|Home | About | Journals | Submit | Contact Us | Français|
Proton Magnetic Resonance Spectroscopic Imaging (1H MRSI) at 7T offers many advantages, including increased SNR and spectral resolution. However, technical di culties associated with operating at high fields such as increased B1 and B0 inhomogeneity, severe chemical shift localization (CSL) error, and converging T1 values, make the suppression of the broad lipid peaks which can obscure targeted metabolite signals, particularly challenging. Conventional Short Tau Inversion Recovery (STIR) can successfully suppress fat without restricting the selected volume, but only with significant metabolite signal loss. In this work, we have designed two new pulses for frequency-selective inversion recovery that achieve B1-insensitive fat suppression without degrading signal from the major metabolites of interest. The first is a spectrally-selective adiabatic pulse to be used in a volumetric 1H MRSI sequence and the second is a spatial-spectral (SPSP) adiabatic pulse geared toward multi-slice 1H MRSI. Partial interior volume selection may be used in addition to the pulses, to exclude areas with severe B0 inhomogeneity. Some differences in the spectral profile as well as degree of suppression make each pulse valuable for different applications. 7T phantom and in vivo data show that both pulses significantly suppress fat, while leaving most of the metabolite signal intact.
1H MRSI is a valuable technique for measuring metabolite levels in the brain in vivo . However, in order to obtain useful metabolite spectra, the suppression of broad, overlapping lipid resonances is usually necessary. 1H MRSI at higher fields, such as 7T, offers the advantages of increased SNR and spectral separation, but also suffers from increased B1 inhomogeneity, B0 inhomogeneity, and chemical shift localization (CSL) errors. Any lipid suppression method used for 1H MRSI at 7T must be robust in the presence of these problems. A commonly used method for fat suppression is to use a Position Resolved Spectroscopy (PRESS) sequence  to limit the selected volume to the interior of the head so that the subcutaneous fat near the skull is not excited. The usable volume is further reduced when edge voxels must be discarded due to CSL errors, which scale with field. As a result, spatial coverage is significantly restricted using this method. B1-insensitive outer volume suppression techniques such as BISTRO  may also be used to eliminate signal from subcutaneous fat. However, the selected volume is dependent on the placement of the BISTRO suppression bands and is therefore still restricted. An alternative, commonly used fat suppression method that does not restrict the selected volume is Short Tau Inversion Recovery (STIR) [4,5]. STIR employs a 180° inversion pulse prior to the 90° excitation pulse in a PRESS sequence with inversion time (TI) adjusted such that the longitudinal magnetization of short T1 species (e.g. fat) passes through a null when primary excitation occurs. Although most metabolites have T1’s significantly longer than fat, they nonetheless suffer from a partial signal loss on the order of 30%-40% at 1.5T and 3T . Due to the convergence of T1 values for di erent species at higher fields [6, 7], this signal loss is closer to 50% at very high fields such as 7T.
Fortunately, the increased spectral separation at 7T makes it possible to design a spectrally-selective adiabatic inversion pulse that only inverts lipid frequencies, leaving most of the metabolites untouched. Two versions of such a pulse have been designed. The first is a spectrally-selective adiabatic pulse that provides B1-insensitive inversion of the lipid resonances in the entire volume prior to slice selection. The second is a spatial-spectral (SPSP) adiabatic pulse that inverts the fat in just the selected slice. The adiabatic SPSP pulse allows simultaneous selectivity in both space and frequency while maintaining insensitivity to B1 inhomogeneities . The spectral selectivity allows selective fat inversion while the spatial selectivity makes the pulse amenable to multi-slice sequences. At 7T, multislice MRSI is particularly attractive, as compared to volumetric methods, due to the ability to individually optimize the B0 homogeneity of each slice using dynamic shimming [9, 10]. In this work, designs for both of these pulses are presented and integrated into a standard PRESS sequence.
First the spectrally-selective adiabatic inversion pulse and the SPSP adiabatic inversion pulse were designed. The two pulses were then inserted into the preparatory pulse section of a standard PRESS sequence, resulting in two versions of a modified PRESS sequence: Version 1, utilizes the spectrally-selective adiabatic inversion pulse and, Version 2, utilizes the SPSP adiabatic inversion pulse for fat suppression.
where the maximum B1 field A0 = 4.3 μT, the modulation angular frequency β = 300 rad/s, the bandwidth determining dimensionless parameter μ = 6, and the pulse duration T = 30 ms.
