Measurement of relative fat content by proton magnetic resonance spectroscopy using a clinical imager

Authors


Abstract

The aim of this study was to determine the applicability of a proton magnetic resonance (MR) spectroscopy-based technique using a clinical 1.5-T MR imager for assessment of relative fat content. Proton MR spectra were obtained from a trunk phantom and 23 volunteers using a single free induction decay measurement. The ratios of fat methyl and methylene proton resonance to the water proton resonance were compared with the ratio of oil weight to water weight for the phantom, and with the ratio of body fat to lean body mass estimated by bioelectrical impedance analysis for the human subjects. Good linear relationships were found between the MR metabolite ratio and the ratio of oil weight to water weight (r = 0.9989), and the ratio of body fat to lean body mass (r = 0.9169). This MR spectroscopy-based technique is sufficiently accurate and may be applicable to assessment of human body composition. J. Magn. Reson. Imaging 2000;11:330–335. © 2000 Wiley-Liss, Inc.

ALTHOUGH ESTIMATES of human body fat content are made using a wide variety of techniques (1, 2), there is no technique optimal for all clinical circumstances. There is a need for a body fat content measurement technique that is rapid, safe, sufficiently accurate, and feasible for clinical practice. MRI-based techniques for measurement of body fat tissue volume have recently been introduced (3–5). Total and regional fat volume can be measured using multiple sections (3–5). MRI-based techniques are safe and feasible with clinical MR imagers (3–5). However, many sections through the body and complicated image analyses for segmentation of fat tissue, correction of image artifacts, and volume calculation are required (3–5). Localized magnetic resonance spectroscopy (MRS) can separate resonance of the methyl and methylene protons of fat from that of water protons, and the ratio of the fat methyl and methylene proton resonance to the water proton resonance indicates relative fat content (6, 7). Although there are a few reports on MRS-based techniques for assessing relative fat content of small animals at high magnetic fields (8, 9), the applicability of this method for assessment of human body composition has not been established. The aim of this study was to determine the accuracy and applicability of an MRS-based technique using a clinical MR imager for assessment of relative fat content.

MATERIALS AND METHODS

A trunk phantom made of acrylic resin 30 cm in width, 30 cm in height, and 20 cm in thickness (Kyoto Kagagu, Kyoto, Japan) was used for MRS experiments. It was filled with rape seed oil and 0.5% NaCl, 0.25 mM MnCl2 solution (T1 420 msec, T2 47 msec; 24°C). Considering the difference in distribution between water and fat in the human body, the oil and water were not homogenized. The oil concentrations by weight were varied from 5.2% to 73% (oil/water W/W, 0.06–2.71) in 13 steps.

MRS experiments were performed with a Magnetom Vision (Siemens, Erlangen, Germany) operating at 1.5 T using an embedded body coil. The phantom was placed at the magnet isocenter. Automatic global shimming was performed using multi-angle projection shim (10). Non-localized MR spectra were acquired from the phantom using single free induction decay (FID) with a flip angle of 90° and 512 data points at a spectral width of 1 kHz at a room temperature of 24°C. Measurements were repeated five times at 1 minute intervals.

Acquired FIDs were processed using Luise software (Siemens). Zero-filling to 1024 data points was applied to the raw data. No apodization was applied. After fast Fourier transformation, spectral data were obtained through manual phase and baseline correction. Peak areas and resonance line widths (full width at half-maximum) for water protons and methyl and methylene protons were calculated by fitting the spectrum to a sum of Lorenzian curves interactively.

The rape seed oil used for the phantom experiments was analyzed by high-resolution 1H nuclear magnetic resonance (NMR) spectroscopy using a Unity-500 spectrometer (Varian, Palo Alto, CA) operating at 499.817 MHz. The oil was dissolved in CDCl3 (99.8 atom% D) with 0.05% tetramethylsilane. A 1H spectrum was acquired with a flip angle of 45°, 32 k data points at a spectral width of 7.5 kHz, a repetition time of 4 seconds, and 128 accumulations at 25°C.

