Alterations in hepatic energy metabolism are typical for inflammatory and neoplastic liver diseases (1–3). Recently, evidence has shown that abnormalities in energy metabolism can also underlie non-alcoholic fatty liver in insulin-resistant and/or type 2 diabetic patients (4). Thus, information on regional alterations in liver metabolism may contribute to early diagnosis of various liver diseases in humans.
During the past decade, in vivo phosphorus magnetic resonance spectroscopy (31P-MRS) has been shown to be a valuable non-invasive research tool to investigate metabolic changes in the liver, specifically with regard to diffuse liver disease (5), viral (6) and alcoholic liver disease (7), cirrhosis (8–11), and liver metastases (12, 13). These studies used metabolite peak ratios as a surrogate for energy metabolism in the human liver. 31P-MRS has also been used as a tool for determining absolute concentrations of metabolites in the healthy human liver (14–18) and can yield more detailed information on liver function than metabolite peak ratios.
Despite the increasing use of this approach there are discrepancies between the reported results (14–18), which may reflect different signal acquisition schemes, different corrections of partial longitudinal saturation effects, as well as differences in postprocessing and quantification protocols (16). Thus, there is a need for a robust, simple, and reproducible method for acquiring 31P metabolite concentrations in the human liver.
The most promising approach is currently magnetic resonance spectroscopic imaging (MRSI). The main advantage of multivoxel MRSI resides in the ability to provide a measure of the spatial distribution of metabolites. On the other hand, time demands of typical MRSI measurements limit their use to one-dimensional (19–21) or two-dimensional MRSI (10, 18) protocols with a maximum of 16 × 16 phase-encoding steps. Three-dimensional (3D) MRSI was used only once by Li et al. (17), with an 8 × 8 × 8 matrix. They employed nuclear Overhauser effect enhancement during acquisition and simulated B1-field correction in the quantification procedure. Despite using the highest spatial resolution (27 cm3), no information was provided for the regional metabolite concentrations in human liver. Of note, these studies had some limitations due to the large volumes of interest (27–1600 cm3), giving low spatial resolution at 1.5 T.
The aims of this study were to introduce an improved protocol (i.e., higher spatial resolution, higher magnetic field, simplified acquisition, and generation of maps of concentration spatial distribution) for the absolute quantification of in vivo hepatic 31P metabolites by employing a high-resolution 31P 3D k-space weighted MRSI using B1-insensitive adiabatic excitation, and also to test the feasibility of quantification of 31P metabolite concentrations in the liver of healthy volunteers.
Ten healthy volunteers (8 males, 2 females; age 25 ± 1 years; body mass index 22.2 ± 0.7 kg/m2) participated in this study, which was approved by the ethics committee of the Medical University of Vienna. Experiments were performed after overnight fasting and informed written consent had been obtained from all participants.
Hardware and MRI Acquisition Protocol
All measurements were performed on a 3-T whole-body scanner (Medspec S30/80; Bruker Biospin, Ettlingen, Germany). A 10-cm-diameter, linearly polarized surface coil, tuned to 1H and 31P resonance frequencies, was used for transmission and signal reception. Volunteers were scanned in the prone position, with the coil positioned under the lateral aspect of the liver (Fig. 1a).
A small cylindrical reference sample (volume = 1 ml, diameter = 10 mm, height = 13 mm) filled with triphenylphosphate (TPP) diluted in chloroform, giving a stable and detectable signal with a chemical shift of approximately −12 ppm, was placed at a fixed location in the center of the surface coil (Fig. 1a,b).
To improve magnetic field homogeneity, linear shimming was performed using the proton signal of the surface coil and the automated, non-localized procedure was provided by the manufacturer.
Axial MRI gradient-echo images (15 slices, slice thickness = 8 mm, field of view = 20 × 20 cm, matrix size = 128 × 96, TE = 3.4 msec, TR = 120 msec) covering the region of interest were acquired during 1 breath-hold immediately before the 31P-MRSI scan (Fig. 1b).
