The importance of fat fraction quantification and distribution in the human tissue is growing with the epidemic of obesity, and the increasing prevalence of diabetes mellitus type 2 (T2DM) [1, 2] and non-alcoholic fatty liver disease (NAFLD) [3–5].

Modern 3T clinical MR systems, providing high signal-to-noise ratio (SNR) and high resolution, allow the measurement of tissue proton density fat fraction (PDFF) [6, 7] even in routine clinical protocols. Multi-echo data is required to achieve accurate fat quantification based on Chemical Shift Encoded Magnetic Resonance Imaging (CSE-MRI) [8–13]. To determine of accurate PDFF distribution in the inner organs, high image resolution is required. Then, however, the SNR is reduced and subsequent data processing, in which the PDFF is calculated for each image voxel/pixel, is adversely affected. In principle, the SNR can be increased by the use of phased array RF (multi-channel) coils [14–16] and/or increased static magnetic field [17–19], both having some practical limits and incurring higher costs of hardware [20]. Recent installations of ultra-high-field (UHF) MR systems (7T or more), along with improvements in RF hardware and acquisition methods, have clearly demonstrated superior data quality for neuro- and musculoskeletal imaging [21, 22]. Nevertheless, problems remain; in particular, the abdominal region is affected by water-fat displacement and stronger susceptibility artifacts. Moreover, at high fields, the chemical shifts between the water peak and the multiple spectral peaks in the fat signal are increased. To minimize chemical shift displacement, strong gradients and large acquisition bandwidths must be used, causing SNR reduction and potentially substantial eddy currents [23, 24]. The other UHF effects are a prolongation of relaxation time T₁ and a shortening of T₂ and T2* relaxation times [25–27]. An important parameter in UHF MR for patient safety is the specific absorption rate (SAR), which, in principle, increases quadratically with B₀ field, but the specific spatial pattern depends in a complex way on the B₁ frequency and a variety of factors, including the subject [28, 29]. These factors are a problem mainly with whole-body or large-volume coverage coils, such as birdcage, saddle, and TEM coils. A related UHF-MR problem is inhomogeneity of the excitation RF-field (B1+), which leads to inhomogeneous excitation and errors in quantitative imaging if not properly addressed. To overcome the problems of heterogeneous SAR, excitation, and detection sensitivity, dedicated multi-channel RF coils in transmit and receive modes [30–33] have been used and field-specific adaptations to MR acquisition protocols have been implemented [34, 35].

CSE-MRI-based quantification of PDFF is a fast and reliable way to determine the distribution of fat in a tissue. This approach, proposed by Dixon [36] in 1984, has undergone considerable changes and evolution [8, 9, 13, 37, 38]. These various changes are frequently called “Dixon” methods even for multi-echo approaches; however, the term “CSE-MRI” is usually used for advanced methods that include a multi-echo acquisition scheme. Generally, the signal model representing the behavior of a vector of magnetization during the measurement sequence is crucial for the calculation of correct MRI-PDFF values. Due to the complex lipid spectrum [6, 39, 40], which contains several spectral lines in the frequency range of ~4.5 ppm (main peak at 1.3 ppm -CH₂-), the signal model must contain prior knowledge about the spectral position and the relative intensity of each peak. A further common problem of all “Dixon”/CSE-MRI water-fat separation methods is field map (B₀) estimation [11, 41]. At higher static magnetic fields, larger local magnetic field gradients result in more phase overflows of 2π in complex images compared to those at lower fields. The occurrence of these phase wraps may lead to water-fat swaps in the calculated MRI-PDFF maps, necessitating the application of phase-unwrapping algorithms. Several approaches have addressed this problem successfully [37, 41]. In hepatic iron overload, especially at excessive concentration, the tissue signal is significantly dephased, and hence, T2* relaxation is distinctly shortened; thus, the R2* = 1/T2* must be included in the signal model [8]; otherwise, substantial errors in PDFF estimation [42, 43] may occur.