To assess coverage of the embedded body coil and changes in resonance frequency at different positions inside the coil, FIDs from a spherical body phantom (NiSO4 • 6H2O/H2O, 1.25/1000 by weight) 25 cm in diameter within an elliptical loaded body phantom (NaCl/MnCl2 • 4H2O/H2O, 5/3/1000 by weight) were measured, moving the phantoms from -35 cm to +35 cm in Z direction from the magnet isocenter in 5-cm increments. Constant coil tune, transmitter amplitude, reciever gain, and shim currents were used throughout the measurements. Zero-filling to 1024 data points was applied to the raw data. No apodization was applied. After fast Fourier transformation, spectral data were obtained through automatic phase correction. Peak areas, resonance frequencies, and line widths for water protons were calculated by fitting the spectra to a Lorenzian curve.

To ensure that the phantom was a good model for the human body, we compared MR spectra from the phantom with those from human subjects. MR spectra were obtained from 23 volunteers [12 men and 11 women; 41.9 ± 13.7 years (mean ± SD)] The mean body mass index was 26.64 ± 4.32 (mean ± SD) kg/m2. They were positioned with their navel at the magnet isocenter. We used liquid fluorocarbon pads (Sat Pad; Alliance Pharmaceutical, San Diego, CA) to improve static magnetic field (B0) homogeneity. The same FID measurement was performed after multi-angle projection shim, as in the phantom experiments. This procedure required approximately 15 minutes of confinement in the magnet. Acquired FIDs were processed in the same fashion as for phantom experiments. The ratio of the fat methyl and methylene proton resonance to the water proton resonance was compared with the ratio of body fat to lean body mass [% body fat content/(100 - % body fat content)]. The body fat content of each subject was estimated by bioelectrical impedance analysis using an HBF-300 (Omron, Tokyo, Japan). These human studies were approved by the faculty Ethics Committee. Written informed consent was obtained from each subject.

General linear model analysis was used to test the linear relationship between the ratio of the methyl and methylene proton resonance to the water proton resonance and the ratio of oil weight to water weight, and to test equality of successive MRS measurements. General linear model analysis was also used to test the relationship between the ratio of the methyl and methylene proton resonance to the water proton resonance and the ratio of body fat to lean body mass. The level of statistical significance was set at P < 0.05. Statistical analyses were performed with SPSS for Windows software (SPSS Japan, Tokyo, Japan).

RESULTS

Figure 1 shows representative proton MR spectra from the phantom experiments. Two resonances corresponding to water protons and methyl and methylene protons were clearly identified 3.5 ppm apart from each other. The observed full line widths at half-maximum ranged from 32 to 49 Hz for the resonance of water protons, and from 35 to 74 Hz for that of methyl and methylene protons. Single FID provided a sufficient signal-to-noise ratio.

Figure 1.

Proton MR spectra from an oil water phantom at three different oil concentrations: (A) 10%, (B) 37%, and (C) 67% oil by weight. The chemical shift axis is displayed in parts per million. The water resonance serves as an internal reference (4.7 ppm). The resonance at 1.2 ppm corresponds to methyl and methylene protons. The resonance from protons of -CH=CH- (5.5 ppm) is observed at high oil concentration (C, arrow).

Figure 2 shows the correlation between the ratio of the oil methyl and methylene proton resonance to the water proton resonance and the ratio of oil weight to water weight, and indicates a good linear relationship. No interaction was found between the ratio of the oil methyl and methylene proton resonance to the water proton resonance and successive measurements (P = 0.709), and the effect of successive measurements on the regression model was not significant (P = 0.993). The estimated slope and intercept were 0.927 (P < 0.001, 95% confidence interval: 0.916–0.938) and −0.006 (P = 0.419, 95% confidence interval: −0.021–0.009), respectively.

