Original Research
Gastrointestinal Imaging
January 23, 2015

Quantification of Hepatic Steatosis With a Multistep Adaptive Fitting MRI Approach: Prospective Validation Against MR Spectroscopy

Abstract

OBJECTIVE. The purpose of this study is to prospectively compare hybrid and complex chemical shift–based MRI fat quantification methods against MR spectroscopy (MRS) for the measurement of hepatic steatosis.
SUBJECTS AND METHODS. Forty-two subjects (18 men and 24 women; mean ± SD age, 52.8 ± 14 years) were prospectively enrolled and imaged at 3 T with a chemical shift–based MRI sequence and a single-voxel MRS sequence, each in one breath-hold. Proton density fat fraction and rate constant (R2*) using both single- and dual-R2* hybrid fitting methods, as well as proton density fat fraction and R2* maps using a complex fitting method, were generated. A single radiologist colocalized volumes of interest on the proton density fat fraction and R2* maps according to the spectroscopy measurement voxel. Agreement among the three MRI methods and the MRS proton density fat fraction values was assessed using linear regression, intraclass correlation coefficient (ICC), and Bland-Altman analysis.
RESULTS. Correlation between the MRI and MRS measures of proton density fat fraction was excellent. Linear regression coefficients ranged from 0.98 to 1.01, and intercepts ranged from −1.12% to 0.49%. Agreement measured by ICC was also excellent (0.99 for all three methods). Bland-Altman analysis showed excellent agreement, with mean differences of −1.0% to 0.6% (SD, 1.3–1.6%).
CONCLUSION. The described MRI-based liver proton density fat fraction measures are clinically feasible and accurate. The validation of proton density fat fraction quantification methods is an important step toward wide availability and acceptance of the MRI-based measurement of proton density fat fraction as an accurate and generalizable biomarker.
MRI measures of proton density fat fraction are becoming important and generally accepted tools in the evaluation of hepatic steatosis [13]. Nonalcoholic fatty liver disease is associated with diabetes, obesity, and hypertension and is estimated to affect up to 80 million Americans [4]. Nonalcoholic fatty liver disease has been shown to be a risk factor for malignancy, cardiovascular disease, and sudden death [58]. Proton density fat fraction, the ratio of fat protons to the sum of fat and water protons in a given volume, is an MRI-based measure of liver fat content that has been strongly correlated with histopathologic measures of steatosis [9, 10]. This important measure is gaining traction in the research community as a biomarker in chronic liver disease. Liver fat quantification may also have a role in oncology, where hepatic steatosis has been shown as a marker of liver injury from certain chemotherapeutic agents [11, 12].
Over the last few decades, substantial advances have been made by the scientific community in developing accurate robust measures of the degree of hepatic steatosis [13, 1330]. It has been well shown that it is possible to measure the severity of liver fatty metamorphosis; however, many of these techniques are technical, require offline processing, and are not widely available in clinical practice [1321]. In fact, in the United States, only one commercial vendor has received approval from the Food and Drug Administration for an MRI-based fat quantification method [2]. For proton density fat fraction to be considered a generalizable biomarker, it must be broadly available from multiple vendors, on multiple hardware platforms. Quantitative methods face a particular challenge as commercial platforms move toward wide-bore (70 cm) MRI systems, which may have difficulties with field inhomogeneity and eddy currents, two important confounders in proton density fat fraction measurement, particularly for methods based on complex-number calculations [26].
The purpose of this study was to prospectively compare MRI-based measures of proton density fat fraction against reference standard MR spectroscopy (MRS) for the quantification of hepatic steatosis severity, which are implemented inline on a commercially available wide-bore MRI system.

Subjects and Methods

Subjects

The institutional review board approved this prospective HIPAA-compliant study. Written informed consent was obtained from all subjects. The author who is not an employee of Siemens Healthcare had control of data and information that might have presented a conflict of interest for the duration of the study.
Forty-two subjects (18 men and 24 women) were consecutively enrolled who presented for clinical abdominal MRI, predominantly for either chronic liver disease or oncologic indications. Subject age, weight, and body mass index (weight in kilograms divided by the square of height in meters) were recorded, and relevant clinical data were collected from the electronic medical record. It should be noted that data from a subset of these subjects (n = 30) were used in a recent technical description of one of the imaging methods presented here [31]. The current work differs from that manuscript in that it is a clinical rather than technical validation with differences in data analysis, involves a larger cohort of subjects, and evaluates or compares multiple image reconstruction algorithms.

MRI

All imaging was performed on one of two identical 3-T clinical MRI systems (Magnetom Skyra, Siemens Healthcare), using anterior 18-channel flexible array coils in combination with the table-mounted spine coil array. Before IV contrast material administration, a whole-liver volume acquisition was performed using a six-echo 3D spoiled gradient-echo acquisition. Two-dimensional parallel acceleration was used to allow whole liver coverage in a single breath-hold (controlled aliasing in parallel imaging results in higher acceleration) [32].
Imaging parameters included TR/first TE of 8.9/1.23, echo spacing of 1.23 ms, flip angle of 4°, receiver bandwidth of 1085 Hz/pixel, FOV of 42 × 32.8 × 24 cm, acquisition matrix of 256 × 160 × 50 interpolated to an image matrix of 256 × 200 with 60 slices, spatial resolution of 1.6 × 2.1 × 4 mm3, parallel imaging factor of 2 × 2, and acquisition time of 21 seconds. In particular, the work of Levin et al. [33] has shown that the use of inand opposed-phase echo pairs can provide accurate proton density fat fraction measurements, and on the basis of the work of Johnson et al. [34], we think that a flip angle of 4° combined with a TR of 8.9 would minimize T1-related effects while providing adequate signal-to-noise ratio (SNR).

