To compare noninvasive MRI and magnetic resonance spectroscopy (MRS) methods with liver biopsy to quantify liver fat content.
To compare noninvasive MRI and magnetic resonance spectroscopy (MRS) methods with liver biopsy to quantify liver fat content.
Quantification of liver fat was compared by liver biopsy, proton MRS, and MRI using in-phase/out-of-phase (IP/OP) and plus/minus fat saturation (±FS) techniques. The reproducibility of each MR measure was also determined. An additional group of overweight patients with steatosis underwent hepatic MRI and MRS before and after a six-month weight-loss program.
A close correlation was demonstrated between histological assessment of steatosis and measurement of intrahepatocellular lipid (IHCL) by MRS (rs = 0.928, P < 0.0001) and MRI (IP/OP rs = 0.942, P < 0.0001; FS rs = 0.935, P < 0.0001). Following weight reduction, four of five patients with >5% weight loss had a decrease in IHCL of ≥50%.
These findings suggest that standard MRI protocols provide a rapid, safe, and quantitative assessment of hepatic steatosis. This is important because MRS is not available on all clinical MRI systems. This will enable noninvasive monitoring of the effects of interventions such as weight loss or pharmacotherapy in patients with fatty liver diseases. J. Magn. Reson. Imaging 2008;28:937–945. © 2008 Wiley-Liss, Inc.
OBESITY-RELATED STEATOSIS is a cofactor in the development or progression of liver injury in a number of chronic liver diseases, including nonalcoholic fatty liver disease (NAFLD) (1) hepatitis C virus (HCV) (2), and hemochromatosis (3). In addition, obesity and steatosis are factors that impair the response to antiviral therapy in patients with chronic HCV (4–6). As a consequence, weight loss is now considered to be an important component of the management of overweight patients with chronic liver disease. In clinical practice, lifestyle and dietary interventions usually produce a modest weight loss (∼5% body weight) and do not result in a “normal” body mass index (BMI). It remains unclear whether this relatively small weight loss is sufficient to mobilize the intrahepatic lipids in patients with chronic liver disease.
It was previously shown that in subjects with chronic HCV, a mean weight loss of 8.1% led to a decrease in steatosis, as assessed in paired liver biopsy specimens, with a median reduction of one grade on a four-point histological scale (7). In addition to concerns regarding sampling error, liver biopsy is invasive and cannot be performed repeatedly to monitor hepatic steatosis following treatment. Noninvasive assessment of steatosis is required to investigate fundamental questions such as stability and mobilization of liver fat content (8).
A noninvasive, quantitative assessment of steatosis is available using proton magnetic resonance spectroscopy (1H MRS). Results of MRS have shown to correlate closely with biochemical and histological assessments of liver triglyceride content (9, 10), and to demonstrate a decrease in liver fat content following rosiglitazone therapy (11) or weight reduction (12, 13). Other imaging modalities, such as ultrasonography and computed tomography (CT), provide limited quantitative assessment (14, 15). However, radiation exposure limits repeated measurements. Although MRI can depict liver fat by determining intensity differences between in-phase (IP) and out-of-phase (OP) images or images acquired with/without fat saturation (±FS), it is not routinely used in the clinical setting for quantitative measurements.
The aim of the current study was to examine the ability of standard fat-edited MRI sequences to monitor liver fat. Intrahepatocellular lipid (IHCL) using MRI techniques was compared with assessment by histology and 1H MRS.
To investigate the correlation between steatosis determined by histology with MRI and MRS, 12 patients (Table 1) who had a percutaneous liver biopsy arranged as part of their clinical management were invited to participate in this study. Sequential patients with no (N = 1), mild (N = 4), moderate (N = 4), or severe (N = 3) steatosis were recruited and underwent hepatic MRI and MRS within six weeks of their liver biopsy.
