Quantification of liver fat in mice: comparing dual-echo Dixon imaging, chemical shift imaging, and 1H-MR spectroscopy

Abstract

We evaluated dual-echo Dixon in-phase and out-of-phase (IP-OP), chemical shift imaging (CSI), and ¹H MRS (hydrogen MR spectroscopy) in estimating fat content (FC) in phantoms and in livers of mice. Phantoms were made according to the volume percentage of fat ranging from 0% to 100%. The three MR methods were performed to measure FC in phantoms and in livers of obese leptin-deficient (ob/ob), human BSCL2/seipin gene knockout (SKO), and wild-type (WT) mice. The results were compared with known FC in phantoms and to a reference standard from mice by histological semiautomatic vacuole segmentation (HIS-S) procedure and liver lipid (LL) chemical analysis. In phantoms, CSI underestimated FC from 50% to 100%, to a lesser extent than IP-OP. In vivo, liver FC in ob/ob and SKO mice measured by the three MR methods were all significantly higher than that in WT mice. Liver FC measured by IP-OP were significantly lower than that measured by CSI and MRS, with no significant difference between CSI and MRS. CSI and MRS showed a linear correlation with LL analysis and with each other. IP-OP underestimated FC, whereas CSI and MRS were more accurate for quantifying FC in both phantoms and liver. CSI and MRS have the potential to replace HIS-S and LL analysis in longitudinal studies.

During the past decade, there has been evidence of an epidemic increase in nonalcoholic fatty liver disease (NAFLD), which affects from 10% to 30% of adults (1–3) and 13% of children (4) in the general population. NAFLD has a strong association with type II diabetes mellitus, obesity, hyperlipidemia, and other diseases of the metabolic syndrome (5). The hallmark of NAFLD is fatty infiltration of hepatocytes. Hepatic steatosis also reflects the toxicity of drugs such as amiodarone, tamoxifen, and antiretrovirals (6).

Thus, liver fat quantification has generated considerable interest; it may be of clinical importance to be able to reliably measure liver fat content (FC). Liver biopsy and histological analysis are considered the diagnostic reference standard. But in humans, the biopsy procedure is both invasive and painful, and it presents risks for patients (7, 8). In addition, a very small liver sample may not be representative of the heterogeneous fat distribution (9). The analogous procedure in animals involves euthanizing the animals and directly analyzing liver FC, which is not ideal for longitudinal follow-up studies.

Hydrogen magnetic resonance (¹H MR) imaging offers several noninvasive methods to obtain separate fat and water images for liver FC quantification. In 1984, the Dixon technique for water and fat imaging was described (10). The authors presented two imaging sequences: one was conventional spin echo imaging with water and fat signals in-phase, and the second had the readout gradient slightly shifted to create 180° out-of-phase of the water and fat signals. Over the last decade, the Dixon technique has been improved extensively in the aspects of phase errors, noise, and artifacts. Triple-echo and multiecho in-phase and out-of-phase MR imaging were proposed, which allowed correction for T2* decay and lipid quantification (11, 12). Iterative decomposition of water and fat with echo asymmetry and least-squares estimation (IDEAL) also improved fat quantification and fat suppression (13, 14). In addition to the Dixon technique, in 1985, Haase et al. proposed the chemical shift selective imaging technique, which relies on a single frequency-selective excitation pulse with a flip angle of 90° followed by a dephasing gradient to distinguish H₂O and CH₂ differences (15). Sbarbati et al. and Lunati et al. applied chemical shift imaging to in vivo quantification of brown adipose tissue lipids (16, 17).

Animal models of hepatic steatosis and steatohepatitis have improved our understanding of the pathogenesis of NAFLD. Continued studies in animals will further clarify the pathogenesis of these disorders and will probably improve the diagnosis and treatment of human NAFLD. The obese leptin-deficient (ob/ob) mouse is an excellent model of hepatic steatosis because of its expression of many NAFLD traits (18). These mice are genetically leptin deficient, which causes excessive overeating and development of obesity, steatosis, steatohepatitis, and diabetes, among other symptoms of NAFLD (19).