Image Reconstruction

Inline image reconstruction was performed using a multistep adaptive fitting algorithm that has been previously described [31]. In brief, this hybrid magnitude-complex technique uses the Levenberg-Marquardt fitting algorithm to solve for the values of water signal intensity (SI), fat SI, and rate constant (R2*), according to a model adapted from that described by Yu et al. [21], and allows the independent estimation of the R2* of fat and R2* of water, first described by Chebrolu et al. [35]. Our model corrects for the R2* of water, the R2* of fat, and the complex multipeak spectrum of fat, as first described by Yu et al. [21] and first shown in the liver by Reeder et al. [36]. A published multipeak spectral model of fat was incorporated as an a priori model, rather than modeling the fat peak spectrum directly, to reduce the number of free parameters in the model; a number of other fat peak models have also been described [21, 31, 37, 38]. Two reconstructions were performed inline: image set A, with water R2* assumed equal to fat R2* (termed the effective R2*), similar to past descriptions of R2* correction [21, 3840]; and image set B, with the R2* values of fat and water assumed to be independent of one another, also as previously described [35]. Precise reconstruction times were not recorded; however, whether the combined reconstruction required more than 4 minutes was assessed, beginning from the end of the pulse sequence acquisition to the time of arrival of the final image in the image archive.
A second image reconstruction was performed offline using the raw datasets from the above acquisitions (image set C) [31]. This complex-based method first determines the field map on a lower resolution based on variable projection, formulated independently of eddy current effects, and then estimates the phase map between images acquired with opposed readout polarity for correction of undesired effects from eddy currents or gradient delays. Next, fitting is performed to the phase corrected complex data to obtain the R2* value, which is assumed to be equal for fat and water in this method, as well as water SI and fat SI.

MR Spectroscopy Technique

A single-voxel MRS acquisition was performed as the reference standard for proton density fat fraction measured with the MRI techniques. A 2.0 × 2.0 × 2.0 cm3 voxel was placed in a homogeneous portion of the mid-to-posterior liver on the basis of scout images, avoiding large vessels, bile ducts, masses and other obvious abnormalities. A stimulated echo acquisition mode acquisition was performed, with the following parameters: TR, 3000; TE, 12, 24, 36, 48, and 72; mixing time, 10 ms; receiver bandwidth, 1200 Hz; 1024 readout points; and breath-hold time, 15 seconds. Reference standard fat fraction values were calculated inline with separate R2 correction of both fat and water peaks, extrapolated to a TE of 0, using the high-speed multiple-echo acquisition method [41].

Image Analysis

Image analysis was performed by a single board-certified abdominal radiologist with 3 years of postfellowship experience in abdominal MRI. Measurements were performed using OsiriX MD (version 1.3, Pixmeo Sarl). Measurements were performed on MR images using 2.0 × 2.0 × 2.0 cm3 volumes of interest (VOIs). Colocalization was performed using the reference image generated by the MRI console, showing the location of the MRS voxel overlaid on the fat fraction images according to the location of the voxel in the DICOM metadata. The VOI was then copied to all fat fraction and R2* series, and mean values were calculated for each VOI.

Statistical Analysis

All statistical analyses were performed using SPSS software (version 20.0, IBM). The proton density fat fraction values measured using each of the MRI techniques were independently compared against the MRS values using a variety of statistical methods. Intraclass correlation coefficients (ICCs) with 95% CIs were calculated using a twoway mixed model to assess agreement between MRI-based fat fraction values and MRS-based values. Linear regression was performed to determine the correlation (R2), slope, and intercept as measures of correlation between MRI-based and MRS-based measures. Bland-Altman analysis was also performed to assess the degree of systematic and random bias between the MRI measures of proton density fat fraction and reference standard MRS. For comparisons that provided p values, p < 0.05 was considered statistically significant.
Because MRS measures R2, not R2*, no reference standard was available for validation of the MRI-derived measurements of R2*. To evaluate agreement among the MRI methods, we calculated ICCs using a two-way mixed model for each pairing, as well as all three of the following R2* values: effective R2* from image set A, water R2* from image set B, and effective R2* from image set C. Finally, we performed a linear regression analysis for all three MRI methods between R2* (effective or water, as appropriate) and proton density fat fraction values to determine whether they were related.