3D Spectroscopic Imaging
We used the 31P 3D k-space weighted MRSI localization technique with an adiabatic B1-insensitive half-passage excitation pulse (2.5-msec sin/cos-modulated AHP, bandwidth = 4300 Hz) (Fig. 2). No pulse adjustment prior to MRSI measurement was necessary because of adiabatic excitation. The pulse was followed by phase-encoding gradients in all three dimensions. The 20 × 20 × 20-cm FOV was encoded using a 13 × 13 × 13 matrix. FIDs (1024 complex points, spectral width (SW) = 10,000 Hz) were acquired after the phase-encoding gradients. The number of times a certain phase-encoding step was averaged was determined by a 100% Hanning filter. TR was 1000 msec and the whole protocol, including set-up, took approximately 45 min.
Phantom Data Set Acquisition and Quantification
For the quantification of hepatic metabolites, a simulated phantom experiment was performed as described by Meyerhoff et al. (14). The measurement protocol with the same geometry and parameters was repeated on a cylindrical phantom containing a solution of KH2PO4 doped with gadolinium diethylenetriamine pentaacetate anions (Gd-DTPA2) to reduce T1 (volume = 4 L, diameter = 20 cm, height = 13 cm, T1 = 2.88 sec) with a known concentration (c = 50 mmol/L) for creating a calibration data set. Identical geometry including the reference sample during all in vivo measurements permits the use of one calibration data set for all quantifications. The temperature stability of the reference sample's signal was confirmed in the temperature range 20°–40°C by non-localized pulse-acquired measurement (TR = 1 sec) to ensure independence of heating by the volunteer's body. Because of the nonuniform excitation profile of the adiabatic pulse, five different pulse-frequency offsets were used to prepare reference data sets with offsets corresponding to the in vivo spectral peak positions of phosphomonoesters (PME), inorganic phosphate (Pi), phosphodiesters (PDE), gamma-adenosine triphosphate (γ-ATP), and 0 ppm.
Data Processing and Quantification
Data were processed offline using the MRSI software tool developed in our laboratory (22) and were quantified in jMRUI (Java-based magnetic resonance user interface) (23) software using AMARES (advanced method for accurate, robust, and efficient spectral fitting of MRS data with use of prior knowledge) (24). The complete quantification of one subject's MRSI data took approximately 20–30 min, depending on the number of quantified spectra. The following steps were performed:
1Postprocessing consisted of 4D Fourier transformation after spatial zero-filling to a 17 × 17 × 17 matrix, spectral zero filling to 4096 complex points, and 10-Hz exponential line broadening. Automatic phasing (zero order) was used.
2The data were visualized using the MRSI tool. The spatial response function (SRF) (Fig. 4) of the reference sample was calculated and maximal SRF was used as a reference point for voxel shifting to precisely adjust alignment of data if the position of the coil with the reference sample was not located in the correct position. The inherent resolution of the k-space weighted experiment, given by the Rayleigh criterion, is indicated by the 64% level of the SRF experiment and was defined previously by Pohmann et al. (25). Therefore, the experimentally determined width at the 64% level of the SRF was used as measure of the true spatial resolution. The signal from the reference sample was assessed by summing over a 5 × 5 × 5 matrix of spectra around the voxel that contained the highest signal of the reference solution. The summed signals across the reference sample were used for the correction of different coil loads in all experiments (26).
3Liver spectra were marked in the MRSI tool and exported as single FIDs. The criteria for using spectra for quantification were as follows: (a) the spectrum had to be located in liver tissue based on 1H images, and no PCr signal was visible; and (b) the spectrum had a sufficient signal-tonoise ratio (SNR).
4jMRUI software (23) with AMARES (24) time-domain fit algorithm were used to evaluate the selected spectra with prior knowledge, as described by Schmid et al. (27).
5The jMRUI results were imported back into the MRSI tool and absolute quantification was performed; the various metabolite concentrations were determined for the chosen voxels via the equation:
where c(x,y,z) is the absolute metabolite concentration (in mmol/L) of the voxel with coordinates x, y, and z; cp is the concentration of phantom solution used for calibration (in mmol/L); I(x,y,z) and Ip(x,y,z) are the corresponding signal integrals from voxels with the same x, y, and z coordinates; S and Sp are the corresponding saturation factors calculated from measured T1 times; and Iref and Iref(p) are the corresponding signals of the TPP reference sample (28). Calibration signal integrals, Ip(x,y,z), were related to the frequency offset of the in vivo signal. Liver metabolite T1 values were used in the calculation of saturation factors as reported previously (27). The signal was not corrected for T2 relaxation because the delay between the end of the excitation pulse and the onset of data acquisition was only 1.26 msec (Fig. 2). The γ-ATP resonance was used for calculation of ATP concentration instead of β-ATP, because of its attenuation due to the limited bandwidth of the excitation pulse.
