeak fat modeling and a multipoint search method. Magn Reson Med. 2019;82(1):460–75.

25. Bray TJP, Bainbridge A, Lim E, Hall-Craggs MA, Zhang H. MAGORINO: Magnitude-only fat fraction and R*2 estimation with Rician noise modeling. Magn Reson Med. 2023 Mar 1;89(3):1173–92.

26. Cui C, Wu X, Newell JD, Jacob M. Fat water decomposition using globally optimal surface estimation (GOOSE) algorithm. Magn Reson Med. 2015 Mar 1;73(3):1289–99.

27. Bydder, M, Hamilton, G, Yokoo, T, Middleton, MS, Chavez, AD, Sirlin C. Fat-fat interactions in Dixon-variant imaging. In: Proc Intl Soc Mag Reson Med. 2007. p. 1632.28. Liu CY, McKenzie CA, Yu H, Brittain JH, Reeder SB. Fat quantification with IDEAL gradient echo imaging: Correction of bias from T1 and noise. Magn Reson Med. 2007;

29. Yu H, McKenzie CA, Shimakawa A, et al. Multiecho reconstruction for simultaneous water-fat decomposition and T2* estimation. Journal of Magnetic Resonance Imaging. 2007;26(4):1153–61.

30. Kühn J-P, Hernando D, Muñoz A, et al. effect of Multipeak spectral Modeling of Fat for liver iron and Fat Quantification: Correlation of Biopsy with MR Imaging Results 1. Radiology. 265.

31. Ruschke S, Eggers H, Kooijman H, et al. Correction of phase errors in quantitative water–fat imaging using a monopolar time-interleaved multi-echo gradient echo sequence. Magn Reson Med. 2017 Sep 1;78(3):984–96.

32. Hansen KH, Schroeder ME, Hamilton G, Sirlin CB, Bydder M. Robustness of fat quantification using chemical shift imaging. Magn Reson Imaging. 2012;30(2):151–7.

33. Navaratna R, Zhao R, Colgan TJ, et al. Temperature-corrected proton density fat fraction estimation using chemical shift-encoded MRI in phantoms. Magn Reson Med. 2021 Jul 1;86(1):69–81.

34. Loughran T, Higgins DM, McCallum M, et al. Improving highly accelerated fat fraction measurements for clinical trials in muscular dystrophy: Origin and quantitative effect of R2∗ changes. Radiology. 2015 May 1;275(2):570–8.

35. Mann LW, Higgins DM, Peters CN, et al. Accelerating MR imaging liver steatosis measurement using combined compressed sensing and parallel imaging: A quantitative evaluation. Radiology. 2016 Jan 1;278(1):247–56.

36. Lohöfer FK, Kaissis GA, Müller-Leisse C, et al. Acceleration of chemical shift encoding-based water fat MRI for liver proton density fat fraction and T2 mapping using compressed sensing. PLoS One. 2019 Nov 1;14(11).

37. Liu D, Steingoetter A, Parker HL, Curcic J, Kozerke S. Accelerating MRI fat quantification using a signal model-based dictionary to assess gastric fat volume and distribution of fat fraction. Magn Reson Imaging. 2017 Apr 1;37:81–9.

38. Hernando D, Kramer JH, Reeder SB. Multipeak fat-corrected complex R2* relaxometry: Theory, optimization, and clinical validation. Magn Reson Med. 2013;70(5):1319–31.

39. Lu F, Zhao YJ, Ni JM, et al. Adding liver R2* quantification to proton density fat fraction MRI of vertebral bone marrow improves the prediction of osteoporosis. Eur Radiol. 2022;

40. Le Ster C, Gambarota G, Lasbleiz J, et al. Breath-hold MR measurements of fat fraction, T1, and T2* of water and fat in vertebral bone marrow. Journal of Magnetic Resonance Imaging. 2016 Sep 1;44(3):549–55.

41. Thompson RB, Chow K, Mager D, Pagano JJ, Grenier J. Simultaneous proton density fat-fraction and R2∗ imaging with water-specific T1 mapping (PROFIT1): application in liver. Magn Reson Med. 2021 Jan 1;85(1):223–38.

42. Jaubert O, Arrieta C, Cruz G, et al. Multi-parametric liver tissue characterization using MR fingerprinting: Simultaneous T1, T2, T2*, and fat fraction mapping. Magn Reson Med. 2020 Nov 1;84(5):2625–35.

43. Jaubert O, Cruz G, Bustin A, et al. Water–fat Dixon cardiac magnetic resonance fingerprinting. Magn Reson Med. 2020 Jun 1;83(6):2107–23.

44. Marty B, Carlier PG. MR fingerprinting for water T1 and fat fraction quantification in fat infiltrated skeletal muscles. Magn Reson Med. 2020 Feb 1;83(2):621–34.

45. Wang N, Cao T, Han F, et al. Free-breathing multitasking multi-echo MRI for whole-liver water-specific T1, proton density fat fraction, and R2∗ quantification. Magn Reson Med. 2022 Jan 1;87(1):120–37.

46. Goldfarb JW, Craft J, Cao JJ. Water–fat separation and parameter mapping in cardiac MRI via deep learning with a convolutional neural network. Journal of Magnetic Resonance Imaging. 2019 Aug 1;50(2):655–65.

47. Andersson J, Ahlström H, Kullberg J. Separation of water and fat signal in whole-body gradient echo scans using convolutional neural networks. Magn Reson Med. 2019 Sep 1;82(3):1177–86.

48. Liu K, Li X, Li Z, et al. Robust water–fat separation based on deep learning model exploring multi-echo nature of mGRE. Magn Reson Med. 2021 May 1;85(5):2828–41.

49. Jafari R, Spincemaille P, Zhang J, et al. Deep neural network for water/fat separation: Supervised training, unsupervised training, and no training. Magn Reson Med. 2021 Apr 1;85(4):2263–77.

Proc. Intl. Soc. Mag. Reson. Med. 31 (2023)