Figure 2.

The relationship between the ratio of the oil methyl and methylene proton resonance to the water proton resonance and the ratio of oil weight to water weight. Error bars indicate 95% confidence interval for means.

Figure 3 shows a high-resolution 1H spectrum from the oil used for the phantom experiments. The assignment, chemical shift, and fractional intensity for each group of resonances are given in Table 1.

Figure 3.

A high-resolution 1H spectrum of the oil used for phantom experiments. Assignments for resonances are given in Table 1. TMS = tetramethylsilane.

Table 1. Assignment, Chemical Shift, and Fractional Intensity for Each Group of Resonances Observed by High-Resolution 1H NMR Spectroscopy*
ResonanceAssignmentδ (ppm)% Fractional intensity
  • *

    Chemical shifts were referred to the resonance of tetramethylsilane.

1−CH30.887.41
2−CH[DOUBLE BOND]CHCH2CH30.970.67
3−(CH2)n1.3055.26
4−CH 2CH2CO−1.616.31
5−CH[DOUBLE BOND]CHCH 22.0310.94
6−CH2CO−2.305.96
7[DOUBLE BOND]CHCH 2CH[DOUBLE BOND]2.791.78
8c1,3 glycerol backbone4.14, 4.293.33
9c2 glycerol backbone5.250.92
10−CH[DOUBLE BOND]CH−5.357.42

Figure 4 shows the relationships between distance from the magnet isocenter and measured signal intensity, resonance frequency, and line width of phantom water proton resonance. Measured signal intensity descended from 15 cm distant from the magnet isocenter, diminished by half at ±25 cm, and reached noise level at ±35 cm (Fig. 4A). Variation of resonance frequency within the active coil volume from the resonance frequency at the magnet isocenter did not exceed 0.34 ppm (Fig. 4B). Measured line width at the magnet isocenter was 11.9 Hz and broadened approximately twofold at ±20 cm, fourfold at −25 cm, and tenfold at −30 cm (Fig. 4C).

Figure 4.

a: The relationship between distance from the magnet isocenter and measured signal intensity of phantom water proton resonance. b: The relationship between distance from the magnet isocenter and measured resonance frequency of phantom water protons. c: The relationship between distance from the magnet isocenter and measured line width of phantom water proton resonance.

Figure 5 shows a representative proton MR spectrum from a healthy volunteer. Two resonances corresponding to water protons and methyl and methylene protons were clearly identified. The observed full line widths at half-maximum measured 57.7 ± 8.0 (mean ± SD) Hz for the resonance of water protons, and 67.3 ± 12.9 (mean ± SD) Hz for that of methyl and methylene protons.

Figure 5.

Proton MR spectrum from human body.

Figure 6 shows the correlation between the ratio of the fat methyl and methylene proton resonance to the water proton resonance and the ratio of body fat to lean body mass, and indicates a good linear relationship. No interaction was found between the ratio of the fat methyl and methylene proton resonance to the water proton resonance and successive measurements (P = 0.999), and the effect of successive measurements on the regression model was not significant (P = 0.999). The estimated slope and intercept were 1.293 (P < 0.001, 95% confidence interval: 1.188–1.398) and 0.163 (P < 0.001, 95% confidence interval: 0.114–0.212), respectively.

Figure 6.

The relationship between the ratio of the fat methyl and methylene proton resonance to the water proton resonance and the ratio of body fat to lean body mass. Error bars indicate 95% confidence interval for means.

DISCUSSION

Although there are a few reports on MRS-based techniques for assessing body fat and water content of small animals at high magnetic fields (8, 9), the applicability of this method for assessment of human body composition has not been established. The results of this study demonstrate that our MRS-based technique using a clinical MR imager has sufficient accuracy for measurement of relative fat content, and suggest that it may be applicable to assessment of human body composition.