Protocol Validation and Reproducibility Tests
For validation of the quantification procedure, a cylindrical phantom containing KH2PO4 (c = 20 mmol/L) doped with Gd-DTPA2− (volume = 4 L, diameter = 20 cm, height = 13 cm, T1 = 2.28 sec, T1 was measured by inversion recovery (IR) experiment) was used. Identical protocols to those used for the in vivo experiments were performed three times with complete phantom replacement. One volunteer was measured consecutively three times with complete repositioning.
Concentrations are presented as weighted mean ± weighted SD of voxels quantified in the experiment (weighting factor was the SNR of the quantified signal). Mean concentrations are presented as mean ± SEM.
An axial image slice through the phantom with a concentration of c = 20 mmol/L is shown in Fig. 3a. The MRSI grid is overlaid on the image and selected voxels are highlighted with yellow borders. A 31P image of the phantom created from the peak estimates by jMRUI is displayed in Fig. 3b, and a 31P absolute concentration map created after the absolute quantification of the selected spectra is shown in Fig. 3c. A typical experimentally determined SRF of reference sample (diameter = 1 cm) is displayed in Fig. 4. Experimental results of three phantom (c = 20 mmol/L) measurements with complete repositioning are summarized in Table 1. There were 286 voxels quantified for each measurement. The mean concentrations (mean ± SD) of three experiments were estimated to be 19.45 ± 0.82 mmol/L, 19.04 ± 0.91 mmol/L, and 19.59 ± 0.74 mmol/L, respectively, and the true concentration of the phantom (20 mmol/L) within 1 SD of the measurements.
Table 1. Estimated Concentrations and Widths at 64% Level of Reference Sample's SRF for Three Phantom Experiments with Complete Repositioning
A total of 286 voxels per measurement were quantified.
Weighted mean based on the S/N ratio of the signal in 286 quantified voxels.
Weighted SD based on the S/N ratio of the signal in 286 quantified voxels.
Width measured at the 64% level of the reference sample's SRF.
In Vivo Experiments
A typical axial image slice within the volunteer's liver is displayed in Fig. 5a. The γ-ATP signal and absolute concentration maps are shown in Fig. 5b and 5c, respectively. An example of the AMARES quantification is given in Fig. 6. The low intensity of the phosphocreatine signal at 0 ppm indicates excellent signal localization within the liver with only negligible contamination by the overlying skeletal muscle. Experimental results of three measurements with complete repositioning of the volunteer are summarized in Table 2. The reproducibility rates for PME, Pi, PDE, and γ-ATP quantification were 1.82%, 7.88%, 2.93%, and 1.55%, respectively. The individual results of the PME, Pi, PDE, and ATP signal-to-noise weighted-average concentrations, number of quantified voxels, and width at the 64% level of the reference sample's SRF are summarized in Table 3. The mean concentrations of PME, Pi, PDE, and γ-ATP from 10 volunteers were 2.24 ± 0.10 mmol/L, 1.37 ± 0.07 mmol/L, 11.40 ± 0.96 mmol/L, and 2.14 ± 0.10 mmol/L, respectively. The mean number of quantified voxels was 57.3 and the true resolution of experiments was estimated to be (2.61 ± 0.01 cm)3, and 17.8 ± 0.22 cm3 in volume.
Table 2. Estimated PME, Pi, PDE, and γ-ATP Concentrations and Widths at the 64% Level of Reference Sample's SRFs of Three In Vivo Experiments with Complete Repositioning of the Volunteer
This study has investigated the feasibility of k-space weighted high-resolution 31P 3D MRSI for measuring 31P-metabolite absolute concentrations in the human liver at 3 T.
