Spatial effects in the detection of γ-aminobutyric acid: Improved sensitivity at high fields using inner volume saturation


  • Richard A.E. Edden,

    1. Russell H. Morgan Department of Radiology and Radiological Science, Johns Hopkins University School of Medicine, Baltimore, Maryland, USA
    2. F.M. Kirby Research Center for Functional Brain Imaging, Kennedy Krieger Institute, Baltimore, Maryland, USA
    Search for more papers by this author
  • Peter B. Barker

    Corresponding author
    1. Russell H. Morgan Department of Radiology and Radiological Science, Johns Hopkins University School of Medicine, Baltimore, Maryland, USA
    2. F.M. Kirby Research Center for Functional Brain Imaging, Kennedy Krieger Institute, Baltimore, Maryland, USA
    • Department of Radiology, MRI 143C, Johns Hopkins University School of Medicine, 600 N Wolfe Street, Baltimore, MD 21287
    Search for more papers by this author


The MEGA-PRESS-IVS method has been developed, which combines MEGA (a frequency-selective editing technique) editing with the point-resolved spectroscopy sequence (PRESS) and inner volume saturation (IVS) localization, reducing the deleterious effects of spatial variation in coupling evolution. The IVS method has been previously described for improved efficiency of lactate detection. The current study demonstrates that the combination of MEGA-PRESS with IVS results in increased sensitivity for edited single-voxel measurements of glutamate and γ-aminobutyric acid (GABA). A four-compartment model of coupling evolution is investigated through simple product operators and full spin-system simulations and the predicted pattern of signal evolution is demonstrated through MEGA-PRESS-MRSI. MEGA-PRESS-IVS is then compared to MEGA-PRESS in a phantom and an average signal increase of 24% is demonstrated in five healthy volunteers. Magn Reson Med, 2007. © 2007 Wiley-Liss, Inc.

The in vivo proton (1H) spectrum of the human brain contains many metabolite signals that can be interpreted in terms of tissue function and cellular composition (1). Of the detectable signals, several are due to substances that are known to act as neurotransmitters, such as glutamate and γ-aminobutyric acid (GABA). Variations in GABA levels in the brain have been reported in epilepsy (2), panic disorder, and neurological disorders, as well as in the brains of alcoholics (3) and puerperal women (4).

Detection of GABA in vivo is hampered by signal overlap and low signal intensity, due both to low concentration and splitting of the signals associated with scalar J-coupling. At moderate field strengths, the multiplets arising from GABA protons overlap at 3.0 ppm with signals due to creatine and phosphocreatine (collectively referred to as Cr), at 2.3 ppm with glutamine and glutamate (Glx) and the GABA multiplet at 1.9 ppm overlaps with signals from N-acetyl aspartate (NAA) and N-acetylaspartylglutamate (NAAG). Although spectral fitting routines, such as the LCModel (5), can be used to quantify the different components of overlapped spectra, more robust approaches to the quantitation of GABA concentration by MRS have involved spectral editing. A number of possible editing methods have been proposed that rely upon the coupled nature of the GABA spin system, some to produce double-quantum coherence (6–8) and some through J-difference editing (9–11). A frequency-selective editing technique (called MEGA) combined with the point-resolved spectroscopy sequence (PRESS) method (MEGA-PRESS) (10) allows the detection of the outer two peaks of the GABA triplet at 3.0 ppm, while editing out signal due to Cr and sacrificing the central peak of the triplet. It should be noted that in addition to providing spectral editing, MEGA-PRESS allows water suppression (10).

A weakness of the MEGA-PRESS sequence at high field is that, when the bandwidth of the slice-selective pulses becomes similar to the chemical shift difference of the coupled spin systems, the scalar-modulation patterns become spatially dependent, which can result in loss of editing efficiency. This effect is well known in the context of the use of the PRESS sequence for lactate detection, where substantial signal losses can occur (12, 13).

In this work, the evolution of coupling in the GABA spin system during the MEGA-PRESS experiment is considered, and it is shown that a significant loss in intensity of the detected signal occurs due to the spatial variation of this evolution. It is then demonstrated that the inner volume saturation (IVS) technique (previously described for lactate detection using the PRESS sequence (14)) can be also applied to the MEGA-PRESS experiment to significantly increase the intensity of the edited signals. In the current example (GABA on a 3T system with a maximum B1 field strength of 14 μT) the use of the IVS method improved sensitivity by an average of 24%; larger improvements may be realized at higher B0 field strengths, lower B1 fields, or for other coupled spin systems with larger chemical shift differences.