|Liver disease||HCV||HCV, HBV||HCV||HCV||HCV||HCV||HCV||HCV||HCV||NAFLD||NAFLD||NAFLD|
|Stage of fibrosis||0||1||1||1||1||4||1||4||4||1||0||2|
|Grade of hepatic iron||0||0||0||0||1||1||0||0||0||0||0||0|
|% Steatotic hepatocytes||3||7||10||15||20||30||35||50||55||90||95||95|
|IHCL by 1H MRS (AU)||1.4||2.5||5.4||9.5||8.3||7.2||14.2||9.8||14.0||31.1||21.6||25.8|
To investigate mobilization of liver fat during weight loss, a group of 10 overweight patients (Table 2) with chronic liver disease who were referred to the dietician for weight reduction underwent hepatic MRI and MRS before and after a six-month weight-loss program.
|Liver disease||HCb||HBV||HCV||HCV||HCV||HCb||NAFLD||NAFLDc||AICAH, NAFLD||HCV|
|Stage of fibrosis||0||2||1||4||2||0||ND||1||4||1|
|Grade of hepatic iron||0||0||0||0||1||4||ND||2||0||0|
|Grade of steatosis||2||2||2||2||2||ND||ND||1||2||1|
|% Weight loss||1.8||3.1||3.5||3.5||4.6||6.1||5.9||10.5||14.7||22.8|
|% Waist loss||1.1||1.9||8.6||3.4||4.0||9.1||10.3||9.1||9.9||24.4|
|IHCL by 1H MRS (AU)|
|% Reduction in liver fat||6.1||24.4||78.9||43.4||39.9||70.9||42.2||69.9||59.9||90|
|Subcutaneous fat (cm2)|
|% Reduction in subcutaneous fat||−0.8||1.5||10.9||7.2||−0.05||17.4||18.8||20.2||25.4||35.6|
|Visceral fat (cm2)|
|% Reduction in visceral fat||−3.4||19.1||−12.4||−4.8||21.4||17.8||15.4||34.6||23.6||46.7|
Informed consent was obtained from each patient and the study protocol was approved by the Princess Alexandra Hospital, Wesley Hospital, and University of Queensland research ethics committees.
Liver fat was measured using three standard protocols on a Siemens Sonata 1.5T MRI system (Erlangen, Germany) with the subjects in a supine position using standard array coils. Localizer images were used for planning of fast imaging with steady-state precession (FISP) (16), IP/OP (Dixon) images (17) acquired simultaneously, and T2-weighted images ±FS acquired with T2 half-Fourier acquisition single-shot turbo spin echo (HASTE) with and without FS. Sequence parameters are presented in Table 3.
|Image||TR (msec)||TE (msec)||Slice thickness (mm)||Slices||Field of view (mm)||Averages||Handling of respiratory motion|
|FISP||3.67||1.88||8||14||400 × 400||1||Single breathhold|
|IP/OP||70||2.38 and 5.04||8||20||400 × 400||1||Three breathholds|
|±FS||Respiratory-dependent||121||8||20||400 × 400||2||Free-breathing using navigator respiratory gating|
Images were analyzed using standard Siemens software by a single operator (G.C.) on completion of each study. Similar slices, which contained liver and spleen for positioning of the region of interest (ROI), were selected from each data set as illustrated by the round ROIs in Fig. 1. A region of homogeneous tissue avoiding obvious vessels was selected, in a similar position in the anterior–posterior direction and a similar distance from the center of the image in the left–right direction. This was done to reduce intensity variations due to varying distances between the ROI and the receive coils. Methods used to calculate liver fat were compared as described in Table 4.
|1. Liver signal-to-noise||(mean signal of liver ROI)/(mean signal of noise ROI)|
|2. Liver signal to spleen signal||(mean signal of liver ROI)/(mean signal of spleen ROI)|
|3. Difference from IP/OP images||(liver IP − liver OP)/(liver IP*100)|
|4. Difference from IP/OP images (spleen-corrected)||[(liver IP/spleen IP) − (liver OP/spleen OP)]/(liver IP/spleen IP*100)|
|5. Difference from FS images||(liver no FS − liver FS)/(liver no FS*100)|
|6. Difference from FS images (spleen-corrected)||[(liver no FS/spleen no FS) − (liver FS/spleen FS)]/(liver no FS/spleen no FS*100)|
Single-voxel spectra were measured using the point-resolved spectroscopy (PRESS) (18) technique with the following parameters: TR = 3 seconds, TE = 30 msec, data points = 2048, bandwidth = 2000 Hz, voxel size = 20 × 20 × 20 mm, no water suppression, and four averages during a single breath-hold on expiration. Using the FISP images, four voxels were positioned within the liver (square ROIs, Fig. 1), avoiding obvious vessels and the edge of the liver. Automatic setup was performed during free breathing with no operator intervention.