Berardinelli-Seip congenital lipodystrophy type 2 (BSCL2) is a recessive disorder characterized by an almost complete loss of adipose tissue, insulin resistance, and fatty liver (20, 21). BSCL2 encodes a protein, seipin, the function of which is largely unknown (22, 23). Recently, Cui et al. created the first murine model of BSCL2 by targeted disruption of seipin and suggested a possible tissue-autonomous role of seipin in liver lipid storage (24). To better understand the effect of seipin levels on fatty liver, we measured liver FC in a human BSCL2/seipin gene knockout (SKO) mouse model in vivo and in vitro.

The purpose of this study was to evaluate the dual-echo Dixon MR imaging (in-phase and out-of-phase, IP-OP), chemical shift imaging (CSI) selective for either fat or water protons, and ¹H MR spectroscopy (MRS) for fat measurement in phantoms and in the liver of ob/ob and SKO mice, and to compare these MR-derived data with that from histological and chemical analysis.

MATERIALS AND METHODS

Animals

All animal experiments were approved by the institutional Committee on Animal Research. Obese leptin-deficient mice (C57BL/OlaHsd-Lep; ob/ob) and WT C57BL/6J mice were obtained from the Laboratory Animal Center of the Academy of Military Medical Science and the Shanghai Model Animal Research Center, respectively. They were fed normally before the experiment and kept at temperature of 20–24°C in 12 h day/night periods. Six 10-week-old male ob/ob and WT mice with an average weight of 47.5 ± 1.54 g and 26 ± 0.71 g, respectively, were used in this study. Four 10-week-old SKO and four WT C57BL/6J mice from Peking University Health Science Center had an average weight of 21.8 ± 1.61 g and 22.6 ± 1.25 g, respectively. After MR scanning, they were euthanized with an overdose of pentobarbital.

Fat/water phantoms

According to Poon et al. (25), phantoms were made by mixing known amounts of water (doped with 0.2 mM MnCl₂) and vegetable oil (extra virgin olive oil, Olivoila, Italy). Percentages of oil by volume ranged from 0% to 100% in steps of 10%. To improve the stability of the mixture, 2% of Tween 80 (polyoxyethylene sorbitan monooleate) by volume of oil was added. The mixture was blended homogeneously using an ultrasonic homogenizer (KQ-400KQE, Kunshan Ultrasonic Equipment Co., Ltd., Kunshan, China). Plastic tubes (15 mm in diameter) containing the suspensions were placed longitudinally into the magnet. Phantoms were scanned and analyzed with the same experimental protocol as that in vivo.

MR imaging and 1H MRS protocol

For in vivo MR acquisition, anesthesia was induced by inhalation of a mixture of oxygen and 5% isoflurane and maintained by a mixture of oxygen containing 0.5-1% isoflurane. All MR experiments were carried out using a 7T small animal MR system (Bruker PharmaScan, Ettlingen, Germany) interfaced to a Bruker console. The horizontal bore system was equipped with a 15 cm diameter gradient set capable of generating 375 mT/m gradient strengths in all three directions. A 31 mm inner diameter transmit-receive quatrature coil was used for MR data collection.

T₁-weighted images (T₁WIs) were acquired with a respiratory-gated spin echo sequence, 500/15 ms; section thickness, 2 mm; matrix, 256 × 256; field of view, 3.5 × 3.5 cm; and number of excitations, 4. The T₁-weighted images were used to study the distribution of fat stores and measure the volume of the liver using ImageJ software.