The MEGA-PRESS experiment (10), shown in Fig. 1, consists of two excitations: in one (when GABA is the editing target), the inversion pulses are applied to the H3 spins at 1.9 ppm, refocusing the scalar modulation of the H2 resonance at 3.01 ppm; in the other, the inversion pulses are either not applied, or are applied at some other frequency, so as to leave evolution of scalar coupling undisturbed. GABA consists of six coupled proton spins; their chemical shifts, as detailed by Govindarju et al. (15), are as follows: H2 and H2′ = 3.01 ppm; H3 and H3′ = 1.89 ppm; and H4 and H4′ = 2.28 ppm.

Figure 1.

Pulse program of the MEGA-PRESS experiment. The maximum RF field is 14 μT; the maximum gradient amplitude is 20 mT/m. The first echo time (TE1) is typically 15 ms and the second (TE2) is 53 ms. IVS is achieved through spatially-selective saturation before the MEGA-PRESS excitation (not shown).

The MEGA-PRESS experiment will be considered in the same manner as our previous treatment of the lactate spin system during the PRESS experiment (14). Because MEGA-editing for GABA detection relies upon either the refocusing, or full evolution of coupling, it is important to be aware of the evolution of coupling in all regions of the PRESS-excited volume for the H2 spins. Depending on the exact location within the volume, the passive spins may undergo both, the first only, the second only, or neither of the refocusing pulses (referred to as Cases I, II, III, and IV, respectively), which will significantly affect the form of the H2 multiplet.

Product operator calculations for a three-spin system that models the pertinent features of the GABA spin system are presented in the Appendix. These results can be characterized by an apparent coupling evolution time (which is the duration of coupling evolution that gives the same result as the product operators calculated for a given case). Case I demonstrates the intended effect of the experiment—that the coupling evolve for TE in the Off experiment and 0 in the On experiment. Since TE1 is usually short compared to TE2, the effect of coupling evolution for TE1 approximates to zero evolution and evolution for TE2 approximates to the full TE. Therefore Cases I and III evolve largely as required, while Cases II and IV evolve to a very similar extent in the On and Off experiment and result in a loss of signal upon subtraction.

However, since this approach is only an approximation of the behavior of the full GABA spin system, the results of the product operator calculations were verified using full numerical density matrix simulations of the six-spin system (NMRSIM; Bruker Biospin). 3T simulations were performed using the inclusion or omission of block refocusing pulses to approximate the four Cases (assuming that H3 and H2 spins undergo the same refocusing pulses in each Case). Idealized multiplets representing the calculated product operator terms (see Appendix) are shown alongside the numerical simulations in Fig. 2. The results show good agreement, suggesting that the product operator description of the three-spin model adequately describes the behavior of the GABA spin system at 3 T.

Figure 2.

Comparison of idealized multiplets based on the three-spin product-operator calculations (shown offset in gray) and simulations of the full GABA spin system (H2, H2′ multiplet shown in black). Calculations were carried out for all four cases (passive spins undergoing both, the first, second or neither refocusing pulse), both with and without the editing pulses. Agreement is strong, supporting the applicability of the three-spin, weakly-coupled model.

At 3T, using the body coil for transmit, the field is limited to 14 μT in our system, resulting in sinc-gauss refocusing pulses with a bandwidth (width at half-height) of 660 Hz, as compared to the 143 Hz chemical shift difference between the H2 and H3 protons. The PRESS-excited volume for the two spins is therefore displaced linearly by 21% of its width in the two dimensions defined by refocusing pulses and the same fraction of the volume corresponds to Case II or IV (which are predicted to result in signal loss). These spatial effects can be demonstrated experimentally using PRESS-MRSI.