All spectra were processed using the standard Siemens software by a single operator (G.C.) on a single day following completion of each study. Automatic processing included phase correction based on the water and 200 msec exponential filter. Manual processing included optimization of the phase and a linear baseline correction if required.
Peak quantification was performed using the Siemens curve-fitting package with the following parameters: water: amplitude 1–100,000, width 1–50 Hz, peak shape Lorentzian, position 4–5.5 ppm; lipid peaks: amplitude 1–100,000, width 1–50 Hz, peak shape Gaussian, position 2–3 ppm (CH2-C=), 1.2–1.8 ppm (-CH2-), 0.7–1.1 ppm (-CH3).
The average liver fat derived from the four voxels was expressed as a percentage using (CH2 + CH3) / (H2O + CH2 + CH3) × 100.
Reproducibility was assessed by six repeated measures on a single subject and by paired measures on six individuals. The six repetitions on the single subject were completed on a single morning (between 9:25 and 12:15). Four ROIs were positioned in the phase-edited and FS-edited images in positions similar to the position of the spectroscopy voxels (square ROIs, Fig. 1). Between each study the subject was repositioned. The data obtained for the paired reproducibility assessments were analyzed by means of the standard techniques.
Breathhold FISP images (Table 3) were centered on the L4 vertebra. If the patient's width was greater than the 400-mm field-of-view (FOV) limit, two offset images were acquired to ensure complete coverage of the subject. The four slices that were best aligned with the L4 vertebra (19, 20) were analyzed by a single operator (J.B.) using the polygon ROI in DicomWorks to define subcutaneous and intraabdominal fat areas as described previously (21). The average results for the four slices were reported.
Liver biopsies were fixed in formalin and embedded in paraffin. Liver biopsy sections were analyzed by an experienced hepatopathologist (A.C.) who was blinded to the laboratory parameters and clinical data. Steatosis was graded as follows: 0 (<5% hepatocytes affected); 1 (mild, 5–29% of hepatocytes affected); 2 (moderate, 30–70% of hepatocytes affected); or 3 (severe, >70% of hepatocytes affected). In addition, % steatotic hepatocytes was determined and fibrosis was staged according to the method of Scheuer (22).
The weight-reduction program was provided as part of standard clinical care for overweight patients with steatosis and chronic liver disease. The intervention involved an intensive period of weekly review by a dietician for 12 weeks, followed by a weight maintenance program involving monthly review for 3 months. The subjects were counseled on achieving an energy-restricted diet and exercise regimen to promote the goal of 0.5 kg weight loss per week.
Blood samples for measurement of glucose and insulin were obtained after an overnight fast before and after the weight loss intervention or at liver biopsy. Routine biochemical tests were performed using a Hitachi 747-100 analyzer (Roche). Circulating insulin was determined using the Tosoh AIA600 analyzer two-site immunoenzymometric assay (Tosoh Medics, San Francisco, CA, USA) with a coefficient of variation (CV) of 4% to 5%.
Statistical analysis was performed using SPSS for Windows version 14.0. Data were summarized as the mean ± standard deviation (SD). The correlations between histology, 1H MRS, and MRI were assessed using Spearman's nonparametric correlation coefficient (rs). Statistical significance was taken at the 5% level.
Demographic and clinical information for the 12 patients who underwent a liver biopsy within six weeks of MRI are given in detail in Table 1. All liver biopsies were intercostal transcutaneous biopsies of the right lobe, and were performed within 25.8 ± 3.2 (11–42) days of MRI. The mean length of the liver biopsies was 23.2 ± 3.04 (13–41) mm. The percentage of hepatocytes that were steatotic ranged from 3% to 95% and five patients had ≤20% steatotic hepatocytes. Three patients had cirrhosis documented on liver biopsy, and all were clinically well compensated. Two patients from group 1 (patients G and H) also enrolled for the weight-reduction program.