A point-resolved spectroscopy sequence for localized ¹H MRS sequence was used with the following parameters: repetition time (TR)/echo time (TE), 2500/20 ms; voxel, 3 × 3 × 3 mm; and number of excitations, 128. To correct for T2 decay, seven consecutive spectra were acquired with echo times of 10, 20, 30, 40, 50, 70, and 90 msec. A 9 mm³ region of interest (ROI) was placed over the left lobe of the liver, avoiding intrahepatic blood vessels. Before measurement, the automatic shimming procedure FASTMAP was used to achieve optimal uniformity of the magnetic field across the voxel volume. Water suppression was not used in any spectroscopy sequences during measurement. The free induction decay signals were Fourier-transformed. The phase and the baseline of the spectra were also corrected with great care using TOPSPIN (Bruker BioSpin MRI GmbH). Spectra were used only if homogeneity after shimming was better than 0.45 ppm, measured as the full width at 50% peak height. Spin-spin relaxation times (T2) were determined for nine different peaks (ranging from 0.9 to 5.32 ppm) by fitting the monoexponential model function M_TE = M₀ × exp(−TE/T2) to the measured peak integrals at the different TEs, and the correction factors (M₀ / M_TE) for the nine different proton resonances of triacylglycerols and the proton resonance of the water peak were calculated (26–28). The degree of FC was calculated using equation 1:

$$\begin{array}{c}{\text{FC}}_{\text{MRS}}\mathrm{=100}\times \text{integral value of fat peak/}\\ \text{(integral value of fat peak}\\ +\text{integral value of water peak)}\end{array}$$ (Eq. 1)

Fast low angle shot imaging of the entire liver were obtained for both in-phase and out-of-phase transverse dual-echoes. Imaging parameters were as follows: 500/1.47 (out-of-phase) or 1.97 (in-phase); flip angle, 40°; section thickness, 2 mm; matrix, 256 × 256; field of view, 3.5 × 3.5 cm. A radiologist used these images to position one voxel in the liver parenchyma outside the area of the great vessels for spectroscopy. And the size and location of the ROI matched those of the original ¹H MRS voxel on three consecutive in-phase and out-of-phase images. Signal intensity (SI) was measured by ParaVision 4.0 (Bruker BioSpin MRI). The SI for each ROI (3 × 3 mm) was recorded separately, and an average SI was calculated from the three images to gather information for the entire ¹H MR spectroscopy voxel volume. We used equation 2:

$${\text{IP}}_{\text{OP−IP}}\text{=100}\times {\text{[(SI}}_{\text{IP}}{\text{−SI}}_{\text{OP}}\text{)/(2}\times {\text{SI}}_{\text{IP}}\text{)]}$$ (Eq. 2)

where SI_IP is SI measured on the in-phase; SI_OP, SI measured on the out-of-phase image. The in-phase and out-of-phase images were used to study the liver FC.

CSI were also used to study liver FC. Two series of chemical shift selective images, one for fat protons and one for water protons, were derived using rapid acquisition with relaxation enhancement (RARE) sequence. Both series were performed by the 180° refocusing pulse (16, 17). Both the 90° and the 180° pulses were band-selective Gaussian pulses; however, the 90° pulse was applied when the slice-selection gradient was on, and when the 180° pulse was applied, all gradients were off to refocus only in the selected chemical-shift range. The 180° pulse was a 2.3 ms Gaussian pulse, which excited a 700 Hz bandwidth. This was appropriate to discriminate fat and water peaks, which, at 7T, have a separation of about 1,000 Hz. All images were acquired using the following parameters: 1,000/9.9 ms, flip angle, 180°; section thickness, 1 mm; matrix, 256 × 256; field of view, 3.5 × 3.5 cm; and number of excitations, 4. An ROI was drawn at the same site as the voxel used for ¹H MRS in the left lobe. The SI in the ROI was recorded for both fat selective images (SI_fat) and water selective images (SI_water). Liver FC was computed as follows (equation 3):

$${\text{FC}}_{\text{CSI}}\text{=100}\times {\text{SI}}_{\text{fat}}{\text{/(SI}}_{\text{fat}}{+\text{SI}}_{\text{water}}\times \text{R)}$$ (Eq. 3)

where R is the ratio of the fat-to-water proton densities in their pure form: a value R = 0.9 has been used in literature (25). The water fraction by volume of the sample is thus 1 − FC.