The resultant pulse had a spectral bandwidth of 573 Hz, which was large enough to invert the full range of lipid frequencies plus an additional margin to accommodate peak shifts anticipated in vivo due to B0 inhomogeneity. The pulse duration was set to 30 ms in order to achieve a spectral profile with transition bands narrow enough to selectively invert lipids. At 7T, NAA (2.0 ppm) and lipids (1.3 ppm) are separated by 210 Hz. The final pulse we designed has a transition width of approx 110 Hz. Thus, with this pulse design, is possible to selectively and adiabatically invert fat without affecting NAA. Because the pulse is not spatially selective, it is not subject to CSL errors. Figure 1 A shows the real and imaginary components of the adiabatic spectrally-selective 180° inversion pulse.
Simulations were performed in MATLAB (The Mathworks, Natick, MA, USA) by explicitly multiplying the rotation matrices to get the final spectral profile of the pulse. T1 and T2 effects were neglected. Figure 1 B is the simulated spectral profile for the inversion pulse showing the location of the NAA and lipid resonances relative to the inversion band. Fig. 2, the simulated spectral profile is shown for a range of B1 overdrive factors above adiabatic threshold. If the nominal B1 is set to be at the adiabatic threshold, the pulse may driven at an overdrive factor of 3.9 before reaching the 17 μT RF peak amplitude limit our 7T RF amplifier. As shown in Fig. 2 the spectral profile stays invariant over a 290% increase in B1, at which point the RF amplifier limit is reached. This is more than sufficient invert lipids in the presence of the 50% B1 variations measured across the human brain 7T using our standard GE volume head coil.
A sech/tanh pulse with a bandwidth of 382 Hz and duration of 37.8 ms was designed using Eqns. 1 and 2 for the amplitude and frequency modulation functions respectively. For this pulse, A0 = 6.5 μT, β = 200 rad/s and μ = 6. The spectral inversion bandwidth of 382 was sufficient to invert the full range of lipid frequencies and the transition bandwidthof 100 Hz was small enough to avoid inverting NAA. The pulse was then sampled with 75 sample points, achieving an optimal trade off between sideband distance and minimum slice thickness. The spectral sidebands had to be placed at a sufficient distance away from the main inversion band such that the most prominent brain metabolites (i.e Cho at 3.2 ppm, Cr at 3.0 ppm and NAA at 2.0 ppm) were not inverted. The final adiabatic SPSP pulse was comprised of 75 small tip-angle subpulses scaled by the sampled values of the adiabatic sech/tanh envelope. The subpulses were a least squares Fourier design with a time-bandwidth product of 2 generated using the firls function in MATLAB. The resultant separation between the main inversion band and sidebands was ±2 kHz. The opposing sidebands were located at ±1 kHz. This separation was large enough to prevent inversion of metabolites between Cho and NAA. Due to the high spatial bandwidth of 4 kHz achieved by the short duration spatial subpulses, the CSL error for this pulse is minimal. Figure 3 A shows real and imaginary components of the adiabatic SPSP 180° inversion pulse. The simulated spectral profile for the inversion pulse showing the main inversion band and sidebands at ±2 kHz is plotted in Fig. 3 B. Figure 3 C is a zoomed in view of the spectral profile showing the main inversion band more clearly. Simultaneous spatial and spectral selectivity is shown in the simulated 2D spatial-spectral profile, Fig. 3 D. Horizontal cross-sections of this 2D SPSP profile show frequency selectivity for a particular spatial location and vertical cross-sections show the selected slice for a particular frequency. Within the main inversion band, the selected slice stays largely invariant, demonstrating the chemical shift immunity of the pulse. Sidebands at ±2 kHz and opposing sidebands at ±1 kHz are visible in the 2D SPSP profile. Opposing sidebands exist in the inversion profile because a SPSP pulse does not trace out a rectilinear excitation k-space trajectory. The even and odd lines of the trajectory are slightly tilted which results in imperfect cancellation of the sidebands at ±1 kHz .
The adiabaticity of the spatial and spectral magnetization profiles of the pulse was verified through simulations. In Fig. 4 A, the simulated spatial profile is shown for a range of B1 overdrive factors above adiabatic threshold. The pulse may be overdriven by 150% (i.e. an overdrive factor of 2.5) before reaching the limit for our 7T RF amplifier. An increase in spatial selectivity with increasing B1 is noticeable in Fig. 4 A. Thus, in regions of higher B1, a sharper slice will be selected by the SPSP inversion pulse. This variation in slice selectivity may be partially accounted for by inverting a slightly thicker slice than that selected by the subsequent PRESS excitation. Figure 4 B shows the main spectral passband of the pulse over the same range of B1 overdrive factors. The spectral profile stays invariant over a 150% increase in B1, at which point the RF amplifier limit is reached. Beyond an overdrive factor of 3.4, there is a slight increase in inversion band ripple.