Results

Forty-two subjects (18 men and 24 women) were enrolled in this study, with a mean age of 52.8 ± 14 years (range, 20–80 years). The mean subject weight was 85.3 ± 22.4 kg (range, 49.9–163.3 kg), and mean body mass index was 29.3 ± 5.9 (range, 17.8–46.2). Twenty-two subjects were being imaged for liver metastasis, 14 for chronic liver disease or hepatocellular carcinoma, three for indeterminate liver lesions, two for pain, and one to evaluate the cause of biliary obstruction. By MRS, proton density fat fraction values ranged from 1.9% to 30.3%; 23 of 42 subjects (55%) had at least mild hepatic steatosis, defined as proton density fat fraction greater than 5.56%.
Figures 1 and 2 show representative images from subjects with hepatic steatosis and normal proton density fat fraction, including proton density fat fraction and R2* maps reconstructed using all three methods. There is slightly greater central noise in proton density fat fraction images generated using the dual-R2* fitting method, likely reflecting the model's instability related to the greater number of degrees of freedom. Agreement with the reference standard proton density fat fraction values was excellent, and although no reference standard R2* value was available, the R2* values (effective or water R2*) calculated with the three methods agreed well with one another. Note that fat R2* values were extremely noisy, likely owing to the combination of model instability and low fat signal; this phenomenon has been described more completely in the work of Horng et al. [42]. Also note that the use of dual-R2* correction introduces spuriously high water R2* values in the subcutaneous and intraabdominal fat, which was also found by Horng et al.
Fig. 1A —50-year-old man with hepatic steatosis.
A, Representative images from MRI are shown. Proton density fat fraction (A) and effective R2* (D) were reconstructed using hybrid method with single R2* fitting. Proton density fat fraction (B) and water R2* (E) were reconstructed using hybrid method with dual R2* fitting. Proton density fat fraction (C) and effective R2* (F) were reconstructed using complex method. Fat R2* was reconstructed using hybrid method with dual R2* fitting (G); note severe noise throughout image. White squares show ROI locations, colocalized with single-voxel spectroscopy, which yielded reference proton density fat fraction value of 23.9%.
Fig. 1B —50-year-old man with hepatic steatosis.
B, Representative images from MRI are shown. Proton density fat fraction (A) and effective R2* (D) were reconstructed using hybrid method with single R2* fitting. Proton density fat fraction (B) and water R2* (E) were reconstructed using hybrid method with dual R2* fitting. Proton density fat fraction (C) and effective R2* (F) were reconstructed using complex method. Fat R2* was reconstructed using hybrid method with dual R2* fitting (G); note severe noise throughout image. White squares show ROI locations, colocalized with single-voxel spectroscopy, which yielded reference proton density fat fraction value of 23.9%.
Fig. 1C —50-year-old man with hepatic steatosis.
C, Representative images from MRI are shown. Proton density fat fraction (A) and effective R2* (D) were reconstructed using hybrid method with single R2* fitting. Proton density fat fraction (B) and water R2* (E) were reconstructed using hybrid method with dual R2* fitting. Proton density fat fraction (C) and effective R2* (F) were reconstructed using complex method. Fat R2* was reconstructed using hybrid method with dual R2* fitting (G); note severe noise throughout image. White squares show ROI locations, colocalized with single-voxel spectroscopy, which yielded reference proton density fat fraction value of 23.9%.
Fig. 1D —50-year-old man with hepatic steatosis.
D, Representative images from MRI are shown. Proton density fat fraction (A) and effective R2* (D) were reconstructed using hybrid method with single R2* fitting. Proton density fat fraction (B) and water R2* (E) were reconstructed using hybrid method with dual R2* fitting. Proton density fat fraction (C) and effective R2* (F) were reconstructed using complex method. Fat R2* was reconstructed using hybrid method with dual R2* fitting (G); note severe noise throughout image. White squares show ROI locations, colocalized with single-voxel spectroscopy, which yielded reference proton density fat fraction value of 23.9%.
Fig. 1E —50-year-old man with hepatic steatosis.
E, Representative images from MRI are shown. Proton density fat fraction (A) and effective R2* (D) were reconstructed using hybrid method with single R2* fitting. Proton density fat fraction (B) and water R2* (E) were reconstructed using hybrid method with dual R2* fitting. Proton density fat fraction (C) and effective R2* (F) were reconstructed using complex method. Fat R2* was reconstructed using hybrid method with dual R2* fitting (G); note severe noise throughout image. White squares show ROI locations, colocalized with single-voxel spectroscopy, which yielded reference proton density fat fraction value of 23.9%.
Fig. 1F —50-year-old man with hepatic steatosis.
F, Representative images from MRI are shown. Proton density fat fraction (A) and effective R2* (D) were reconstructed using hybrid method with single R2* fitting. Proton density fat fraction (B) and water R2* (E) were reconstructed using hybrid method with dual R2* fitting. Proton density fat fraction (C) and effective R2* (F) were reconstructed using complex method. Fat R2* was reconstructed using hybrid method with dual R2* fitting (G); note severe noise throughout image. White squares show ROI locations, colocalized with single-voxel spectroscopy, which yielded reference proton density fat fraction value of 23.9%.
Fig. 1G —50-year-old man with hepatic steatosis.