Histopathology

Mice were perfused transcardially with phosphate-buffered saline, followed by freshly prepared 4% paraformaldehyde in 0.1 M phosphate buffer (pH 7.4). Parts of liver were fixed, dehydrated, embedded, and transversely sectioned into 5 μm pieces for hematoxylin and eosin (H and E) staining. Other parts of liver were frozen in dry ice and then cryostat-sectioned at a thickness of 7 um onto poly-l-lysine slides for lipid deposition analysis by Oil Red O and hematoxylin staining.

All histopathology slides were examined and FC analysis was performed with a semiautomatic vacuole segmentation procedure (HIS-S) developed with the MATLAB software (MathWorks, Natick, MA) as describe by Gaspard et al. (29). The artificial areas, such as blood vessels, were manually excluded by a pathologist. The percent fat fraction (FC_HIS-S) was calculated by equation 4:

$${\text{FC}}_{\text{HIS−S}}\text{=the area of fat / the total tissue area}$$ (Eq. 4)

Liver lipid analysis

Approximately 100 mg of liver (wet weight) was homogenized in 1 ml of PBS. Lipid was extracted by homogenizing with 2:1 chloroform-methanol (v/v) and separated into three phases by centrifugation: the upper and lower liquid phases and the middle solid phase. The upper phase was removed as much as possible by siphoning and extracted once more. The lower phase was reserved and dried. Lipid was dissolved in 100 μl of 3% Triton X-100. The determination of triglyceride was carried out using enzymatic methods as described by Folch et al. (30).

Statistical analysis

All statistical analyses were performed using SPSS software (SPSS for Windows, version 11.0, 2001; SPSS, Chicago, IL). Numeric data were reported as mean values ± SD. For statistical comparisons, the independent-sample t-test, paired-sample t-test, and correlation test were applied. A P value of less than 0.05 indicated a statistically significant difference.

The MR methods determined the volume fractions of lipids in the liver tissue, the histological method calculated the percentage of hepatocytes showing visible fat droplets in the microscopic view, and the chemical method measured the TG content of fresh liver tissue. For statistical comparisons, a correlation test was applied to assess the relationship between LL and FC measured by the three MR methods and the histological method, and both the paired-sample t-test and correlation test were applied to assess difference and relationship between the MR methods and the histological method.

RESULTS

Quantitation in fat/water phantoms

Three MR-based methods of quantitative FC evaluation were tested in fat/water phantoms (Figs. 1, 2). Results for fat and water quantitation are shown in Table 1. The dual-echo method significantly underestimated the fat concentration compared with the gravimetric reference standard when FC ranged from 10% to 100%. There was a weak correlation between FC calculated with the dual-echo method and the known gravimetric reference standard (r = 0.03, P > 0.05). When the actual FC was less than 50% (i.e., water content was greater than FC), fat concentration was calculated using the formula: FC_IP-OP = 100 × [(SI_IP - SI_OP)/ (2 × SI_IP)], but when the actual FC was greater than 50% (i.e., water content less than FC), fat concentration was calculated using W_IP-OP = 100 × [(SI_IP − SI_OP) / (2 × SI_IP)], then FC_IP-OP = 1 − W_IP-OP. With the visual analysis of ¹H MRS fat-water peak size to guide for fat and water dominance on IP-OP images, the corrected IP-OP underestimated fat concentration when FC ranged from 10% to 60%: 4.5 ± 0%, 9.2 ± 2.0%, 16.0 ± 1.0%, 26.2 ± 0.1%, 31.9 ± 1.0%, and 56.1 ± 0.2%, respectively. There was a strong correlation (r = 0.972, P < 0.01) between FC calculated with corrected IP-OP (FC_{IP-OP correction}) and the known gravimetric reference standard.

Fig. 1.

Quantitative MR imaging of diluted fat/water phantoms. A: Chemical shift imaging (selective fat protons imaging and selective water protons imaging, CSI-fat and CSI-water). B: Dixon dual-echo MR imaging (in-phase and out-of-phase).

Fig. 2.

A: ¹H-MRS of fat/water phantoms. B: Results of fat content calculated by using IP-OP, corrected IP-OP, CSI, and MRS.

TABLE 1.