Higher magnetic field (i.e., 3 T), improved spatial resolution (i.e., 17.8 ± 0.22 cm3 for experimentally determined true resolution), k-space weighted acquisition (“side-lobe-free” SRF; Fig. 4), B1 homogeneity-insensitive excitation, possibility of creating detailed maps of the spatial distribution of human liver metabolite concentrations (Fig. 5), and high reproducibility of the protocol (Tables 1 and 2) are the main advantages of the approach presented. The mean concentrations of PME, PDE, and γ-ATP are in good agreement with earlier studies: PME 0.7–3.8 mmol/L; Pi 1.4–2.8 mmol/L; PDE 3.5–12.5 mmol/L; and ATP 1.6–3.8 mmol/L (14–18, 29). Both the 3-T magnetic field and smaller voxel volumes with lower intervoxel dephasing effect and lower susceptibility changes led to improved spectral resolution in this study. Although the better spectral resolution of PME–Pi–PDE resonance lines was achieved, this spectral region still suffered from partial overlapping of signal peaks. This resulted in the highest variation of Pi concentration in reproducibility tests in vivo and could explain why average concentration of Pi shifted closer to the lower end of published values.
Previous studies that used 31P MRSI to estimate liver metabolite concentrations (17, 18, 30), had relatively low spatial resolution (27–64 cm3) and thus were limited to using only a single spectrum for the quantification. In this study, on average, 57 ± 7 spectra with sufficient SNRs were quantified per subject. This fact could potentially reduce systematic and statistical errors due to the postprocessing, fitting algorithms, and absolute quantification, and allows for development of reliable metabolic maps.
Surface coils are frequently used to investigate the liver and produce an inhomogeneous B1 field and varying flip angle distributions across the coil (31). We employed a B1-insensitive excitation pulse, which has an advantage over conventional pulses because it produces 90° flip angles regardless of B1. Adiabatic pulses are not slice-selective, which is often the case in the 3D MRSI technique where phase encoding is used in all three dimensions. The sensitivity volume of the surface coils is approximately a hemisphere with the coil radius. This is an advantage because 3D MRSI protocols can be used without the need for using outer volume suppression or another prelocalization technique to avoid a fold-over effect. On the other hand, it restricts FOV size and limits applications to superficial parts of the liver. From this point of view, the 10-cm surface coil used in this study seems to offer a sensible compromise.
The k-space weighted acquisition scheme is the technique of choice for measurements where signal averaging per phase-encoding step is used (32, 33). The k-space weighted MRSI significantly reduces the spatial cross-talk between adjacent voxels. It was shown in human heart in vivo that, at an identical degree of sensitivity, the side lobes of the SRF are substantially attenuated, and the reduced spatial contamination improves the experimental reproducibility (25). Although the liver is more homogeneous than the heart, the k-space weighted acquisition minimizes contamination of the spectra by signals from adjacent muscle tissue and provides localized information on liver tissue. This is of increased relevance when studying the focal liver lesions.
The disadvantage, of course, is reduced spatial resolution. We considered this limitation and calculated the real spectral resolution based on individual experimental data (Tables 1–3 and Fig. 4), which was still higher than in previous studies, with nominal resolution and a less perfect SRF shape.
3D k-space weighted MRSI localization fulfilled a prerequisite of good separation of metabolically and magnetically different signals from liver and adjacent muscle tissue. Our method allows for estimation of 31P metabolite concentrations in a measurement time of 34 min plus set-up and image acquisition (∼10 min), an easily tolerable duration for most patients. 31P spectra were acquired without respiratory or motion triggering. Effects of breathing motion could have broadened the spectral lines and changed them toward Gaussian line shapes. However, the motion of the liver could be expected to occur mostly in the superoinferior direction as movements of the patient were restricted by scanning them in the prone position, as the muscle–liver border runs mostly in the coronal plane.
In conclusion, a high-resolution 3D 31P MRSI technique that allows quantification of hepatic 31P metabolite concentrations and their spatial distribution in human liver was successfully implemented. Owing to the simple and robust nature of the experimental design, this method should not only be suitable for studies on physiologic regulation of hepatic energy metabolism but also for future clinical applications. In particular, this approach should contribute to a better understanding of abnormalities underlying metabolic liver diseases such as non-alcoholic liver disease and focal (neoplastic) liver diseases.