G, Representative images from MRI are shown. Proton density fat fraction (A) and effective R2* (D) were reconstructed using hybrid method with single R2* fitting. Proton density fat fraction (B) and water R2* (E) were reconstructed using hybrid method with dual R2* fitting. Proton density fat fraction (C) and effective R2* (F) were reconstructed using complex method. Fat R2* was reconstructed using hybrid method with dual R2* fitting (G); note severe noise throughout image. White squares show ROI locations, colocalized with single-voxel spectroscopy, which yielded reference proton density fat fraction value of 23.9%.
Fig. 2A —26-year-old woman without hepatic steatosis.
A, Representative images from MRI are shown. Proton density fat fraction (A) and effective R2* (D) were reconstructed using hybrid method with single R2* fitting. Proton density fat fraction (B) and water R2* (E) were reconstructed using hybrid method with dual R2* fitting. Proton density fat fraction (C) and effective R2* (F) were reconstructed using complex method. Fat R2* was reconstructed using hybrid method with dual R2* fitting (G); note severe noise throughout image. White squares show ROI locations, colocalized with single-voxel spectroscopy, which yielded reference proton density fat fraction value of 2.1%.
Fig. 2B —26-year-old woman without hepatic steatosis.
B, Representative images from MRI are shown. Proton density fat fraction (A) and effective R2* (D) were reconstructed using hybrid method with single R2* fitting. Proton density fat fraction (B) and water R2* (E) were reconstructed using hybrid method with dual R2* fitting. Proton density fat fraction (C) and effective R2* (F) were reconstructed using complex method. Fat R2* was reconstructed using hybrid method with dual R2* fitting (G); note severe noise throughout image. White squares show ROI locations, colocalized with single-voxel spectroscopy, which yielded reference proton density fat fraction value of 2.1%.
Fig. 2C —26-year-old woman without hepatic steatosis.
C, Representative images from MRI are shown. Proton density fat fraction (A) and effective R2* (D) were reconstructed using hybrid method with single R2* fitting. Proton density fat fraction (B) and water R2* (E) were reconstructed using hybrid method with dual R2* fitting. Proton density fat fraction (C) and effective R2* (F) were reconstructed using complex method. Fat R2* was reconstructed using hybrid method with dual R2* fitting (G); note severe noise throughout image. White squares show ROI locations, colocalized with single-voxel spectroscopy, which yielded reference proton density fat fraction value of 2.1%.
Fig. 2D —26-year-old woman without hepatic steatosis.
D, Representative images from MRI are shown. Proton density fat fraction (A) and effective R2* (D) were reconstructed using hybrid method with single R2* fitting. Proton density fat fraction (B) and water R2* (E) were reconstructed using hybrid method with dual R2* fitting. Proton density fat fraction (C) and effective R2* (F) were reconstructed using complex method. Fat R2* was reconstructed using hybrid method with dual R2* fitting (G); note severe noise throughout image. White squares show ROI locations, colocalized with single-voxel spectroscopy, which yielded reference proton density fat fraction value of 2.1%.
Fig. 2E —26-year-old woman without hepatic steatosis.
E, Representative images from MRI are shown. Proton density fat fraction (A) and effective R2* (D) were reconstructed using hybrid method with single R2* fitting. Proton density fat fraction (B) and water R2* (E) were reconstructed using hybrid method with dual R2* fitting. Proton density fat fraction (C) and effective R2* (F) were reconstructed using complex method. Fat R2* was reconstructed using hybrid method with dual R2* fitting (G); note severe noise throughout image. White squares show ROI locations, colocalized with single-voxel spectroscopy, which yielded reference proton density fat fraction value of 2.1%.
Fig. 2F —26-year-old woman without hepatic steatosis.
F, Representative images from MRI are shown. Proton density fat fraction (A) and effective R2* (D) were reconstructed using hybrid method with single R2* fitting. Proton density fat fraction (B) and water R2* (E) were reconstructed using hybrid method with dual R2* fitting. Proton density fat fraction (C) and effective R2* (F) were reconstructed using complex method. Fat R2* was reconstructed using hybrid method with dual R2* fitting (G); note severe noise throughout image. White squares show ROI locations, colocalized with single-voxel spectroscopy, which yielded reference proton density fat fraction value of 2.1%.
Fig. 2G —26-year-old woman without hepatic steatosis.
G, Representative images from MRI are shown. Proton density fat fraction (A) and effective R2* (D) were reconstructed using hybrid method with single R2* fitting. Proton density fat fraction (B) and water R2* (E) were reconstructed using hybrid method with dual R2* fitting. Proton density fat fraction (C) and effective R2* (F) were reconstructed using complex method. Fat R2* was reconstructed using hybrid method with dual R2* fitting (G); note severe noise throughout image. White squares show ROI locations, colocalized with single-voxel spectroscopy, which yielded reference proton density fat fraction value of 2.1%.
Results of the statistical analyses comparing MRI-derived measures of proton density fat fraction and MRS measures are shown in Table 1. Note that ICCs were high for all three comparisons (0.99 for all three measures).
TABLE 1: Performance of Three MRI-Based Measures of Proton Density Fat Fraction Compared With Reference Standard MR Spectroscopy
Statistical AnalysisHybrid Method (Single R2*)Hybrid Method (Dual R2*)Complex Method
Intraclass correlation coefficient (95% CI)0.99 (0.99–1.00)0.99 (0.98–1.00)0.99 (0.99–1.00)
Linear regression   
 Slope1.01 ± 0.03 (0.96–1.07)1.01 ± 0.03 (0.95–1.08)0.98 ± 0.03 (0.93–1.03)
 Intercept (%)−1.12 ± 0.34 (−1.80 to −0.43)0.49 ± 0.39 (−0.29 to 1.30)−0.10 ± 0.32 (−0.80 to 0.50)
Correlation coefficient (R2)0.970.960.97
Bland-Altman bias (%)−1.0 ± 1.4 (−3.7 to 1.7)0.6 ± 1.6 (−2.4 to 3.7)−0.4 ± 1.3 (−2.9 to 2.2)

