Simultaneous comprehensive liver T₁, T₂, , T_1ρ, and fat fraction characterization with MR fingerprinting

1 INTRODUCTION

Chronic liver disease is defined as a progressive destruction of liver tissue over a period greater than 6 months. One of the main causes of chronic liver disease, with a 25% prevalence in the world’s population,¹ is nonalcoholic fatty liver disease. This disease can progress through different stages, from inflammation (nonalcoholic steatohepatitis or NASH) to hepatocellular carcinoma, passing through liver fibrosis and liver cirrhosis.² Liver biopsy, despite its invasiveness, is considered to be the current gold standard for staging and assessment of liver fibrosis. Still, many factors such as sampling errors and interrater variability can affect the accuracy and diagnostic value of this technique.^3-5 To overcome the limitations of liver biopsy, many noninvasive alternatives have been proposed.^{6, 7} Among these, quantitative MRI arises as a promising imaging tool for the assessment and follow-up of liver disease. Particularly, quantitative maps of T₁ and T₂ (for fibrosis and inflammation detection^{8, 9}), (for hepatic iron content visualization¹⁰), and fat fraction (FF, for hepatic lipid content assessment^{11, 12}) are promising imaging biomarkers for the assessment and follow-up of liver disease. Multiparametric information is conventionally obtained by performing sequential scans during different breath-holds,¹³ which leads to long scan times, bias due to inter-parameter dependencies, patient discomfort, and potential multiparametric image misalignments due to subject’s bulk and respiratory motion. Many efforts have been undertaken to develop multiparametric methods that provide joint parameter mapping in a single scan.^{14, 15} In recent years, MR fingerprinting (MRF)¹⁶ has shown to be a powerful technique to provide multiparametric information in abdominal scans.^17-20 Particularly, T₁, T₂, and B₁ maps have been obtained with MRF in a single breath-hold scan of about 19 seconds.¹⁷ We recently proposed an MRF framework able to quantify T₁, T₂, , and fat fraction (FF) from a single 14-second breath-hold scan,²⁰ whereas simultaneous T₁, T₂, and T_1ρ MRF relaxation mapping was recently proposed by Wyatt et al.¹⁸ T_1⍴ mapping has been shown to be of importance in liver fibrosis assessment^{21, 22} and function.²³ T_1⍴ relaxation correlates with physical processes that take place within the low-frequency range (0.1-1 kHz),²⁴ which are related to interactions between bulk water and extracellular matrix macromolecules, such as collagen or proteoglycans.^25-27 Therefore, it could provide additional information to standard T₁ and T₂ imaging regarding macromolecular composition of liver tissue, especially when these three relaxation parameters are measured in the same scan. A comprehensive scan that is able to simultaneously quantify not only T₁, T₂, and T_1ρ (as proposed by Wyatt et al¹⁸), but also providing and FF quantification, is desired.

In this work, we build on our previous work²⁰ to propose a simultaneous co-registered T₁, T_2,, T_1ρ, and FF abdominal MRF approach to enable a fully comprehensive liver-tissue characterization in a single breath-hold scan.

2 METHODS

2.1 Magnetic resonance fingerprinting acquisition scheme

An eight-echo gradient-echo (GRE) liver MRF sequence for simultaneous quantification of T₁, T₂, , T_1⍴, and FF is proposed (Figure 1), based on our previous work.²⁰ The acquisition includes two inversion-recovery (IR) (TI = {1,100} ms) pulses, five T₂-prepared pulses (TEs = {30,60} ms) with adiabatic refocusing pulses, five spin-lock (SL) pulses (SL times [SLT] = {30,60} ms) of SL amplitude (SLA) = 350 Hz, and six additional blocks in which no magnetization preparation is applied, leading to a total of 18 acquisition blocks of 507-ms duration each. Acquisition blocks are spaced regularly every 1 second, allowing for magnetization recovery before the next block. The T_1ρ preparation pulses robust to both B₁ and B₀ inhomogeneities were considered.^{28, 29} Acquisition parameters included golden-angle radial (~111°) readout; fixed low flip angle (15°); TR/TE₁/ΔTE = 16.9/1.6/1.95 ms (in which all of the echoes are sampled along the same radial spoke); FOV = 350 × 350 mm²; voxel size = 2 × 2 mm²; slice thickness = 8 mm; number of readouts per block = 30; one time-point image per radial readout, leading to a total of 540 timepoints; and scan time of 18 seconds. Acquisitions were performed on phantoms and 10 healthy subjects on a 1.5T scanner (Ingenia, Philips Healthcare).

