High-Resolution Echo-Planar Spectroscopic Imaging of the Human Calf

Background This study exploits the speed benefits of echo-planar spectroscopic imaging (EPSI) to acquire lipid spectra of skeletal muscle. The main purpose was to develop a high-resolution EPSI technique for clinical MR scanner, to visualise the bulk magnetic susceptibility (BMS) shifts of extra-myocellular lipid (EMCL) spectral lines, and to investigate the feasibility of this method for the assessment of intra-myocellular (IMCL) lipids. Methods The study group consisted of six healthy volunteers. A two dimensional EPSI sequence with point-resolved spectroscopy (PRESS) spatial localization was implemented on a 3T clinical MR scanner. Measurements were performed by means of 64×64 spatial matrix and nominal voxel size 3×3×15 mm3. The total net measurement time was 3 min 12 sec for non-water-suppressed (1 acquisition) and 12 min 48 sec for water-suppressed scans (4 acquisitions). Results Spectra of the human calf had a very good signal-to-noise ratio and linewidths sufficient to differentiate IMCL resonances from EMCL. The use of a large spatial matrix reduces inter-voxel signal contamination of the strong EMCL signals. Small voxels enabled visualisation of the methylene EMCL spectral line splitting and their BMS shifts up to 0.5 ppm relative to the correspondent IMCL line. The mean soleus muscle IMCL content of our six volunteers was 0.30±0.10 vol% (range 0.18–0.46) or 3.6±1.2 mmol/kg wet weight (range: 2.1–5.4). Conclusion This study demonstrates that high-spatial resolution PRESS EPSI of the muscle lipids is feasible on standard clinical scanners.


Introduction
Proton ( 1 H) magnetic resonance spectra of lipids in skeletal muscle are dominated by two spectral lines originating from the methylene protons (-CH 2 -)n of extra-(EMCL) and intramyocellular (IMCL) lipids (EMCL CH2 , IMCL CH2 ). Less intense resonances are also observable, originating from the methyl protons (CH 3 ), methylene protons a and b to COO (-CH 2 -CH 2 -COO), a to C = C (-CH 2 -CH = CH-CH 2 -), diallylic methylene ( = CH-CH 2 -CH = ), and methine protons (-CH = CH-) [1,2]. IMCL are stored in the form of small droplets in a magnetic susceptibility homogeneous environment (cytoplasm) of muscle cells. In such circumstances, the positions of IMCL spectral lines are independent in orientation of the myocytes relative to the external magnetic field B 0 [2,3]. The bulk fat (EMCL) is, in contrast, stored subcutaneously, in bone marrow and as interstitial (adipose) tissue. The anisotropic shape and geometrical arrangement of EMCL compartments result in a bulk magnetic susceptibility (BMS) shift of its spectral lines [3]. If the shape of the EMCL compartments resemble ''infinitely long'' cylinders, the spectral line shifts range from 0.2 ppm to 20.1 ppm between parallel and perpendicular orientations of the cylinders relative to B 0 [3][4][5][6]. However, EMCL compartments in skeletal muscle do not follow a strict cylindrical shape. BMS effects could be therefore more complex. This expectation is supported by the study of Steidle et al [7]. The authors proposed a deconvolution method to compute magnetic field distribution (MFD) inside the voxel using the spectra of muscle lipids and bone marrow. Spectra of soleus muscle revealed that magnetic field deviations experienced by the EMCL compartments were in the interval from ca 0.6 to 20.2 ppm in respect of IMCL CH2 . From Steidle et al study [7] it follows that EMCL spectral line shifts somewhat larger than 0.2 ppm or less than 20.1 ppm relative to IMCL can be expected.
The quantification of IMCL became important mainly because of its relation to insulin resistance in obese and non-obese adults as well as in non-diabetic offspring of type 2 diabetic subjects [8][9][10]. IMCLs are involved in lipid metabolism with a turnover of some hours. IMCL concentrations vary during physical exercise and also depend on training status, diet, fasting, oxygen supply, etc [4].
Fitting of IMCL resonances is often hampered by overlap with intense EMCL spectral lines. Single-voxel MRS with its large voxel size (.1 cm 3 ) is the most sensitive to this effect [2,6]. The voxel must be placed in the muscle regions free from large deposits of EMCL tissues. Improper voxel positioning will result in a spectrum with indistinguishable IMCL lines. To overcome this limitation, magnetic resonance spectroscopic imaging techniques (MRSI) were proposed [11][12][13][14][15][16][17]. The reduction of IMCL contamination was achieved by using smaller voxel sizes and by a post-acquisition selection of voxels with acceptable separation of IMCL CH2 line from EMCL CH2 . Conventional STEAM, PRESS or slice selective 2D MRSI techniques reached nominal voxel sizes 565610 mm 3 using spatial matrices 32632 or 36636 [11][12][13][14][15]. Acquisition time was 20-35 minutes. The fundamental problem of MRSI is inter-voxel signal contamination (signal bleeding) of the strong EMCL signals from subcutaneous fat, bone marrow or larger facial layers into the adjacent regions of muscle due to limited k-space sampling. One solution to this problem is presaturation of large fat structures followed by removing the residual signals by a lipid extrapolation procedure [14,15]. Another solution is to acquire MRSI data set with high spatial resolution [16,17]. We previously used 2D MRSI based on a gradient echo sequence with step increments of TE [16]. Spectral matrix 1286256 led to the acquisition voxel size 1.260.6615 mm 3 . Newcomer et al [17] used a conventional 2D MRSI sequence with decreased flip angles, short TR (146 ms), spectral matrix 64664, and voxel size 2.562.5610 mm 3 . The disadvantages of both high-resolution methods were poor water suppression, decreased spectral quality, and the inability to detect other than most intense EMCL CH2 and IMCL CH2 lines.
The focus of this study was to develop and implement a highresolution 2D PRESS EPSI technique on a clinical MR scanner, to visualise the bulk magnetic susceptibility (BMS) shifts of EMCL spectral lines, and to investigate the feasibility of this method for the assessment of IMCL lipids.