MRS without localization has several technical problems. The magnet and embedded body coil are too small to measure an FID from the whole human body at one time. It is impossible to determine which parts of the body are contributing to the observed signal. The result of measurements of the phantoms at different positions suggests that the majority of an observed signal originates within 25 cm from the magnet isocenter in Z direction. Thus, a volume ranging from the chest to pelvic region is thought to contribute to the observed signal from the human body. The bodily extent of the volume contributing to the observed signal varies with the individual. A large active volume of the embedded body coil may cause considerable resonance frequency shift because of B0 inhomogeneity. The result of measurements of the phantoms at different positions showed that variation of resonance frequency within the active volume of the embedded body coil did not exceed one-tenth the resonance frequency difference between water protons and methyl and methylene protons. Considerable resonance line broadening was observed at both ends of active coil volume. However, the effect of B0 inhomogeneity on observed human body spectra may be smaller than expected from these results because of insensitivity of the embedded body coil at positions far from the magnet isocenter.

Resonance frequency also shifts at interfaces of regions having different magnetic susceptibility. This effect is most apparent at air-body interfaces. The contours of the human body are much more complex than those of the phantom used here. The combined use of liquid fluorocarbon pads and multi-angle projection shim could minimize this problem (10–12).

The observed line widths of the resonance from the oil methyl and methylene protons in the phantom experiments were so broad that it was difficult to resolve resonances from protons of -CH3 (0.88 ppm), -CH=CHCH2CH3 (0.97 ppm), -(CH2)n- (1.30 ppm), -CH2CH2CO- (1.61 ppm), -CH=CHCH2- (2.03 ppm), -CH2CO- (2.30 ppm), and =CHCH2CH= (2.79 ppm), which were clearly resolved by high-resolution NMR spectroscopy. Thus, calculated peak areas for the resonance of the oil methyl and methylene protons in the phantom experiments included these resonances, which accounted for 88.33% of the resonances from protons of the oil. For the same reason, it was difficult to resolve resonances from protons of the glycerol backbone (4.14, 4.29, and 5.25 ppm), which accounted for 4.25% of the resonances from protons of the oil, from the resonance of water protons. Adequate restriction of fitting parameters could minimize effects of the resonance from protons of -HC=CH- (5.35 ppm), which was observed at high oil concentration, on the calculated peak area of the water proton resonance. The ratio of the oil methyl and methylene proton resonance to the water proton resonance could therefore be used as an indicator of relative oil content within the range of the phantom experiments. Saturated positions account for approximately 90% of positions in fatty acid chains in human adipose tissue (13). Resonances from protons of saturated positions therefore represent the majority of protons in fatty acid in human adipose tissue (14). Our MRS-based technique may be useful for assessing human body fat as well as the experimental phantom tested here, which contained vegetable oil.

The human body is structurally and chemically much more complex than the phantom used here. Body water consists of multiple fractions with different relaxation rates (15). Some fractions of tightly bound water may be invisible to our technique. Body fat does not have the fluidity of vegetable oil. Each human subject has a unique physique and body composition. Despite the structural and chemical complexity of the human body, the observed human body spectra were quite similar to those for the phantom, permitting curve fitting and area calculation as in the phantom experiments. There was a good linear relationship between the ratio of the fat methyl and methylene proton resonance to the water proton resonance and the ratio of body fat to lean body mass. This MRS metabolite ratio can readily be used as an indicator of body fat content. Our MRS-based technique, however, permits not direct measurement of body fat content but content of body fat relative to that of visible water, and therefore requires calibration by other methods for estimation of body fat content.

Our MRS-based technique using a clinical MR imager provides a noninvasive method for evaluation of relative fat content and may therefore be useful in clinical practice. MRI studies for assessment of body fat distribution can be performed consecutively. Relative fat content and body fat distribution can be rapidly assessed by combined use of our MRS-based technique and MRI. Comparisons with other methods for estimates of body fat are needed to determine the validity and usefulness of our MRS-based technique for assessing human body composition.

Ancillary