In fact, the current clinical 3T protocols for PDFF measurements collect data in low resolution due to the requirement for a short measurement time (breath-hold) and a good SNR. The full coverage of the abdominal space is standard at 3T; however, the lower resolution smooths the information about the fat or iron distribution in the liver. This is a potential problem for the detailed study of fat distribution in the liver. Higher resolution that prolongs measurement time influences other measuring parameters (TE, TR, BW, and many others), and mainly deteriorates SNR in acquired data, which leads to noisy parametric maps (PDFF, R2^*). Generally, quantitative magnetic resonance imaging (qMRI) is a crucial component of the many therapies and diagnostic [44], and provides the relatively stable and reproducible results [45]. The using of high field potentially yield benefit in the form of higher SNR compare to a low field, and it can improve the qMRI [46]. We have to note at the outset that due to incomparable coil configuration, the quality of images is not compared (SNR) in the study (it would not be objective); and the main focus is on the comparison of the quantitative results from both magnetic field, 3 and 7T. This study is the first step in exploring the possibilities of abdominal quantitative MRI (qMRI) that could provide improvement in diagnostic accuracy for a wide range of chronic liver diseases due to higher sensitivity at 7T in combination with the appropriate hardware equipment.

The purpose of this study was to assess the feasibility of proton density fat fraction (PDFF) quantification using multi-echo MRI at 7T with a 32-channel phase-array coil without B₁ shimming. To demonstrate the potential value of the method at UHF with the best possible available hardware configuration at our institution (at the time of the study), PDFF derived from 7T MRI measurement (MRI-PDFF) was compared to PDFF determined by 7T MRS (MRS-PDFF) and to gold-standard [47, 48] multi-echo MRI-based 3T measurements [49, 50] on the same group of subjects.

Examples of the acquired MRI-PDFF volumes for two subjects with different body sizes and composition, and with high (S6) and low (S3) BMI, at 3 and 7T, are shown in Figure 1 (and related figures of PDFFs with corresponding 7T anatomical images in Supplementary Material 1). The position of the spectroscopic volume of interest (VOI) and the ROI used for the comparison is delineated for each subject by the white box in the respective PDFF map; the calculated field maps are shown in Supplementary Material 2. The effective liver volume coverage for MRI-PDFF showed 100% of the liver volume, as well as the whole abdomen for all subjects at 3T. In the case of the 32-channel RF coil used at 7T, the effective liver volume coverage varied from ~30 to 90% based on body size and composition. Liver coverage was highest in the subjects with low BMI and smaller body (torso) size (example in Figure 1 S13, Supplementary Figures 1, 3 in Supplementary Material 1). A case of a subject with very high BMI and non-optimal body size is shown in Figure 1 S6 and Supplementary Figures 1, 2 in Supplementary Material 1. The whiskers diagrams (Figure 2) show the distributions of MRI-PDFF within the VOI and the related MRS-PDFF values acquired from similar or almost identical volumes.

The correlation analyses for the data presented in Figure 2 are shown in Figure 3, where the 7T-MRI-PDFFs are compared with (Figure 3A) 3T-MRI, and (Figure 3B) 7T-MRS PDFF values. In the both cases A and B (Figure 3), high R²-values with low p-values, indicating strong agreement between the 3T and 7T measurements, were observed. The calculated R²-values (with p-values) of A and B (Figure 3) cases were 0.936 (p ≈ 2.75·10⁻⁷; 7T MRI vs. 3T MRI) and 0.970 (p ≈ 1.32·10⁻⁶; 7T MRI vs. 7T MRS), respectively; in addition, the slopes were calculated. Then, the Bland-Altman (BA) analyses of the previous (A) and (B) cases were also performed, which can be seen in Figure 4. In the first case, the BA plot shows a mean difference (bias) of ≈-0.60% and upper and lower confidence intervals (CI of 95%) of 0.58 and −1.79%, respectively. In the second case (Figure 4 on the right), there was a mean difference (bias) of ≈-0.54% and upper and lower confidence intervals (CI of 95%) of 0.52 and −1.60%, respectively. The extended boxplots and related analyses of PDFF maps that were calculated by the offline toolbox using Graph-Cut algorithm/approach from 3T-MRI data are given in Supplementary Material 3.