Ethics Statement
The study was approved by The Regional Ethical Review Board in Uppsala and written consent was obtained from all participants.

Subjects
Six male volunteers participated in this study. The subjects were healthy, non-obese, passive in free time physical activity and with sedentary occupations. Median age and body mass index were 30.5 years (range: 22-61), and 23.361.5 kg/m 2 (range: 21-25.5), respectively.

Data Acquisition
All the experiments were performed by using a 3 T scanner (Achieva, Philips Healthcare, Best, Netherlands). The trapezoidal echo-planar readout gradient train was implemented into a manufacturer point-resolved spectroscopy (PRESS) localization module (Fig. 1). It enabled utilization all possibilities of the manufacturer spectroscopic imaging package including water/fat and outer volume suppression modules, excitation pulse shapes, etc. The PRESS localization technique was chosen because this method effectively reduces risk of signal contamination from outside the VOI, e.g. from the other leg (in our case). However, PRESS excitation ( Fig. 1) was effective only for the slice selection ( Fig. 2), because our intention was to perform spectroscopic imaging in the whole slice (calf). A whole-body coil was used for excitation, and the standard, manufacturer provided flexible twoelement coil (20 cm diameter) placed in front of and behind the calf served as the receiver. The subject lay in a supine position with the most extended part of the calf in the center of the receiver coil. 2D PRESS EPSI measurement was performed in the transversal plane by means of 64664 spatial matrix (Fig. 2); FOV 192 mm; nominal voxel size 363615 mm 3 (0.135 cm 3 ); acquisition bandwidth 128 kHz; TR 1500 ms, and TE 38 ms (minimum TE). One and four signal averages were used for non-water and water suppressed scans, respectively. Typical size of the excitation volume (PRESS box) was 1606160 mm in the transversal plane (Fig. 2). Thickness of the PRESS box was 15 mm in feet-head direction. Chemical shift displacement between PRESS box of trimethyl ammonium containing compounds (TMA) at ca 3.2 ppm and methylene (-CH 2 -)n PRESS box (,1.5 ppm) was 44.9, 16.4, and 4.2 mm in anterior-posterior, left-right, and feet-head directions, respectively (Fig. 2). 90 o RF-pulse with broader bandwidth (BW) produces lower chemical shift artefact in the left-right direction than 180 o RF-pulse with narrow BW in the anterior-posterior direction (Fig. 2). Radiofrequency of the PRESS pulses was centered to ,1.5 ppm in water-suppressed scans, i.e. there was no chemical shift displacement between the water and methylene (-CH 2 -)n PRESS boxes. The period of readout gradients was 1.6108 ms and readout gradient train consisted of 128 positive and 128 negative lobes. Trapezoidal readout gradient waveforms consisted of linear ramps of 0.1566 ms duration, a constant plateau of 0.4922 ms duration and an amplitude 15.66 mT/m. The duration of the linear ramps was limited by the slew rate of 100 mT/m/s fixed by manufacturer. Conventional phaseencoding was performed in the anterior-posterior direction (Fig. 1). Magnetic field homogeneity in the whole slice was improved by iterative first order shimming (Fig. 2). Typical linewidth of the water line was between ,15 and ,25 Hz. Water suppression was performed by applying two bandwidth selective RF pulses and spoiler gradients. Suppression was achieved by adjusting the tip angle of the second RF pulse such that the longitudinal magnetization of the water signal was minimal at the time of the first PRESS excitation RF pulse. Water suppression bandwidth was 140 Hz. An acquisition of two interleaved gradient echo trains was performed in order to increase spectral bandwidth [21]. The beginning of the second echo train was shifted by a half period (0.8054 ms) of the readout gradients. Combination of both echo trains resulted in the effective inter-echo time 0.8054 ms corresponding to a spectral bandwidth of 9.72 ppm. The total net measurement time of our sequence with two interleaves of gradient echo train was 3 min 12 sec for non-water-suppressed (1 acquisition) and 12 min 48 sec for water-suppressed scans (4 acquisitions).