3 RESULTS

3.1 Phantom study

Correlation analyses for the average T₁, T₂, , T_1⍴, and FF maps of MRF against their respective reference measurements (i.e., T₁-IRSE, T₂-multi-echo SE, eight-echo GRE, T_1⍴-prepared SE, and six-echo GRE) are shown in Figure  2A. The corresponding Bland-Altman plots are shown in Figure 2B. Linear fits show excellent coefficients of determination (r² > 0.9) for each parameter, while a zero bias is always included in the 95% confidence interval of their respective Bland-Altman analyses. Best fits, coefficients of determination, biases, and limits of agreement are reported in Supporting Information Table S2. Mean ± SD values measured inside each vial are reported in Supporting Information Table S3. Mean ± SD T₁ and T₂ MRF values compared with T₁-MOLLI and T₂-GRASE are reported in Supporting Information Table S4.

3.2 In vivo study

A comparison between the maps obtained with the proposed MRF framework and reference maps is shown in Figure 3 for 2 representative healthy subjects and in Figure 4 for 2 subjects who presented with previously diagnosed mild hepatic steatosis. This mild elevation of fat percentage in liver tissue can be observed in the proposed FF-MRF map (13.3 ± 4.4% and 20.5 ± 2.7% for each subject, respectively) and was confirmed by the reference six-echo GRE in both cases (16.4 ± 5.3% and 18.6 ± 10.5% for each subject, respectively). The quantitative analysis of MRF-derived T₁, T₂, , T_1⍴, and FF values measured within ROIs manually drawn inside the liver, spleen, erector spinae muscle, and subcutaneous fat is shown in Figure 5A, and the corresponding values in Supporting Information Table S5. The average values measured within each ROI are statistically different from the reference scans (p < 0.05) for T₁, T₂, and T_1⍴ but not for and FF, whereas correlation between MRF and reference values (shown in Figure 5B) is excellent for each parameter (r² = 0.95, r² = 0.97, r² = 0.92, r² = 0.98, and r² = 0.99 for T₁, T₂, , T_1⍴, and FF, respectively).

4 DISCUSSION

An eight-echo MRF approach is proposed to enable simultaneous (inherently co-registered) and comprehensive water-specific T₁, T₂, , and T_1⍴ maps and FF liver maps in a single breath-hold scan. The framework relies on a radial multi-echo readout to estimate and B₀ and separate the signal in water-based and fat-based fingerprints, and on IR, T₂, and T_1⍴ preparations to encode T₁, T₂, and T_1⍴. The number of echoes was chosen based on previous optimization studies shown in Jaubert et al.²⁰ The method provides a set of reconstructed multi-echo singular images, from where water-fat separation is performed. Subsequently, T₁, T₂, and T_1⍴ parametric maps are estimated by matching the water-based fingerprints to a precalculated dictionary, whereas FF is estimated from the water and fat M₀ maps. Because the multiparametric quantification is performed on several tissues, including liver, the dictionary T₁, T₂, and T_1⍴ combinations were chosen to represent a broad range of combinations that covered those tissues. Different to our previous work,²⁰ five SL pulses are applied here with different SLT in order to provide additional T_1⍴ encoding. The SL pulses used in this work are totally balanced,^{29, 44, 45} which have been shown to be robust to B₁ and B₀ inhomogeneities.^46-48 A SL amplitude of SLA = 350 Hz was chosen for these experiments. The SLA is known to have a positive correlation with maximum contrast achievable between healthy and fibrotic tissue^{49, 50}; hence, the application of SL pulses at the highest possible SLA is desired. However, several factors limit the maximum SLA achievable, including the maximum power of the RF amplifiers and specific absorption rate restrictions, which are especially stringent in abdominal scans. To address these limitations, a compromise between maximum SLA and maximum SLT applied must be found. Therefore, an SLA = 350 Hz was chosen to make SLT = 60 ms feasible, despite scanner and specific absorption rate limitations.

Phantom experiments showed excellent agreement between MRF and reference measurements for all parameters. The T_1⍴ and T₂ quantitative results in the agar-based phantom were similar between them for both reference and MRF measurements. This is in line with previous MRF-based studies^{18, 51, 52} and indicate that the mechanisms that model T_1⍴ and T₂ relaxations are the same in agar-based phantoms.