Known gravimetric FC and FC measured by MR imaging and MR spectroscopy in phantoms

Known FC	FCIP-OP	P	Corrected FCIP-OP	P	FCCSI	P	FCMRS	P
%	%		%		%		%
0	0. 1 ± 0	0.004a	0. 1 ± 0	0.004a	0	1.000	0	1.000
10	4.5 ± 0	0.003a	4.5 ± 0	0.003a	10.0 ± 1.0	1.000	9.8 ± 0.9	0.825
20	9.6 ± 2.0	0.085	9.6 ± 2.0	0.085	20.0 ± 1.0	1.000	19.7 ± 0.7	0.585
30	16.0 ± 1.0	0.033a	16.0 ± 1.0	0.033a	29.0 ± 1.0	0.184	29.4 ± 0.3	0.363
40	26.2 ± 0.1	0.004a	26.2 ± 0.1	0.004a	39.0 ± 1.0	0.225	40.7 ± 1.2	0.452
50	31.9 ± 1.0	0.029a	31.9 ± 1.0	0.029a	48.0 ± 1.0	0.038a	50.6 ± 1.7	0.638
60	43.9 ± 0.2	0.007a	56.1 ± 0.2	0.016a	57.0 ± 0.5	0.015a	60.8 ± 1.2	0.363
70	37.4 ± 1.8	0.025a	62.6 ± 1.8	0.073	67.3 ± 0.6	0.015a	71.2 ± 0.9	0.160
80	22.0 ± 0.3	0.002a	77.0 ± 0.3	0.063	77.3 ± 0.6	0.015a	81.6 ± 0.6	0.045a
90	2.6 ± 1.0	0.006a	97.4 ± 1.0	0.072	83.3 ± 0.6	0.002a	91.6 ± 1.6	0.215
100	0.4 ± 0	0.000a	99.6 ± 0	0.258	91.0 ± 1.0	0.008a	1	1.000

All values are means ± SD. Statistical analysis was done with independent-sample t-test. Number of tested phantoms, n = 3.

^aP< 0.05.

CSI underestimated fat concentration when FC ranged from 50% to 100%: 48.0 ± 1.0%, 57.0 ± 0.5%, 67.3 ± 0.6%, 77.3 ± 0.6%, 83.3 ± 0.6%, and 91.0 ± 1.0%, respectively (P < 0.05). MRS slightly overestimated fat concentration when FC was approximately 80% (81.6 ± 0.6%, P < 0.05). CSI and MRS had a high linear correlation with the known gravimetric reference standard (CSI: r = 0.998, MRS: r = 0.999, P < 0.01), but the correlation between the corrected IP-OP and the known FC was slightly weaker (r = 0.972, P < 0.01) (Fig. 3).

Fig. 3.

Correlation between known FC and FC measured by (A) CSI (r = 0.998, P< 0.01) and (B) MRS (r = 0.999, P< 0.01) in phantom, and (C) relationship between FC_CSI and FC_MRS (r = 0.998, P< 0.01).

There were strong correlations between FC calculated with corrected IP-OP and MRS (r = 0.972, P < 0.01), between CSI and MRS (r = 0.998, P < 0.01), and between corrected the IP-OP and CSI (r = 0.963, P < 0.01) (Fig. 3).

In vivo MR imaging and 1H MRS of mice

Table 2 lists liver FC measured by the three MR methods, as well as HIS-S and LL analysis (Figs.4–6). There were significant differences in liver FC between ob/ob mice and WT mice and between SKO mice and WT mice using all methods (P < 0.001), with liver FC of the ob/ob mice and SKO mice all being higher than that of WT mice. Both in vivo T1-weighted imaging and in vitro liver volume calculation proved the liver volumes of ob/ob and SKO mice were significantly greater than that of WT mice(P < 0.001), and there was no significant difference between these two methods (P = 0.798). Liver index (LI) was calculated as the volume of liver from MR or gravimetric measure divided by the total body weight (LI_MR, LI_GM). Both LI_MR and LI_GM of ob/ob and SKO mice were significantly higher than that of WT mice (P < 0.001).

TABLE 2.