Data Processing
Measured raw data were transferred to a Linux workstation for processing. Data processing software was developed in-house. Measured data were reorganized to the data sets containing either odd or even echoes. The first 128 echoes arising either from positive or negative readout gradients were processed for each element of the receiver coil separately, i.e., four (k read , k phase , k t ) = (64, 64, 128) data matrices were processed. Measured matrices were zero-filled to size (64, 64, 512). Data processing began with optimized 2D Hanning filter [24] applied across the k read and k phase directions to reduce signal bleeding. Radial profile (Fig. 3a) of the filter (Fig. 3b) was computed using the weighting function [24]: w(k n ) = b/2[1+ cos(2pk n Dx/a)], where k n = [n -(N +1)/2]Dk; n = 1, 2, …, N; N is the number of phase-encoding steps; Dk = 1/FOV is the k-space increment; Dx = FOV/N is the nominal spatial resolution, a = 1.71, and b = 1.47. The first FFT was performed along the k t axis. The chemical shift artifacts caused by readout gradients were removed using a first-order phase correction [25,26]. Data processing continued with 2D FFT along k read and k phase dimensions. The spatial distribution of the magnetic field DB(x, y) in the measured slice was computed from the position of the highest spectral line of non-water suppressed magnitude spectra [25,26]. The voxel spectra were then corrected by shifting about DB/B 0 along the spectral axis (Fig. 4). The magnitude spectra of the resulting four spectral matrices were then averaged. Averaging of real spectra was avoided because large quantities of voxel spectra (4664664) cannot be reliably phasecorrected before summation. Eddy currents corrections were unnecessary. Unsuppressed water line (not shown) as well as the most intense lipid lines did not reveal signs of eddy current distortions.

Results
The proposed 2D PRESS EPSI technique and data processing provided spectroscopic fat images (Fig. 3e) with acceptable spatial resolution and magnitude spectra with a very good signal-to-noise ratio (SNR) in all subjects. Figure 3a shows the profile of the usual 2D k-space weighting functions: standard Hanning, Hamming and optimized Hanning [24]. Optimized and standard 2D Hanning filters are shown in Fig. 3b and c, respectively. Figure 3d represents the fat image computed from the data set without application of k-space filter. From Fig. 3d it follows that the signal bleeding is a non-negligible problem although k-space sampling was increased to 64664. Corresponding fat images of filtered data sets, using an optimized and standard 2D Hanning filters, are shown in Fig. 3e and f, respectively. Figures 3e, f demonstrate that k-filters reduce inter-voxel signal contamination, but decrease  spatial resolution. In this study, a 2D optimized Hanning filter (Fig. 3b) was applied as a best compromise between signal bleeding suppression, spectrum quality and spatial resolution [24]. This filter increased the nominal voxel size by a factor 1.71 in both spatial directions [24]. Our ''true'' voxel shape could be best approximated as a cylinder with a diameter of 5.13 mm and height 15 mm (volume 0.31 cm 3 ). From comparison of Fig. 3e and f, it is evident that standard Hanning filter produces significantly larger true voxel size than optimized Hanning. The representative spectra of a normal volunteer are shown in Fig. 5 and 6. The most pronounced lines are residual water (4.7 ppm), singlet of creatine methylene protons (3.9 ppm), TMA (3.2 ppm), total creatine (tCr) (3 ppm), broader EMCL CH2 line at ,1.5 ppm, and IMCL CH2 at 1.3 ppm. Figure 5 demonstrates the impact of k-space filters on the spectrum of soleus muscle. The spectrum computed without application of k-space filter (Fig. 5a) reveals increased EMCL CH2 line at ,1.5 ppm and a little shoulder at ,1.8 ppm. This is due to the signal bleeding from the neighboring voxels to the selected VOI. Signal bleeding was suppressed by optimized (Fig. 5b), as well as by standard Hanning filter (Fig. 5c), as follows from missing shoulder at ,1.8 ppm and reduced EMCL CH2 line. From a comparison of Fig. 5b and c, it follows that optimized Hanning filter (Fig. 3b) sufficiently suppressed inter-voxel signal contamination because significant sharpening of k-space filter, by standard Hanning (Fig. 3c), did not lead to a further decrease in the EMCL CH2 line. In fact, EMCL CH2 intensity increased slightly (Fig. 5c) due to larger true voxel (VOI) size. Figure 6 shows the reference fat spectrum computed from the central tibial bone marrow and muscles in the tibialis anterior,