Note—Except where noted otherwise, data are mean ± SD (95% CI).

Linear regression showed excellent correlation between each imaging measure and the reference spectroscopy, with slopes of 1.01 ± 0.03 (image set A), 1.01 ± 0.03 (image set B), and 0.98 ± 0.03 (image set C), with 95% CIs all containing slope = 1. The intercepts were −1.12% ± 0.34% (image set A), 0.49% ± 0.39% (image set B), and −0.10% ± 0.32% (image set C), respectively. Notably, the intercept using the hybrid MRI method with single R2* modeling was significantly less than 0% at the 95% confidence level, showing a small but statistically significant bias in this measure. Figure 3 shows plots of the MRI-based measurements against MRS from all three methods, with lines for the least-squares regression and unity on each plot.
Fig. 3A —Results of linear regression analysis for three methods.
A, Scatterplots show results for hybrid method with single R2* fitting (A), hybrid method with dual R2* fitting (B), and complex method (C). Solid lines delineate linear regression lines, and dashed lines show unity or perfect correlation for reference. MRS = MR spectroscopy.
Fig. 3B —Results of linear regression analysis for three methods.
B, Scatterplots show results for hybrid method with single R2* fitting (A), hybrid method with dual R2* fitting (B), and complex method (C). Solid lines delineate linear regression lines, and dashed lines show unity or perfect correlation for reference. MRS = MR spectroscopy.
Fig. 3C —Results of linear regression analysis for three methods.
C, Scatterplots show results for hybrid method with single R2* fitting (A), hybrid method with dual R2* fitting (B), and complex method (C). Solid lines delineate linear regression lines, and dashed lines show unity or perfect correlation for reference. MRS = MR spectroscopy.
Bland-Altman analysis also showed excellent agreement between MRI- and MRS-based proton density fat fraction measures (Fig. 4 and Table 1). The largest systematic bias of −1.0% was again shown by the hybrid MRI method (image set A) with single R2* modeling. Random error was also small for all three methods, with SDs of 1.4%, 1.6%, and 1.3%, respectively. Bland-Altman plots comparing each of the three MRI-based methods against MRS are shown in Figure 4.
Fig. 4A —Results of Bland-Altman analysis for three methods.
A, Scatterplots show results for hybrid method with single R2* fitting (A), hybrid method with dual R2* fitting (B), and complex method (C). Solid lines delineate 95% CIs, and dashed lines show central bias. MRS = MR spectroscopy.
Fig. 4B —Results of Bland-Altman analysis for three methods.
B, Scatterplots show results for hybrid method with single R2* fitting (A), hybrid method with dual R2* fitting (B), and complex method (C). Solid lines delineate 95% CIs, and dashed lines show central bias. MRS = MR spectroscopy.
Fig. 4C —Results of Bland-Altman analysis for three methods.
C, Scatterplots show results for hybrid method with single R2* fitting (A), hybrid method with dual R2* fitting (B), and complex method (C). Solid lines delineate 95% CIs, and dashed lines show central bias. MRS = MR spectroscopy.
By MRS, R2 values of water ranged from 20.8 to 50.2 s−1. Using the three image reconstruction methods, R2* values (either single effective R2* or water R2*) ranged from 21.8 to 75.9 s−1, a range of values similar to those published by other groups [42, 43]. Although no reference standard for R2* was available, there was strong agreement among the R2* values measured within the VOIs using the three image reconstruction methods. For pairings, ICC values were 0.997 (image sets A vs B), 0.994 (image sets B vs C), and 0.993 (image sets A vs C). The ICC for R2* values from all three methods was 0.996.
Linear regression analysis between R2* and proton density fat fraction values showed a moderate correlation (R2 = 0.36–0.42; p < 0.001 for all three MRI methods). A scatter-plot of the effective R2* versus proton density fat fraction data for image set A is shown in Figure 5. Regression line slopes were 0.88–0.95 and intercepts were 37.7–38.3% across the three image reconstruction methods.
Fig. 5 —Scatterplot showing results of linear regression analysis of effective R2* versus proton density fat fraction for hybrid image reconstruction method with single R2* fitting. R2* and proton density fat fraction values were significantly correlated (r2 = 0.42; p < 0.001). Regression line slope was 0.95, and intercept was 38.3%.