Liver FC, liver volume, and LI in ob/ob and WT mice

	ob/ob	WT	P	SKO	WT	P
FCIP-OP (%)	23.62 ± 5.18	1.50 ± 1.90	0.000a	2.90 ± 0.30	1.70 ± 1.40	0.000a
FCCSI (%)	54.44 ± 5.02	4.90 ± 1.10	0.000a	17.66 ± 4.45	3.04 ± 0.80	0.000a
FCMRS (%)	55.83 ± 7.01	3.30 ± 0.99	0.000a	14.78 ± 4.10	1.31 ± 0.88	0.000a
FCHIS-S (%)	23.03 ± 2.59	0.53 ± 0.19	0.000a	7.81 ± 0.89	0.49 ± 0.18	0.000a
LL (mg/g)	55.30 ± 4.3	5.40 ± 0.78	0.000a	18.50 ± 6.42	4.81 ± 0.93	0.000a
Body weight (g)	47.50 ± 1.54	26.00 ± 0.71	0.000a	21.80 ± 1.61	22.60 ± 1.25	0.534
VolumeMR (mm )	4260 ± 547	1503 ± 199	0.000a	2100 ± 221	922 ± 183	0.000a
VolumeGM (mm )	4397 ± 416	1410 ± 205	0.000a	1720 ± 328	1109 ± 123	0.000a
LIMR	89.54 ± 9.22	56.77 ± 6.85	0.000a	96.77 ± 13.40	40.79 ± 7.79	0.000a
LIGM	92.45 ± 6.10	54.25 ± 7.68	0.000a	78.95 ± 6.19	49.04 ± 2.94	0.000a

Liver FC was measured by IP-OP, CSI, MR spectroscopy, HIS-S, and LL. Liver volume and LI were obtained by MR T₁-weighted imaging and volumetric methods. All values are means ± SD. Statistical analysis was done by inde­pendent-sample t-test. LI_GM , volume from gravimetric measure/body weight; LI_MR, volume from MR measure/body weight; Volume_GM, volume from gravimetric measure; Volume_MR, volume from MR measure.

^aP < 0.05.

Fig. 4.

In vivo Dixon dual-echo IP-OP MR imaging, CSI, and ¹H MRS in mice liver.

Fig. 6.

In vivo and in vitro measurements of liver FC and LL in mice. A, B: Liver FC measured by IP-OP, CSI, and ¹H MRS in vivo and by HIS-S in vitro from ob/ob and SKO mice, respectively. C: LL calculated by chemical method. D, E: Correlation between liver FC by chemical method and FC measured by CSI (r = 0.986, P< 0.001) and MRS (r = 0.977, P< 0.001). F: Relationship between CSI and MRS (r = 0.992, P< 0.001).

Fig. 5.

Histology of liver. A: H and E staining (400×). B: Oil Red O staining (400×). Scale bar, 20 μm.

Liver FC measured by the histological method and the corrected IP-OP was less than that by CSI and MRS (P < 0.001), but no significant difference was observed between FC_{IP-OP correction} and FC_HIS-S (P = 0.556). Liver FC measured by CSI and MRS were also not significantly different (P = 0.230) (Table 3). There was a strong correlation between FC calculated with the histological method and the three MR methods (corrected IP-OP: r = 0.882; CSI: r = 0.984; MRS: r = 0.978; P = 0.000), and a significant linear correlation was observed between LL measured by the chemical method and the FC calculated by all three MR methods and HIS-S (corrected IP-OP: r = 0.867; CSI: r = 0.986; MR spectroscopy: r = 0.977; HIS-S: r = 0.960; P = 0.000) (Table 4, ).

TABLE 3.

Differences between MR and histological HIS-S measured for mice

Parameter 1	Parameter 2	t, P
FCHIS-S	FCIP-OP	−0.600, 0.556
	FCCSI	4.394, 0.000a
	FCMRS	3. 548, 0.002a
FCMRS	FCIP-OP	−3.695, 0.002a
	FCCSI	2.494, 0.230
FCCSI	FCIP-OP	4.432, 0.000a

Statistical analysis was done with the paired-sample t-test.