Demographic and clinical data for the 10 patients who participated in a weight-reduction program are given in detail in Table 2. The mean BMI was 33.9 kg/m2 (range = 27.2–42.1 kg/m2) and the mean waist circumference was 115 cm (range = 105–137.5 cm). Two patients had type 2 diabetes according to defined criteria (23). None of the patients with viral hepatitis had received antiviral therapy. Two patients (patient H from group 1, and patient S) had cirrhosis documented on a previous liver biopsy and were clinically well compensated. Biopsy data for the weight-loss group was not used for correlation with MR data.
Following the six-month lifestyle intervention, the mean weight loss was 7.8 kg (range = 1.9–25.6 kg) and mean waist loss was 9.9 cm (range = 1.1–33.5 cm).
Figures 2 and 3 illustrate representative images and spectra from subjects with MRS-calculated liver fat of 1% and 26%, respectively. Figure 4 plots IHCL (assessed by MRS) with % steatosis (determined by histology). A close correlation was measured (rs = 0.928, P < 0.0001). MRI determination of IHCL, by methods 3–6 (Table 4), also demonstrated close correlation with the histological assessment of steatosis (Table 5).
|3. IP/OP images||0.942||<0.0001|
|4. IP/OP images (spleen-corrected)||0.948||<0.0001|
|5. FS images||0.935||<0.0001|
|6. FS images (spleen- corrected)||0.961||<0.0001|
Data from the biopsy study and two time points for each patient from the weight loss study were used to investigate the correlation between assessment of IHCL by MRS and MRI. The liver intensity was correlated with MRS in the FISP, OP, and non-FS image sequences using the following internal standards: noise intensity (FISP, rs = 0.82; OP, rs = 0.63 and non-FS rs = 0.60) or spleen intensity (FISP, rs = 0.83; OP, rs = 0.78 and non-FS rs = 0.70). Calculation of IHCL content from the intensity difference between corresponding pairs of phase-edited or direct fat-suppression images greatly improved the correlation between the imaging and spectroscopy data (IP/OP rs = 0.959, P < 0.0001; FS rs = 0.910, P < 0.0001, Fig. 5). Inclusion of the spleen intensity as an internal standard in each fat-edited image did not improve the correlation of the imaging data with the spectroscopy data (IP/OP rs = 0.950, P < 0.0001; FS rs = 0.910, P < 0.0001). In patients with cirrhosis, assessment of liver fat content by either imaging technique appeared to be appropriate.
A single subject in the weight-loss study was excluded from the phase-edited data. This subject had elevated liver iron content, which would alter the relaxation times that are critical for phase-edited imaging. None of the other subjects had elevated hepatic iron content at the time of liver imaging.
The reproducibility of the measurement of steatosis by spectroscopy and imaging using the difference calculation methods is summarized in Table 6. Determination of IHCL by MRS generally gave a lower CV in the four regions investigated. The left lobe gave the largest CV for all three MR techniques. Minor regional variation in the calculated liver fat content was evident for spectroscopy and ±FS but not for IP/OP imaging. For spectroscopy, the three regions of the right lobe were not significantly different; however, the value for the left lobe was significantly lower than the anterior and central ROIs (P < 0.01) and the posterior ROI (P < 0.001). For ±FS, the anterior and central ROIs of the right lobe were significantly different to the left lobe (P = <0.001 and P < 0.01, respectively). The posterior ROI was significantly different to the anterior ROI (P < 0.01) in the right lobe.
|Right lobe, posterior (%)||Right lobe, anterior (%)||Right lobe, central (%)||Left lobe (%)|
|Spectroscopy||11.2 ± 0.4 (3)||10.7 ± 0.8 (7)||10.8 ± 0.7 (6)||8.6 ± 1.1 (13)|
|Phase-edited||30.7 ± 3.1 (10)||33.3 ± 2.6 (8)||33.7 ± 4.4 (13)||30.3 ± 5.2 (17)|
|Fat saturation–edited||27.6 ± 1.7 (6)||32.3 ± 2.2 (7)||30.4 ± 2.4 (8)||26.5 ± 2.4 (9)|
The calculated liver fat content for the six subjects ranged from 0% to 13% for spectroscopy, and –5% to 30% for IP/OP imaging and ±FS imaging. The linear correlation between the paired data was calculated as follows: spectroscopy (y = –0.11 + 0.96x × r = 0.997), IP/OP imaging (y = –0.66 + 0.97x × r = 0.98) and ±FS imaging (y = –3.63 + 1.23x × r = 0.96).