Discussion

This study found excellent agreement among three methods of measuring hepatic proton density fat fraction by MRI and reference standard R2-corrected single-voxel spectroscopy. An extensive body of literature has shown the accuracy of using T1-insensitive chemical shift methods that incorporate R2* correction, correction for the complex multipeak lipid spectrum, and eddy current compensation for measuring hepatic proton density fat fraction [2, 3, 2127, 29, 39]. T1 weighting is minimized during the acquisition step by using low flip angles relative to the TR. Two of the methods evaluated in this study were hybrid magnitude-complex techniques, which are inherently insensitive to eddy currents, incorporate a published fat spectral model, and model R2*-based signal decay [31, 37]. The third was a complex-based method that incorporated field map correction and eddy current compensation by variable projection at a low-resolution level [31].
Two of these three image reconstruction techniques were implemented to run inline on the MRI systems at the time of our study, with no user interaction beyond prescription of the scan volume; the third (complex method) was implemented shortly after our original data collection concluded, and results were reconstructed using same raw data. All image reconstructions processed in a clinically feasible amount of time. Thus, these techniques are feasible for use in routine clinical imaging.
Our results show small errors in agreement between the MRI-based methods and reference standard MRS, characterized by small deviations of the slopes of the regression lines from 1.00 and the intercepts from 0.0%. In principle, regression analyses are prone to overstating the strength of correlations in the presence of outlying values, but the data in this study did not contain any such outliers. Also, the ICC and Bland-Altman analyses, which are more robust measures of agreement, show excellent results. In particular, the Bland-Altman analyses show that the absolute amounts of systematic bias (≤ 1.0%) and random error (SD ≤ 1.4%) are small and are unlikely to be important in the clinical setting.
Notably, we found a significant correlation between in vivo R2* and proton density fat fraction values in our study for all three image reconstruction techniques, with similar correlation coefficients and regression lines among the three techniques. This is in contrast to a recent study by Kühn et al. [44], which found no association between R2* values and grades of hepatic steatosis. During the testing and development of the MRI techniques used in the current work, we found no statistically significant correlation between measured R2* and proton density fat fraction values in phantoms, in which the ground truth R2* and proton density fat fraction values were known to be independent. The fact that water R2* and proton density fat fraction were correlated when water and fat R2* were calculated separately further suggests that this correlation is not caused by weighting of the effective R2* toward fat R2* as proton density fat fraction increases, when a single effective R2* is calculated.
Our data therefore suggest that the correlation between R2* and proton density fat fraction may represent a true biologic relationship in vivo. It has been shown that iron can accumulate within the liver with progression of chronic liver disease, though the relationship between progressed liver disease (with greater amounts of inflammation and fibrosis) and liver fat and iron deposition is complex [45, 46]. The difference between our finding and that of Kühn et al. [44] could be explained by differences in patient population. Our study population was largely composed of patients with extrahepatic primary tumors (52%) and chronic liver disease (mainly hepatitis C and nonalcoholic steatohepatitis, 33%), whereas the prevalence and type of chronic liver disease was not reported in the article by Kühn et al. [44]. Nonetheless, additional in vivo investigations are needed to further define the relationship between R2* and proton density fat fraction.
A large proportion of the in vivo validations of MRI-based techniques for quantification of hepatic steatosis have been performed at a small number of sites, on a limited number of MRI platforms, and, in some cases, in a research rather than a clinical setting [2, 17, 21, 25, 26]. To be widely usable, these methods are evolving to become robust to MRI vendor, MRI system, field strength, site, and operator [30]. The image acquisition in our study was performed by clinical MRI technologists, with fully automated image postprocessing (for image sets A and B) yielding proton density fat fraction maps without the need for user interaction with the raw or image data. In addition, this study was performed on widely available commercial MRI systems in a busy clinical practice and shows the suitability of these techniques in a clinical setting. Simple workflow and reliable functionality of the technique are important for such methods to gain wide acceptance in fast-paced clinical settings, and such implementations have been shown by other groups [2, 21, 22, 24, 25]. In theory, further automation could be achieved by combining these techniques with liver sampling or segmentation methods, further simplifying workflow and providing user-independent measures of proton density fat fraction [4751].
For MRI-based measures of proton density fat fraction to become widely accepted biomarkers in clinical medicine and multicenter trials, their accuracy and robustness must be established across a variety of platforms. For this reason, we designed our study and portions of our statistical analysis in a manner similar to previous validations, which were performed using MRI systems from a different manufacturer [2, 26]. Mashhood et al. [30] have also shown the robustness of methods across commercial MRI platforms, albeit with variations on only a single image reconstruction method. Additional trials are needed to establish the reproducibility of proton density fat fraction measures obtained with variations in scanner hardware, pulse sequence parameters, and reconstruction techniques.
This study has several limitations. First, we did not directly compare these techniques with methods developed using other vendor platforms. We did not directly evaluate the contribution of individual components of the image reconstruction chain on the accuracy of the results, because this has been performed extensively in previous studies [2, 2427]. In addition, we did not perform direct parameter optimization on our pulse sequence, but rather used published data to inform our choices of TE and various other parameters [24, 34, 38, 52]. An additional limitation is that we did not explore the effect of low SNR scenarios on the image reconstruction. Our technologists were instructed to place the spectroscopy voxels in the posterior portion of the liver, close to the spine coil array, where SNR is relatively high and respiratory motion less pronounced. Prior work by Liu et al. [17] has shown that low SNR can cause substantial bias in the proton density fat fraction measurement, and this is a potential area for future technical development with our method.
However, this work also has several notable strengths. First, the MRI technologists participating in this study received no special training beyond instruction on placement of the spectroscopy voxels. Image reconstructions were accomplished rapidly on commercial MR reconstruction engines with no modification beyond the installation of the pulse sequence and reconstruction program. Data acquisition was performed on wide-bore MRI systems, and the results were accurate despite the technical challenges inherent in such systems, such as phase errors related to eddy currents and field inhomogeneity. The methods thus performed well even in the fast-paced clinical setting, with equipment that accepts certain technical limitations to improve patient comfort.
In conclusion, accurate quantification of hepatic steatosis can be performed on widely available commercial wide-bore MRI systems in a busy clinical practice. These techniques can be used without advanced postprocessing or user interaction. The validation of proton density fat fraction quantification methods in this and other works represents an important step toward wide availability and acceptance of these MRI-based measures of hepatic steatosis as an accurate biomarker. Further trials will focus on directly establishing agreement between fat fraction measures obtained using different equipment and reconstruction algorithms. In addition, the reproducibility of these methods should be confirmed through validation of the techniques at other centers.