^aP < 0.05.

TABLE 4.

Correlations among MR, histological HIS-S, and chemical LL analysis for mice

Parameter 1	Parameter 2	r, P
FCHIS-S	FCIP-OP	0.882, 0.000a
	FCCSI	0.984, 0.000a
	FCMRS	0.978, 0.000a
LL	FCIP-OP	0.867, 0.000a
	FCCSI	0.986, 0.000a
	FCMRS	0.977, 0.000a
	FCHIS-S	0.960, 0.000a
FCMRS	FCIP-OP	0.917, 0.000a
	FCCSI	0.992, 0.000a
FCCSI	FCIP-OP	0.891, 0.000a

Statistical analysis was done with correlation test.

^aP < 0.05.

A significant linear correlation was observed between liver FC calculated by three MR methods. The FC_CSI and FC_MRS correlations were slightly stronger (FC_CSI versus FC_MRS: r = 0.992, P < 0.001) than the FC_{IP-OP correction} and FC_CSI or FC_MRS (FC_{IP-OP correction} versus FC_CSI: r = 0.891; FC_{IP-OP correction} versus FC_MRS: r = 0.917; P = 0.000) (Table 4, Fig. 6).

DISCUSSION

There are currently three different types of imaging techniques that allow noninvasive detection of liver FC: i) ultrasound, ii) computed tomography (CT), and iii) MR methods. However, ultrasound is not considered as a sufficient quantitative tool for fat determination in liver. One major weakness of ultrasound is its operator dependency. And CT requires undesired radiation exposure in the examination of subjects (31). MR methods, therefore, become the most desirable and useful technique. A number of MR methods have been proposed for the detection of fatty liver infiltration: i) fat-sensitive MR imaging techniques based on signal phase, two-point Dixon and three-point IDEAL (also known as iterative decomposition of water and fat with echo asymmetry and least-squares estimation); ii) fat-selective MR imaging based on frequency selective excitation or chemical shift selective images (also called spectral fat or water selective imaging); and iii) ¹H-MRS (31). ¹H MRS has been by far the most promising and most sensitive noninvasive method to assess liver FC (32).

In our study, liver FCs measured by two-point Dixon IP-OP, CSI, and ¹H-MRS were compared. Two-point Dixon is a routine clinical method for the semiquantitative assessment of liver FC. However, this method neglects the T₂* signal decay and assumes that the signal difference between IP and OP echoes is due only to fat-water signal interference. With IP then OP sequential acquisition, the confounding T₂* effect is known to cause fat fraction underestimation (33–35), which was also shown in this study.

At 7T MR, the resonance frequencies of methylene and water have a separation of about 1,000 Hz, the first echo time of in-phase time is 0.5 ms, and of out-of-phase time, 1.0 ms. Our MR machine could only reach the echo time of in-phase time at 1.5 ms and out-of-phase time at 2.0 ms. Thus the corresponding in-phase and out-of-phase image signals were lower and the difference between them became smaller (12). Our results agreed with our expectation by proving that the two-point Dixon method systematically resulted in an underestimation of FC in both phantom and in vivo evaluations and that the degree of underestimation increased along with increasing FC.

In an in-phase image, the fat and water signals within a voxel are additive; in an opposed-phase image, they are subtracted from each other. The decrease in signal intensity from the in-phase to an opposed-phase image indicates the presence of both water and fat within a voxel. When the actual FC is higher than 50% (water content lower than FC), fat concentration was corrected using the formula above. But IP-OP method also presented greater underestimation compared with other methods due to the higher FC in the liver of ob/ob and SKO mice.

CSI includes two series, selective fat-protons imaging and selective water-protons imaging. Our CSI method was derived partly from the method described by Sbandrea et al. (16), who evaluated the accuracy of CSI on brown adipose tissue in rats using a 4.7T MR scanner. In phantoms, we proved using a 7T MR system that CSI was accurate in lipid quantification when FC was lower than 50%, and approximately 9% underestimation of fat occurred from 50% to 100% of FC. As centers of excitation frequency are methylene and water proton, the degree of FC underestimation increased in higher FC. Liver FC usually is lower than 50%, so CSI is able not only to accurately measure FC but also to evaluate its distribution.