In subject S2, the 7T MRS measurement was not performed because of health problems (cramps) not related to the measurement. The reconstructed (7T) MRI-PDFF maps (mainly the volume of interest) from S10 were affected by strong water-fat swaps that could not be removed by changing the parameters of the algorithm used, and therefore, the results were excluded. The 7T spectra from subjects S3 and S9 were not included in the analysis due to movement artifacts in the data.

In our study, we tested the feasibility of MRI-PDFF assessment at 7T. Even-though MRS was considered for the “reference method” in hepatic fat content quantification, recent developments in MRI-PDFF at 3T have substantially improved the accuracy of the approach and made it the method of choice for hepatic fat content quantification [47, 48] in clinical settings.

The results depicted in Figure 2 demonstrate an agreement between the MRI-PDFF distributions from 3T and 7T measurements in the investigated volumes. In most cases, the spectroscopic values were within the 25th and the 75th percentile of MRI-PDFF. In only one case was the MRS value not in this interval (Figure 2), probably due to subtle motion of the subject between the MRS and MRI measurements. Another possible deteriorating effect is intra-voxel inhomogeneity, due to movement-related changes in local B₀, which can lead to an improper estimation of PDFF. In 7T, MRI-DFF distributions show a larger range of values, and the number of outliers (red crosses, Figure 2) is higher than at 3T. This can be caused by the inhomogeneity of B1- over the FOV if the investigated volume is relatively distant from the coil surface, and hence, the water-fat separation process is more prone to errors due to lower SNR.

Correlation analyses confirm good agreement between MRI-PDFF 7T and 3T measurements in the investigated volumes. The level of agreement is attributable to measurement conditions, such as the measurement sequences (minimizing the acquisition time), patient measurement management (short time period between the 3T and 7T measurements–not more than 1 h), and exhalation breath-hold. From our practical experience, the bias in BA-plots appears relatively low, given that we are comparing measurements at different magnetic field. Although initial preliminary studies concerning hepatic liver fat quantitation by MRI at 7T have already been published [64, 65], in this study, a larger number of subjects was investigated, the distribution of MRI-PDFF in the investigated volumes was analyzed and compared to already established 7T MRS, and 3T MRI measurements were performed.

Our results confirm the feasibility of hepatic fat content quantification by MRI-PDFF based on a multi-gradient-echo acquisition method at ultra-high field (7T) using a 32-channel Tx/Rx array coil. In addition to a coil configuration and other hardware equipment, the success of fat quantification using MRI-PDFF is based on the water-fat separation algorithm, including prior knowledge of the fat spectral model. In the presence of rapid field changes at 7T, a robust solution that avoids or at least minimizes water/fat swaps in the reconstructed MRI-PDFF maps (most importantly in the regions of interest) is required. The advanced coil configuration with the further envisioned hardware improvement will provide the opening for further improvement of whole-abdomen imaging and liver fat quantification for patients with a higher BMI.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by Ethics Committee, Medical University of Vienna. The patients/participants provided their written informed consent to participate in this study.

RK and MKrš designed the study, processed data, drafted the initial manuscript, reviewed and revised the manuscript, and coordinated and supervised data collection. RK, LP, HB, and MKre performed the measurements and ensured the patients for the study. KE and SR are responsible for ASPIRE reconstruction. MKrš, ZS, ST, and SR critically reviewed the manuscript for important intellectual content. All authors approved the final manuscript as submitted and agree to be accountable for all aspects of the work.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.