References

1.
Noureddin M, Lam J, Peterson MR, et al. Utility of magnetic resonance imaging versus histology for quantifying changes in liver fat in nonalcoholic fatty liver disease trials. Hepatology 2013; 58:1930–1940
2.
Meisamy S, Hines CD, Hamilton G, et al. Quantification of hepatic steatosis with T1-independent, T2-corrected MR imaging with spectral modeling of fat: blinded comparison with MR spectroscopy. Radiology 2011; 258:767–775
3.
Reeder SB, Cruite I, Hamilton G, Sirlin CB. Quantitative assessment of liver fat with magnetic resonance imaging and spectroscopy. J Magn Reson Imaging 2011; 34:729–749
4.
Clark JM, Diehl AM. Defining nonalcoholic fatty liver disease: implications for epidemiologic studies. Gastroenterology 2003; 124:248–250
5.
Adams LA, Lymp JF, St Sauver J, et al. The natural history of nonalcoholic fatty liver disease: a population-based cohort study. Gastroenterology 2005; 129:113–121
6.
Matteoni CA, Younossi ZM, Gramlich T, Boparai N, Liu YC, McCullough AJ. Nonalcoholic fatty liver disease: a spectrum of clinical and pathological severity. Gastroenterology 1999; 116:1413–1419
7.
Targher G, Bertolini L, Poli F, et al. Nonalcoholic fatty liver disease and risk of future cardiovascular events among type 2 diabetic patients. Diabetes 2005; 54:3541–3546
8.
Rubinstein E, Lavine JE, Schwimmer JB. Hepatic, cardiovascular, and endocrine outcomes of the histological subphenotypes of nonalcoholic fatty liver disease. Semin Liver Dis 2008; 28:380–385
9.
Idilman IS, Aniktar H, Idilman R, et al. Hepatic steatosis: quantification by proton density fat fraction with MR imaging versus liver biopsy. Radiology 2013; 267:767–775
10.
Tang A, Tan J, Sun M, et al. Nonalcoholic fatty liver disease: MR imaging of liver proton density fat fraction to assess hepatic steatosis. Radiology 2013; 267:422–431
11.
Stock W, Douer D, DeAngelo DJ, et al. Prevention and management of asparaginase/pegasparaginase-associated toxicities in adults and older adolescents: recommendations of an expert panel. Leuk Lymphoma 2011; 52:2237–2253
12.
Sahoo S, Hart J. Histopathological features of L-asparaginase-induced liver disease. Semin Liver Dis 2003; 23:295–299
13.
Dixon WT. Simple proton spectroscopic imaging. Radiology 1984; 153:189–194
14.
Glover GH, Schneider E. Three-point Dixon technique for true water/fat decomposition with B0 inhomogeneity correction. Magn Reson Med 1991; 18:371–383
15.
Ma J. Breath-hold water and fat imaging using a dual-echo two-point Dixon technique with an efficient and robust phase-correction algorithm. Magn Reson Med 2004; 52:415–419
16.
Hussain HK, Chenevert TL, Londy FJ, et al. Hepatic fat fraction: MR imaging for quantitative measurement and display—early experience. Radiology 2005; 237:1048–1055
17.
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; 58:354–364
18.
Reeder SB, McKenzie CA, Pineda AR, et al. Water-fat separation with IDEAL gradient-echo imaging. J Magn Reson Imaging 2007; 25:644–652
19.
Schuchmann S, Weigel C, Albrecht L, et al. Noninvasive quantification of hepatic fat fraction by fast 1.0, 1.5 and 3.0 T MR imaging. Eur J Radiol 2007; 62:416–422
20.
Kim H, Taksali SE, Dufour S, et al. Comparative MR study of hepatic fat quantification using single-voxel proton spectroscopy, two-point dixon and three-point IDEAL. Magn Reson Med 2008; 59:521–527
21.
Yu H, Shimakawa A, McKenzie CA, Brodsky E, Brittain JH, Reeder SB. Multiecho water-fat separation and simultaneous R2* estimation with multifrequency fat spectrum modeling. Magn Reson Med 2008; 60:1122–1134
22.
Yokoo T, Bydder M, Hamilton G, et al. Nonalcoholic fatty liver disease: diagnostic and fat-grading accuracy of low-flip-angle multiecho gradient-recalled-echo MR imaging at 1.5 T. Radiology 2009; 251:67–76
23.
Börnert P, Koken P, Eggers H. Spiral water-fat imaging with integrated off-resonance correction on a clinical scanner. J Magn Reson Imaging 2010; 32:1262–1267
24.
Hines CD, Frydrychowicz A, Hamilton G, et al. T1 independent, T2* corrected chemical shift based fat-water separation with multi-peak fat spectral modeling is an accurate and precise measure of hepatic steatosis. J Magn Reson Imaging 2011; 33:873–881
25.
Yokoo T, Shiehmorteza M, Hamilton G, et al. Estimation of hepatic proton-density fat fraction by using MR imaging at 3.0 T. Radiology 2011; 258:749–759
26.
Yu H, Shimakawa A, Hines CD, et al. Combination of complex-based and magnitude-based multiecho water-fat separation for accurate quantification of fat-fraction. Magn Reson Med 2011; 66:199–206
27.
Hernando D, Hines CD, Yu H, Reeder SB. Addressing phase errors in fat-water imaging using a mixed magnitude/complex fitting method. Magn Reson Med 2012; 67:638–644
28.
Reeder SB, Hu HH, Sirlin CB. Proton density fat-fraction: a standardized MR-based biomarker of tissue fat concentration. J Magn Reson Imaging 2012; 36:1011–1014
29.
Sharma SD, Hu HH, Nayak KS. Accelerated water-fat imaging using restricted subspace field map estimation and compressed sensing. Magn Reson Med 2012; 67:650–659
30.
Mashhood A, Railkar R, Yokoo T, et al. Reproducibility of hepatic fat fraction measurement by magnetic resonance imaging. J Magn Reson Imaging 2013; 37:1359–1370
31.
Zhong X, Nickel MD, Kannengiesser SA, Dale BM, Kiefer B, Bashir MR. Liver fat quantification using a multi-step adaptive fitting approach with multi-echo GRE imaging. Magn Reson Med 2013 [Epub ahead of print]
32.
Breuer FA, Blaimer M, Mueller MF, et al. Controlled aliasing in volumetric parallel imaging (2D CAIPIRINHA). Magn Reson Med 2006; 55:549–556
33.
Levin YS, Yokoo T, Wolfson T, et al. Effect of echo-sampling strategy on the accuracy of out-of-phase and in-phase multiecho gradient-echo MRI hepatic fat fraction estimation. J Magn Reson Imaging 2014; 39:567–575
34.
Johnson BL, Schroeder ME, Wolfson T, et al. Effect of flip angle on the accuracy and repeatability of hepatic proton density fat fraction estimation by complex data-based, T1-independent, T2*-corrected, spectrum-modeled MRI. J Magn Reson Imaging 2014; 39:440–447
35.
Chebrolu VV, Hines CD, Yu H, et al. Independent estimation of T*2 for water and fat for improved accuracy of fat quantification. Magn Reson Med 2010; 63:849–857
36.
Reeder SB, Robson PM, Yu H, et al. Quantification of hepatic steatosis with MRI: the effects of accurate fat spectral modeling. J Magn Reson Imaging 2009; 29:1332–1339
37.
Ren J, Dimitrov I, Sherry AD, Malloy CR. Composition of adipose tissue and marrow fat in humans by 1H NMR at 7 Tesla. J Lipid Res 2008; 49:2055–2062
38.
Bydder M, Yokoo T, Hamilton G, et al. Relaxation effects in the quantification of fat using gradient echo imaging. Magn Reson Imaging 2008; 26:347–359
39.
Yu H, McKenzie CA, Shimakawa A, et al. Multi-echo reconstruction for simultaneous water-fat decomposition and T2* estimation. J Magn Reson Imaging 2007; 26:1153–1161
40.
Hernando D, Kramer JH, Reeder SB. Multipeak fat-corrected complex R2* relaxometry: theory, optimization, and clinical validation. Magn Reson Med 2013; 70:1319–1331
41.
Pineda N, Sharma P, Xu Q, Hu X, Vos M, Martin DR. Measurement of hepatic lipid: high-speed T2-corrected multiecho acquisition at 1H MR spectroscopy—a rapid and accurate technique. Radiology 2009; 252:568–576
42.
Horng DE, Hernando D, Hines CD, Reeder SB. Comparison of R2* correction methods for accurate fat quantification in fatty liver. J Magn Reson Imaging 2013; 37:414–422
43.
Kühn JP, Hernando D, Mensel B, et al. Quantitative chemical shift-encoded MRI is an accurate method to quantify hepatic steatosis. J Magn Reson Imaging 2014; 39:1494–1501
44.
Kühn JP, Hernando D, Munoz del Rio A, et al. Effect of multipeak spectral modeling of fat for liver iron and fat quantification: correlation of biopsy with MR imaging results. Radiology 2012; 265:133–142
45.
Younossi ZM, Gramlich T, Bacon BR, et al. Hepatic iron and nonalcoholic fatty liver disease. Hepatology 1999; 30:847–850
46.
Bonkovsky HL, Banner BF, Rothman AL. Iron and chronic viral hepatitis. Hepatology 1997; 25:759–768
47.
Ruskó L, Bekes G. Liver segmentation for contrast-enhanced MR images using partitioned probabilistic model. Int J Comput Assist Radiol Surg 2011; 6:13–20
48.
Bashir MR, Dale BM, Merkle EM, Boll DT. Automated liver sampling using a gradient dual-echo Dixon-based technique. Magn Reson Med 2012; 67:1469–1477
49.
Chen G, Gu L, Qian L, Xu J. An improved level set for liver segmentation and perfusion analysis in MRIs. IEEE Trans Inf Technol Biomed 2009; 13:94–103
50.
Bashir MR, Zhong X, Dale BM, Gupta RT, Boll DT, Merkle EM. Automated patient-tailored screening of the liver for diffuse steatosis and iron overload using MRI. AJR 2013; 201:583–588
51.
Bashir MR, Merkle EM, Smith AD, Boll DT. Hepatic MR imaging for in vivo differentiation of steatosis, iron deposition and combined storage disorder: single-ratio in/opposed phase analysis vs. dual-ratio Dixon discrimination. Eur J Radiol 2012; 81:e101–e109
52.
Yokoo T, Collins JM, Hanna RF, Bydder M, Middleton MS, Sirlin CB. Effects of intravenous gadolinium administration and flip angle on the assessment of liver fat signal fraction with opposed-phase and in-phase imaging. J Magn Reson Imaging 2008; 28:246–251