Following the weight-reduction program, IHCL decreased in all subjects, with marked interindividual differences. Figure 6a and b illustrate the relationship between % reduction in IHCL by 1H MRS and % weight loss and % waist loss, respectively. A relatively small weight loss was associated with a substantial reduction in liver fat content. A decrease in IHCL of ≥50% was seen in four of five patients with a weight loss of >5%, and in five of six patients with a waist loss of >5%.
In this heterogeneous population of patients, there were no statistically significant relationships between IHCL and various anthropometric or metabolic parameters. The correlation between IHCL and BMI (rs = 0.432, P = 0.057) and visceral fat (rs = 0.40, P = 0.08) approached significance.
There were significant correlations between the % weight loss and either the % change in visceral fat (rs = 0.661, P = 0.038) or % change in subcutaneous fat (rs = 0.952, P < 0.001). In this small group of subjects there was a trend between the % change in serum insulin levels and change in visceral fat (rs = 0.60, P = 0.08) but not with the change in subcutaneous fat (rs = 0.183, P = 0.7).
The % change in IHCL was associated with the % change in subcutaneous fat (rs = 0.75, P = 0.013) but not the % change in visceral fat (rs = 0.24, P = 0.51) or the change in serum insulin levels (rs = 0.286, P = 0.46).
Recognition of the role of steatosis in liver injury and impaired response to treatment has led to a need for accurate, noninvasive methods to quantify liver fat content. Liver biopsy remains the reference standard and, in addition to assessing the severity of steatosis, it provides information about the extent of fibrosis and disease activity. However, the procedure is invasive and cannot be performed repeatedly to monitor changes in liver fat content during therapeutic interventions.
To our knowledge, this is the first comparison of the simultaneous assessment of steatosis by liver biopsy, MRS, IP/OP, and ±FS imaging. Quantitation of steatosis using 1H MRS or MRI was closely correlated with histological assessment in patients with chronic liver disease of varying etiology and severity, and serial examinations showed a reduction in hepatic fat content with weight loss. This is consistent with previous studies reporting a correlation between MRS and histomorphometric data (9, 10, 24).
Estimation of IHCL from the intensity difference between corresponding fat-edited images, either phase-edited or fat-suppression-edited, gave excellent correlation with MRS data. This is important because 1H MRS is not universally available on clinical MR systems. Compared to MRI, spectroscopy is more influenced by system optimization and data processing, both of which are influenced by operator experience (25). In contrast, MRI is less sensitive to system optimization, and as a difference method it only requires the placement of a single ROI in a homogeneous region of the liver. The difference method gave superior results compared with using the spleen or noise as an internal standard.
Previous results have also demonstrated a strong correlation between spectroscopy and IP/OP imaging on a Philips system in normal subjects (26). Regression line equations similar to our data suggest that the correlation between IP/OP imaging and liver steatosis is not dependent on the system type. A strong correlation between IP/OP imaging with histology (27) and MRS (28) has also been reported. Qayyum et al (29) investigated a group of subjects with and without cirrhosis by phase-edited and FS-edited imaging using the spleen as an internal standard on a GE system. They reported that ±FS imaging was better correlated with histopathology than in-opposite phase imaging in noncirrhotic subjects (29). In subjects with cirrhosis, only the ±FS imaging was correlated with the histopathology. Variable hepatic iron levels (particularly in cirrhotic patients), which affect the relaxation constants, were discussed as a potential reason for the lower correlation of phase-edited MRI with liver biopsy results (29). This is consistent with the single subject in our study with elevated hepatic iron levels that gave large negative values for phase-edited IHCL. Acquisition of liver fat intensity images using spectral-spatial excitation was correlated with MRS (30). As an alternative to ±FS imaging, paired images with and without water saturation for measurement of liver steatosis were correlated with histology and demonstrated excellent reproducibility in phantom investigations (31). Taken together, these results suggest that fat or water saturation edited liver imaging may have more general application.