Discussion
Slice selective EPSI of the human brain was performed in the majority of previous studies [20,22,23,[29][30][31][32]. Maximum size of spatial matrices was 32632 and 32632616 in 2D and 3D experiments, respectively. Subcutaneous fat suppression was achieved by outer volume suppression (OVS) slices and/or by a broadband inversion pulse. Acquisition times varied between ,8 and ,30 minutes. With the recent improvements in gradient hardware performance, it was demonstrated, that spectral quality (linewidths) of brain spectra and SNR per unit time and unit volume were similar to those obtained with conventional phaseencoded PRESS MRSI techniques [23]. Low spectral intensities of brain metabolites hindered utilizing the speed of EPSI techniques. It should be noted, that readout bandwidth (echo sampling frequency), instead of spectral bandwidth, is the determining factor for SNR of EPSI techniques [33]. This leads to unavoidable loss in SNR compared to conventional MRSI with pure phase-encoding of signals and readout in the absence of imaging gradients.
EPSI techniques are, however, useful in the applications where the penalty paid in SNR loss is an acceptable trade for the considerable acquisition time reduction. Such applications are spectroscopy of lipids in skeletal muscle, liver, heart, etc. The net measurement time of our 2D PRESS EPSI sequence with two interleaves of gradient echo train was 12 min 48 sec for watersuppressed scans (4 acquisitions). However, the spectra of the bone marrow and muscles in Fig. 6 reveal SNR 114, and $28, respectively. SNR was computed by LCModel. The measurement time of water-suppressed scans could therefore be reduced by factor 2 (2 acquisitions). For comparison, minimum measurement time needed for conventional 2D MRSI (TR 1500 ms, spectral matrix 64664, 1 acquisition) is 102 min 24 sec for non-watersuppressed and the same time for water-suppressed scans.
Our module spectra have sufficient spectral resolution to distinguish TMA from tCr, IMCL CH2 from EMCL CH2 , and to detect EMCL CH2 splitting. TMA and tCr spectral intensities reveal differences between muscles (Fig. 6, 8, 9) and dipolar coupling effects, in agreement with previous studies [34,35]. Both spectral lines are highest in soleus muscle and lowest in gastrocnemius medialis and tibialis anterior. Sensitivity of our  EPSI sequence was not sufficient to detect TMA and tCr lines in all muscles due to small voxel size and a short acquisition time. TMA and tCr intensities were at the level of the noise in gastrocnemius medialis and tibialis anterior muscles, as follows from the individual voxel spectra shown in Fig. 4. Summation of such voxel spectra cannot improve the detectability of TMA and tCr.
Broad EMCL CH2 and narrow IMCL CH2 lines dominate in typical spectra of m. soleus (Fig. 5, 6). EMCL CH2 originates from extracellular fatty strands located between muscle fibers. It was suggested that these strands can be modeled by ''infinite long'' cylinders with the long axis at an angle h relative to the direction of the static magnetic field B 0 [5,6]. According to this model, the EMCL CH2 shift ranges from 0.2 ppm to 20.1 ppm relative to IMCL CH2 corresponding to a change in orientation 0 o # h #90 o [5,6]. EMCL CH2 position and linewidth are determined by the dominant angle h and by a degree of dispersion in orientations of fatty strands within the volume of interest, respectively [36]. The model of long cylinders agrees well with the spectra obtained from the large volumes of interest (Fig. 5, 6). However, this model is unable to explain spectra originating from small VOIs shown in Fig. 8 and 9. EMCL CH2 signals are split into more well resolved peaks. Moreover, EMCL CH2 shifts are increased up to 0.4-0.5 ppm relative to IMCL CH2 . These results are in agreement with the theoretical simulation and experimental evidence published by Steidle et al [7]. The authors found that EMCL compartments in soleus muscle experienced magnetic field deviations in the interval from ca 0.6 to 20.2 ppm in respect of IMCL CH2 . It should be also noted that EMCL CH2 spectral line splitting and shifts up to ,0.3 ppm relative to IMCL CH2 can be seen in the spectra of Hwang et al [12]. However, the authors did not point out or discuss these spectral features.