The final spectrally-selective adiabatic inversion pulse and SPSP adiabatic inversion pulses were both integrated into a standard PRESS sequence directly before the CHESS water suppression pulses. A PRESS excitation was still used to provide the option of partial interior volume selection. This may be necessary to exclude lipids from regions of severe B0 inhomogeneity (e.g., near the skull and in the neck), which are shifted out of the inversion band of the pulses. Optimum lipid suppression was achieved when the inversion pulses were placed 300 ms before the 90° pulse (i.e. TI = 300 ms). Figure 5 A is a diagram of the modified PRESS sequence with fat-selective inversion recovery included. Figures 5 B and C show the waveforms for the spectral and spatial-spectral versions of the adiabatic inversion pulses, inserted into version 1 and 2 of the sequence respectively.
Phantom experiments were conducted to verify pulse performance. To test pulse selectivity, single-voxel spectra were obtained from the GE MRS sphere phantom using the two versions of the sequence in Fig. 5 A. The first version uses the waveforms in Fig. 5 B and the second uses the waveforms in Fig. 5 C for fat suppression. The spectral inversion pulse was overdriven by 180% while the SPSP inversion pulse was overdriven by 80%. The results were compared to those obtained using a conventional PRESS sequence as well as a conventional PRESS sequence with standard non-selective STIR. To test the degree of fat suppression achieved by the pulses, the same experiment was repeated for a phantom containing water and canola oil. All pulse parameters, including overdrive factor and spectral offset where kept constant between the two phantom experiments. The scans were performed on a 7T scanner (Echospeed whole-body magnet; GE Healthcare, Waukesha, WI, USA) using a standard GE volume head coil. Reconstruction was performed using the Probe Single Voxel Reconstruction on SAGE/IDL Dev2005.1 (GE Medical Systems). The scale for each phantom experiment was adjusted such that the noise, as determined from the spectral region without peaks, was equivalent for all spectra. The acquisition parameters for all 1H MRS scans were: 3.4 cc voxel, TI: 300 ms, TE/TR: 90/2000 ms, 4 NEX, 1:44 min scan time.
The inversion pulses were tested in vivo by exciting a single slice through the brain of a normal volunteer using the two versions of the sequence in Fig. 5 A. The spectral offset (i.e. transmit frequency) of the inversion pulses was shifted until maximum suppression of lipid peaks was achieved without a reduction in the NAA signal. As in the the phantom experiments, the spectral inversion pulse was overdriven by 180% and the SPSP inversion pulse was overdriven by 80%. The results were compared to those obtained using a conventional PRESS sequence. The scans were performed on our 7T scanner using a standard GE volume head coil. The acquisition parameters for the 1H MRSI scan were: slice thickness = 1.4 cm, FOV = 20×20 cm, matrix size = 12×12 (7×9 voxels within the PRESS box), voxel volume = 3.9 cc, TE/TR = 80/3000 ms, NEX = 1 and scan time = 7:24 min. B1 and B0 maps of the imaged slice were also obtained. Reconstruction was performed using the CSI Reconstruction on SAGE/IDL Dev2005.1. The scale was adjusted such that the noise, as determined from the spectral region without peaks, was equivalent for all data sets.
Results of the phantom experiments are shown in Fig. 6. Single-voxel spectra obtained from the GE MRS phantom using a conventional PRESS sequence and the conventional PRESS with non-selective STIR can be seen in Figs. 6 A and B respectively. As expected, the use of the STIR pulse significantly reduces the metabolite signal. It should be noted that this signal reduction is higher than that seen in vivo because the T1’s of metabolites in the GE MRS Sphere phantom are generally shorter than those in the brain. When the same voxel is excited using version 1 and 2 of the PRESS sequence with fat-selective inversion recovery, the spectral grids in Fig. 6 C and D are obtained. Signal from the most prominent singlet metabolite resonances in the brain (i.e Cho at 3.2 ppm, Cr at 3.0 ppm and NAA at 2.0 ppm) are unaffected by either inversion pulse. However, some signal from metabolites resonating between water (4.7 ppm) and Cho (3.2 ppm) is lost when using the SPSP adiabatic inversion pulse due to the partial inversion of these resonances by the opposing sidebands in the SPSP inversion profile (Fig. 3 D). This is evidenced by the diminished myo-inositol (Myo) peak at 4.0 ppm and Cr peak at 3.9 ppm in Fig. 6 D as compared to Figs. 6 A and C.