An improved correlation between MRS and MRI using the difference method compared to either noise or spleen intensity as an internal standard was not evident. This may result from the image signal intensity not being homogeneous through the liver or spleen, as illustrated in Figs. 2 and 3, due to surface coil reception and proximity of large patients to the transmit body coil. Therefore, changes in the relative position of either the liver or spleen ROI alter the corrected value of liver intensity. These points are consistent with a single voxel, placed in an identical position of the liver, giving superior results. Continued improvement of suppression techniques (32) should improve image homogeneity, and benefit the reproducibility of imaging techniques. The selection of fast T2 HASTE images rather than standard SE sequences may contribute to the regional variation in the ±FS technique.
In this study we did not specifically seek to accurately align the MR ROI or spectroscopy voxel with the biopsy position, although the posterior region of the right lobe was the general region selected for the biopsy and MR measures. This is a limitation of this study, as regional differences in the calculation methods are suggested by the reproducibility data from the four different regions of the liver (square ROIs, Fig. 1). This requires further investigation to determine whether these minor variations reflect a true heterogeneity in the extent of steatosis or results from instrumental sources.
For an MR technique to become a clinical tool for measurement of liver steatosis, standardized methods are required. The correlation between the MR measures and biopsy is dependent upon multiple factors, including the technique, spectroscopy or imaging protocol, calculation method (for example, the inclusion of CH2 or CH2+CH3 integrals for spectroscopy), and relaxation effects. In the current study, approximately 100% steatosis determined by histology is equivalent to a value of ∼30% steatosis calculated by MRS (Fig. 4), ∼90% by IP/OP, or ∼70% by ±FS. Histological grading of the severity of steatosis is determined by calculating the percentage of hepatocytes with visible steatosis rather than the quantity of lipid per hepatocyte. The higher numerical values of the imaging methods, compared to MRS, is a result of the difference method, rather than independent measures of fat and water as in MRS. The combined integrals of the CH2 and CH3 peaks were chosen as a better estimation of total fat proton signal for comparison with the imaging techniques.
For the imaging methods, a crossover point exists where the dominant signal changes from water to fat. At this point, the calculated values start to decrease, introducing a potential source of a false reading. This was addressed recently for IP/OP imaging by Hussain et al (33) by using dual pulse angles, which removed this ambiguity at high fat levels. In addition, that technique minimized the effect of T1 and T2* relaxation upon the estimation of liver steatosis by IP/OP imaging.
In our patient cohort, a relatively small amount of weight loss (≥5% body weight) had the ability to produce a >50% reduction in liver fat. In the subject with HCV genotype 3, in which the virus is considered to have a steatogenic effect, loss of 5% body weight resulted in a 79% reduction in steatosis to “normal levels.” These results confirm earlier data regarding the beneficial effect of weight reduction on steatosis in patients with chronic HCV (7). In addition to loss of liver fat, weight loss led to a significant reduction in visceral and subcutaneous fat.
Previous studies have demonstrated that moderate weight reduction (around 8% body weight) leads to a marked reduction of IHCL content in subjects with type 2 diabetes (12) and women with previous gestational diabetes (13). Petersen et al (12) found that a reduction in IHCL content was associated with a marked improvement in hepatic insulin sensitivity, as reflected by an increase of insulin suppression of glucose production to 93% ± 5% during a hyperinsulinemic clamp compared with 29% ± 22% before the weight loss (P = 0.04). These authors postulated that the IHCL may be a key factor contributing to hepatic insulin resistance and increased gluconeogenesis in patients with type 2 diabetes.
In conclusion, these findings suggest that MRI provides a widely available, safe, and quantitative assessment of hepatic steatosis. This has the potential to be an important clinical tool to noninvasively monitor the effect of interventions such as weight loss or pharmacotherapy in patients with fatty liver diseases.