In vivo experiments demonstrated the feasibility of the proposed MRF approach to provide simultaneous quantification of T₁, T₂, , T_1⍴, and FF in a single breath-hold scan of 18 seconds. The MRF parametric maps (as those shown in Figures 3 and 4) presented similar, if not better, visual image quality than the conventional mapping approaches, with the added value of being inherently co-registered and obtained in a single scan, thus reducing considerably the scan time from five acquisitions (plus their respective preparation times) to just one. Considering the quantitative results, and FF showed a very good correlation and low bias between MRF and the conventional mapping techniques, with MRF measurements not significatively different from the reference values. Very good agreement was also found in FF quantification; this was also observed in the liver FF values measured with the proposed MRF and the conventional proton density FF approach in 2 subjects who presented with mild hepatic steatosis. T₁ values obtained with MOLLI were slightly lower than those obtained with the proposed MRF framework, whereas T₂ and T_1⍴ values measured with the conventional mapping approaches were higher than those measured with the proposed MRF. The MOLLI technique has a known negative bias, and this underestimation has been previously reported.^{20, 53} This underestimation is also observed in the phantom study, as shown in Supporting Information Figure S1. The MRF underestimation bias for T₂ has also been previously reported in several studies.^{18, 20, 34, 39, 54, 55} Diffusion effects⁵⁶ or imperfect T₂ relaxation during the adiabatic T₂ preparation pulses could be some of the sources of this underestimation. A clinical reference technique for T_1⍴ quantification does not currently exist in abdominal imaging. To enable in vivo liver T_1⍴ validation, a fast, single-breath-hold T_1⍴ reference scan, similar to the one used by Wyatt et al,¹⁸ was used as reference; however, this approach has its own limitations and may present a bias. Liver T_1⍴ values obtained with this reference technique were 58.0 ms (54.1-65.8 ms) expressed as median (first to third quartiles), which are higher than those reported for healthy livers at 1.5 T by Allkemper et al²² (T_1⍴ = 40.9 ± 2.2 ms) and Kritsaneepaiboon et al⁵⁷ (T_1⍴ = 54.7 [49.6-56.6]). The T_1⍴ values measured with the proposed MRF approach were 41.8 ms (38.9-47.1 ms), which are in agreement with those reported by Allkemper et al.²² However, further studies are still required to determine the degree of bias of the proposed MRF approach with respect to the ground-truth T_1⍴ values. Finally, T_1⍴ values measured with the proposed MRF approach were higher than the T₂-MRF values, as expected in liver tissue and observed¹⁸ in previous liver MRF studies.

This study has several limitations. The proposed MRF approach is based on a previously introduced MRF sequence for T₁, T₂, , and FF quantification, which has been extended to further include T_1⍴ encoding. However, although simulations were performed (data not shown) to heuristically optimize SLT and the position of SL pulses in the sequence, an extensive optimization study (e.g., using Cramer-Rao lower bound) would be required to find the optimal configuration and will be investigated in future studies. Additionally, the proposed approach requires the subject to hold their breath for 18 seconds. Although this breath-holding duration is usually feasible, this may be challenging for some patients; thus, efforts to reduce the scan time will be considered in future studies. Regarding the amplitude of the SL pulses, it is well known that there is a positive correlation between SLA and T_1⍴, known as T_1⍴ dispersion. Although T_1⍴ dispersion has been previously studied for cardiac,⁵⁰ muscle,⁵⁸ cartilage,⁵⁹ and brain⁶⁰ imaging, to the best of our knowledge, T_1⍴ dispersion has not been studied on the human liver yet. Hence, the combination of SLA and SLTs used in this work (SLA = 350 Hz for maximum SLT = 60 ms) may be suboptimal, and a better compromise between RF and specific absorption rate limitations and T_1⍴ encoding capability of the sequence could be yet to be found. In addition, this T_1⍴ preparation approach fails to handle the chemical shift in fat, as it ignores the different behavior that each fat peak would have with respect to the SL pulse. However, the method proposed here is intended to provide water-specific T₁, T₂, , and T_1⍴ maps, considering fat tissues only through FF quantification. Hence, quantification of T_1⍴ in fat is not performed. In the case in which a nonnegligible amount of fat co-exists with water in the same voxel (such as in hepatic tissue that presents steatosis), a small bias may still be introduced due to fat-signal distortion if the SL pulses used do not effectively lock the spins of the off-resonant frequencies of fat. Alternative strategies for SL, such as adiabatic SL T_1⍴ preparation pulses, shall be explored to deal with chemical shift at higher fields and multipeak encoding⁶¹; and future studies are warranted in this direction. Spatial resolution in this study (2 × 2 mm², slice thickness of 8 mm) may be considered in the lower end for liver imaging. Optimal sequence design for this acquisition could lead to more efficient parametric encoding/scan efficiency, thus allowing a higher resolution. In addition, further development of stronger regularizers or deep-learning reconstruction methods could also enable higher resolution. The use of a multi-echo SE sequence for reference T₂ mapping in the phantom study is also acknowledged as a limitation, as this technique is sensitive to effects like stimulated echoes. In the future, single-echo SE sequences will be used instead to remove this possible source of bias. Another limitation of this proof-of-concept study is the evaluation in phantom and 10 healthy subjects only. Future studies should be performed in patients with liver disease to evaluate the ability of the proposed MRF approach to provide fully comprehensive liver characterization in a single-breath-hold scan.

5 CONCLUSIONS

The proposed eight-echo IR, T₂, and T_1⍴ prepared liver MRF sequence presented in this work allows for quantitative T₁, T₂, , T_1⍴, and FF liver-tissue characterization in a single-breath-hold scan of 18 seconds. This approach has been investigated in phantoms and healthy subjects, showing good agreement and excellent correlation with reference SE measurements and conventional clinical maps. The additional information provided by T_1⍴ quantification may improve the diagnostic power of multiparametric liver tissue characterization, especially in the presence of liver fibrosis. Future work will aim to validate the proposed approach in patients with liver disease.