Figures 6 E and F are single-voxel spectra obtained from the fat/water phantom using the conventional PRESS sequence and the conventional PRESS with non-selective STIR respectively. Spectra from the same voxel obtained using version 1 and 2 of the PRESS sequence with fat-selective inversion recovery are shown in Figs. 6 G and H. The integral under the main lipid peak at 1.3 ppm was approximately 7.5 times smaller for the spectra in Figs. 6 G and H than the unsuppressed spectrum in Fig. 6 E. The combined results of the two phantom experiments verify that that both the spectral and SPSP pulses are sufficiently selective to exclude metabolites while providing significant suppression of major lipid resonances.
Figure 7 A shows the image of a single slice through the brain of a normal volunteer scanned at 7T. 1H MRSI data were obtained for this slice with a conventional PRESS sequence and versions 1 and 2 of the modified PRESS sequence with fat-selective inversion recovery. A large PRESS box was chosen with corners close to the edge of the skull, and the excited slice was placed near the top of the head, resulting in significant lipid contamination both in the center and edge voxels. In this way, the effectiveness of the fat suppression over a range of B1 values could be tested (B1 is highest at the isocenter of our head coil). The PRESS box and spectral grid location for all 1H MRSI experiments are shown on the image in Fig. 7 A. The measured B1 map for the same slice, acquired using the double-angle method [13, 14], can be seen in Fig. 7 B. A severe reduction in B1 from the center to the periphery of the brain is evident. The B0 map in Fig. 7 C was obtained after using first and second order shimming to optimize the B0 homogeneity.
The data obtained for the spectral grid location shown in Fig. 7 A, using a standard PRESS sequence, can be seen in Fig. 7 D. When the same region is excited using version 1 and 2 of the PRESS sequence with fat-selective inversion recovery, the spectral grids in Fig. 7 E and F are obtained respectively. In Fig. 7 E, lipids are inverted in the entire volume, but in Fig. 7 F lipids are only inverted within the excited slice. Figures 7 E and F show that significant fat suppression is achieved over the entire excited region for both the spectral and SPSP adiabatic inversion pulses. Hence, for this particular slice, the overdrive factor used for the SPSP inversion pulse, combined with it’s slightly higher selectivity were sufficient to generate results that are almost as good as those obtained using a spectral inversion pulse. On average there was approximately an 7-fold reduction in lipid signal in the peripheral voxels of the spectral grid. Greater lipid suppression was achieved for the anterior voxels (9-fold) when compared to the posterior voxels (6-fold). All spectra are plotted to the same vertical scale, chosen to emphasize the broad lipid resonances.
In Figs. 7 D-F, lipid signals are stronger on the posterior side of the brain mainly due to shifted lipid peaks from anterior voxels aliasing into the posterior voxels. Since linear-phase SLR pulses with limited spatial bandwidth are used for localization in the sequence in Fig. 5 A, the selected PRESS boxes for metabolites and lipids are significantly shifted with respect to each other at high fields. For the data in Fig. 7, a 90° excitation pulse was used to localize along the left-right (L-R) dimension and a 180° refocusing pulse was used to localize along the anterior-posterior (A-P) dimension. As a result, more severe CSL error occurred in the A-P dimension. The shifted lipid volume combined with a FOV that is only 17% greater than the PRESS box in the A-P dimension, resulted in aliasing of lipid signal from anterior voxels into posterior voxels. In addition to this effect, B0 and B1 inhomogeneity are most significant near the skull, further hampering the performance of the inversion pulses.
Figure 8 shows detailed spectra from four selected voxels from the data set depicted in Fig. 7. Spectra have been phased and scaled to show the Cho, Cr and NAA peaks more clearly. Figure 8 shows the location of the selected voxels on the water image. Figures 8 B, C and D are spectra for these voxels obtained using standard PRESS, version 1 of the modified PRESS sequence, and version 2 of the modified PRESS sequence, respectively. Both inversion pulses achieve significant fat suppression and consequently reduce the baseline distortion caused by the tail of the broad lipid resonance. The spectral pulse is slightly more successful than the SPSP pulse in this respect, especially for the edge voxels. The higher overdrive factor, and therefore a greater range of adiabaticity for the spectral pulse is the probable reason for this difference. The Lipid Suppression Factor (LSF) calculated by taking the ratio of the integrals under the unsuppressed and suppressed lipid peaks is labeled on the peripheral voxels in Fig 8 C and D. Signal from Cho at 3.2 ppm, Cr at 3.0 ppm and NAA at 2.0 ppm stays intact for both inversion pulses. However, there is partial loss of the Cr signal at 4.0 ppm in Fig. 8 D resulting from inversion caused by the opposing sidebands in the SPSP inversion profile.