Besides liver histological and chemical analysis as current standards for diagnosis and grading of hepatic FC, ¹H MRS also allows noninvasive studies in the molecular composition of tissues in vivo. ¹H MRS recorded from liver tissue usually showed two dominant signal portions, namely, the water signal (positioned at 4.7 ppm) and the signal from methylene protons of fatty acids (positioned at 1.3 ppm). In ¹H MRS, the peak area is proportional to concentration of the metabolite containing the relevant nuclei, which is also influenced by T₁ and T₂ relaxation times (36). To minimize T₁ effects, one sequence with a long repetition time (10 s) and one sequence with 1.8 s of repetition time were used by Strobel et al (26). A correction factor was also used on the difference of the two spectra. To correct T₂ relaxation, a sequence with a series of different echo times was used to calculate the spin-spin relaxation time (T₂). We also used a series of different echo times to correct the spectral data. In the phantom experiments, we proved that there was greater accuracy quantifying FC with the MRS method than with CSI when FC ranged from 50% to 100%. MRS allows quantifying FC in different tissues, and CSI allows identifying the fat distribution. Therefore, CSI and MRS should be used together to evaluate liver FC comprehensively.

Interestingly, the percentage values of liver FC from MRS and CSI were slightly different compared with HIS-S examination. Gaspard et al. suggested that this might be due to the fact that MR techniques determined the volume fractions of lipids in the liver tissue rather than the percentage of hepatocytes showing visible fat droplets in the microscopic view (29). In addition, this discrepancy also might be explained by the difficulty of quantifying microvesicular fat. In fact, software estimated the area occupied by fat droplets with a minimum diameter of 5 μm. Fat existing in droplets less than 5 μm in diameter was, therefore, not included. As the histological image was obtained by light transmitting through the slice, droplets that failed to pass through the slice completely would not be quantified.

Similarly, the results calculated by software from histogical slides and by isolation and purification of LL were not comparable because the software measured the specific surface area of fat droplets in the microscopic field, whereas LL analysis measured the quantity of lipid. However, the data of LL chemical analysis and CSI and ¹H MRS correlated very well. This may be because volume fractions of lipid in liver tissue were measured by MR methods and the TG content of fresh liver tissue was detected by the chemical method, instead of the surface area of fat droplets by the HIS-S method.

In addition to the abovementioned MR methods, the traditional T₁-wighted imaging method was used to observe the anatomic details of liver, calculate liver volume, and identify the distribution of fat in subcutaneous and internal organs (data not shown). Ob/ob mice had significantly larger livers than other groups, indicating that the severity of fatty liver positively correlates with the liver volume.

This study provided some information about fatty liver that may be useful when ob/ob mice are used as an animal model of disease. The liver is the organ most often involved in fat accumulation. In its parenchyma, the fat depot appears to be homogeneous, and it is quantified by in vivo localized spectroscopy, MR imaging, and anatomical histological and chemical surveys. Given that ob/ob mice are a model for hepatic steatosis and SKO mice are frequently a model for fat accumulation in liver, ¹H MR spectroscopy and imaging seem to be promising tools to follow the time course of fat accumulation.

The limitations of our study must be acknowledged. First, the number of mice used in the experiments was fairly small. Second, the amount of fat measured by HIS-S and liver lipid analysis might not be representative of the amount of fat throughout the entire liver (9, 37). Third, the FC measured using our MR-based methods represents the signal derived from protons contained in fatty acid molecules, as opposed to the weight of lipid per unit of liver volume.

CONCLUSION

T₁WI can be used to measure liver volume and observe fat distribution; the IP-OP method gives a FC that is significantly lower than the actual value, especially in higher fat concentrations; CSI and ¹H MRS are accurate in quantifying fat in both phantoms and liver in vivo. Given their excellent correlation and concordance with LL analysis, CSI and MRS may replace liver fat histological and chemical analysis in longitudinal studies.