FOR YOUR INFORMATION

The comprehensive book based on the ARRS 2014 annual meeting categorical course on The Radiology M and M Meeting: Misinterpretations, Misses, and Mimics is now available! For more information or to purchase a copy, see www.arrs.org.

Information & Authors

Information

Published In

American Journal of Roentgenology
Pages: 297 - 306
PubMed: 25615751

History

Submitted: January 6, 2014
Accepted: June 27, 2014

Keywords

  1. fat fraction
  2. hepatic steatosis
  3. proton density fat fraction

Authors

Affiliations

Mustafa R. Bashir
Department of Radiology, Duke University Medical Center, DUMC Box 3808, Durham, NC 27710.
Xiaodong Zhong
MR R&D Collaborations, Siemens Healthcare, Atlanta, GA.
Marcel D. Nickel
Siemens AG Healthcare Sector, Erlangen, Germany.
Ghaneh Fananapazir
Department of Radiology, Duke University Medical Center, DUMC Box 3808, Durham, NC 27710.
Stephan A. R. Kannengiesser
Siemens AG Healthcare Sector, Erlangen, Germany.
Berthold Kiefer
Siemens AG Healthcare Sector, Erlangen, Germany.
Brian M. Dale
MR R&D Collaborations, Siemens Healthcare, Morrisville, NC.

Notes

Address correspondence to M. R. Bashir ([email protected]).
M. R. Bashir is a consultant to Siemens Healthcare and receives research support from Siemens Healthcare.
X. Zhong, M. D. Nickel, S. A. R. Kannengiesser, B. Kiefer, and B. M. Dale are employees of Siemens Healthcare.

Metrics & Citations

Metrics

Citations

Export Citations

To download the citation to this article, select your reference manager software.

Articles citing this article

View Options

View options

PDF

View PDF

PDF Download

Download PDF

Media

Figures

Other

Tables

Share

Share

Copy the content Link

Share on social media