Phantom and In vivo data demonstrate that both spectrally-selective and SPSP adiabatic inversion pulses successfully suppress fat while not degrading signal from the Cho, Cr and NAA peaks in the brain. Similar fat suppression may be achieved by using a non-selective adiabatic inversion pulse; however the lack of spectral selectivity with this approach results in approximately a 50% metabolite signal loss at 7T.
Both types of fat-selective inversion pulses presented in this work have their utility. The nature of the application will dictate which pulse is most advantageous. The spectral inversion pulse achieves a higher overdrive factor before reaching the RF peak amplitude limit for our 7T magnet and is therefore more robust to B1 inhomogeneity. The pulse also has a spectral profile that is exclusively lipid-selective, allowing full excitation of all metabolites resonances between water and NAA. Since the pulse is not spatially selective, CSL error is not a consideration. Lower overdrive factors are possible for the SPSP adiabatic inversion pulse than the spectral inversion pulse, resulting in less immunity to B1 variations. However, an advantage of using the SPSP pulse is that it provides spatial selectivity, making it possible to use a multi-slice approach for excitation where each slice may be shimmed individually . This is advantageous at high fields because B0 inhomogeneity scales with field dictating more effective shimming to prevent the shifting of the lipid peaks out of and NAA peak into the inversion passband. Individually optimizing the B0 for each slice is particularly useful for spectroscopic imaging of inferior brain regions where the B0 inhomogeneity can be severe. Furthermore, the use of a SPSP pulse results in high spatial bandwidth, making the selected slice highly immune to CSL errors as compared to a conventional slice-selection approach. Due to the opposing sidebands present at ±1 kHz in the spectral profile of the pulse shown in Fig. 3 D, partial inversion of metabolite peaks between water and Cho will occur; however, all other peaks including Cho, Cr (at 3.0 ppm) and NAA will experience no inversion. For both pulses, the degree of fat suppression may be traded of for robustness to B0 shifts. If the inversion band is placed closer to the NAA peak, increased suppression of lipid resonances in this range will be achieved. However, the chance of the NAA peak shifting into the inversion band due to B0 inhomogeneity also increases.
Because both inversion pulses have a long duration (30 ms for the spectral pulse and 37.8 ms for the SPSP pulse), there will be some degradation of the inversion profile for short-T2 lipid components. However, since the echo time for the sequence in Fig. 5 A is 80 ms, any residual short-T2 lipid signal will be strongly reduced in intensity. Thus, these inversion pulses are most suitable for echo times longer than 40 ms. Specific Absorption Rate (SAR) values were calculated for the two pulses and compared to the 10 ms, adiabatic, non-selective STIR pulse available with the conventional PRESS sequence. SAR for the conventional STIR pulse was 1.17 times the SAR of the spectrally selective adiabatic inversion pulse and 0.89 times the SAR for the SPSP adiabatic inversion pulse. In the case of the spectrally selective pulse, spreading the RF energy over a longer pulse duration reduces the overall SAR. Though the duration of the the SPSP pulse is also long, increased RF amplitude due to the inclusion of subpulses causes an overall increase in SAR.
Both pulses will allow for increased volumetric coverage without degraded metabolite signal for 1H MRSI at 7T. The pulses are readily integrated into a standard PRESS sequence used for 1H MRSI, but may also be used for B1-insensitive fat suppression in high-field imaging sequences for which a TR > TI is tolerable. For imaging applications, the duration of both pulses can be shortened, as the requirement for spectral selectivity is relaxed.
Future work is focused on finding the optimal amplitude and frequency modulation functions for the inversion pulses such that minimum transition widths can be achieved. Other options such as adiabatic inversion pulses with asymmetrical profiles [15, 16] or selective adiabatic excitation pulses  will also be explored and compared in terms of selectivity, suppression efficiency and Specific Absorption Rate (SAR) levels.
This work was supported by NIH-RR09784 and The Lucas Foundation.