It should be emphasized that EMCL CH2 spectral line splits and their shifts up to 0.4-0.5 ppm relative to IMCL CH2 cannot be explained by signal bleeding from regions with strong lipid signals, in combination with B 0 shifts in the voxels. High effectiveness of signal bleeding suppression using Hanning filters was demonstrated theoretically as well as experimentally in the previous studies [24,37]. In addition, the significance of signal bleeding artefacts decreases with increasing size of acquisition matrix. In our case was signal bleeding negligible due to combination of large acquisition matrix 64664 and application of the Hanning k-filter. Figures 3e, f, and spectra in Fig. 4, 5, 8, 9 demonstrate high effectiveness of the Hanning k-filters in signal bleeding suppression. Figure 4 shows BMS shifts of the EMCL CH2 lines in more detail. This phenomenon is best seen in the voxel spectra of the soleus muscle. While TMA, tCr, and IMCL CH2 intensities are in the correct positions (in line), EMCL CH2 peaks at the same voxels show splitting, dispersion and shifts in vertical (ppm) direction. Some EMCL CH2 sub-peaks can be seen between 0.4 and 0.5 ppm from the correspondent IMCL CH2 line (arrows in Fig. 4). This finding reveals that BMS effects are more complex than previously demonstrated. Spectra measured from the large VOIs (.1 cm 3 ) reveal EMCL CH2 shifts up to 0.2 ppm. In such large volumes dominates long plate EMCL structures which resemble ''infinite  long'' cylinders. However, very small and isolated EMCL clusters could exist in the lean muscle. These spatial structures do not resemble infinite long cylinders. We suppose that another kind of BMS effects could occur on the discontinuities or interfaces of such fat clusters.
Quantification of IMCL content was restricted to m. soleus because this muscle group was studied in many previous studies, and because of its center position between the flexible receiver coils, i.e. in the most homogeneous region of the receiver coils sensitivity. Mean IMCL concentration of our six normal volunteers was, 0.3060.10 vol% or 3.661.2 mmol/kg, wet weight. This result agrees well with published values [12,15,16]. Despite good agreement, our result has to be taken with some caution because spectral intensities were not corrected for spatial sensitivity of the receiver coils. Measurements of coils sensitivities are challenging. In our case, this is because the mutual position of the flexible coils is different in each measurement session and the water/fat intensity ratio is different in each voxel. This problem could be circumvented by using a quadrature receiver knee coil. Such coil was not available in our laboratory at the time of this study.
To our knowledge, this is the first application of the EPSI technique in quantification of muscle lipids. PRESS localization was used only for slice selection. We were not forced to select PRESS box inside the slice because of negligible signal bleeding from subcutaneous fat and bone marrow. It enabled selection the bone marrow spectrum as the internal concentration reference. Combination of EPSI and PRESS VOI selection is relatively unusual. Only three research groups used the PRESS EPSI approach previously [38][39][40][41]. PRESS excitation improves VOI definition, reduces risk of signal contamination from outside the VOI, enables omitting inversion fat suppression pre-pulse and reduces the number of OVS slabs or even avoids them in spectroscopy of the brain [39,41], prostate [40], heart [42], etc. More complicated data processing of EPSI techniques is in our application counterbalanced by high spatial resolution and by significantly shorter measurement time compared to conventional MRSI. The high spatial resolution enables noncontiguous and irregularly shaped VOIs (Fig. 5, 6). It was possible to improve the reliability of the muscle lipid estimations by avoiding the influence of relaxation effects because the fat (bone marrow) spectral line was used as the internal concentration reference.

Conclusion
The EPSI sequence with PRESS volume of interest localization was developed and implemented in a 3T scanner. The use of a large spectral matrix 64664, together with the optimized Hanning k-filter significantly reduced inter-voxel signal contamination of the strong EMCL signals from subcutaneous fat, bone marrow or layers of adipose tissues. The small voxels enabled visualisation of EMCL methylene (-CH 2 -)n spectral lines splitting and their BMS shifts up to 0.5 ppm relative to correspondent IMCL methylene line. This study demonstrates that high-spatial resolution PRESS EPSI of the muscle lipids is feasible on clinical scanners. Figure 8. Magnitude spectra computed from the VOIs V1-V5 (four voxels). Left panel represents spectra computed without application of k-space filter during data processing. Middle and right panel depict spectra computed from filtered data set using an optimized and standard 2D Hanning k-space filter, respectively. Figure 6 shows VOIs. doi:10.1371/journal.pone.0087533.g008