Because the phase of the outer peaks differs in Cases II, III, and IV from that in Case I, this spatial variation will result in a loss of signal intensity from the MEGA-PRESS experiment. Application of the PRESS-IVS method can reduce this loss by saturating regions with unwanted modulation patterns. By placing the transmitter frequency midway between the H3 and H2 chemical shifts (at 2.45 ppm), the PRESS-excited volumes for the H3 and H2 spins will be symmetrically disposed about the prescribed volume, and, more importantly, the volume corresponding to Case I, in which the passive spins undergo both refocusing pulses, lies centrally within the prescribed volume. It is therefore possible to enlarge the PRESS-excited volume, and then trim it using high-bandwidth IVS pulses, symmetrically disposed around (14), yet overlapping with, the PRESS-excited volume, as described previously. The amount by which the volume must be enlarged before trimming is principally determined by the size of the fractional chemical shift displacement δ between the H2- and H3-excited regions, and is equal to a minimum of 1/(1 – δ). The enlargement should be sufficient that the subvolume in which the evolution of coupling is well-behaved, should have the same volume as the intended volume of acquisition (14) and the width of the saturation bands should be sufficient to saturate all regions with unwanted modulation, but not significantly more so as to keep the transition edge as steep as possible. Because of the discrepancy between the prescribed and actual band of action of the saturation pulses, it is necessary to calibrate the trimming to ensure that the same volume has been excited in each case. This calibration, which needs only be performed once upon initial implementation of the IVS method, can be done in a phantom by using the sum of the MEGA-PRESS experiments (which contains the unedited information) to compare the MEGA-PRESS excited volume with the MEGA-PRESS-IVS excited volume.


Single-voxel MEGA-PRESS spectroscopy of a phantom containing 140 mM GABA was performed on an Intera 3T systems (Philips Medical Systems, Best, The Netherlands) with a receive-only six-channel head coil and using the body coil for transmit; the maximum transmit B1 field of the body coil is 14 μT (∼600 Hz). Spectra were acquired at an echo time of 68 ms (TE1 = 15 ms, TE2 = 53 ms) and a recycle time of 2000 ms. Refocusing in the PRESS sequence was achieved with a three-lobe sinc-gauss pulse, of length = 6 ms and bandwidth = 600 Hz. Coherence pathway selection was performed using trapezoidal gradients of maximum amplitude 20 mT/m and length = 3 ms. Chemical shift selective (CHESS) water suppression was optimized before each acquisition.

Editing in the MEGA-PRESS experiment was achieved using two Gaussian inversion pulses of length = 14 ms; use of longer pulses is prevented by the 68 ms echo time. The two experiments of the MEGA editing sequence were performed in an interleaved fashion, one with the transmitter frequency at 1.9 ppm and one at 7.6 ppm (to be symmetrical about the water frequency).

For the control MEGA-PRESS acquisition, a 3 × 3 × 3 cm3 voxel was excited; then, in a second experiment, the voxel size was enlarged to 4.5 × 4.5 × 3 cm3 and four IVS pulses (30-mm thickness) were applied to saturate regions of unwanted modulation, and to reduce the effective voxel size back to 3 × 3 × 3 cm3. The saturation pulses were applied in the 35 ms directly before the MEGA-PRESS excitation, with a 4.3-kHz bandwidth and a duration = 2 ms (each followed by a 3 ms crusher gradient). To offset the effects of T1 relaxation, the pulses were applied with flip angles of 95°. Each experiment was also performed without water suppression, so that the acquired data could be normalized to account for any difference in the volume excited.

To demonstrate the spatially dependent modulation effects in the MEGA-PRESS sequence, a MEGA-PRESS-MRSI experiment of a phantom containing 200 mM GABA was performed, using a 4-inch × 6-inch surface coil and the body coil for transmit. Data were acquired over a field of view = 80 mm × 80 mm with a slice thickness = 10 mm, defined by the slice-selective excitation pulse. In the direction defined by slice-selective refocusing pulses, the prescribed volume was 60 mm × 60 mm and two-dimensional phase-encoding was performed in this plane. MRSI data were acquired using 28 phase-encoding gradients in both dimensions with a 2 s recycle time, giving a total experiment time of 26 min. Separate MRSI datasets were acquired with and without the editing pulses applied at 1.9 ppm.

Prior to Fourier transformation, MRSI data were processed in the time domain with 3-Hz line broadening and eightfold zero-filling and in the spatial dimensions with a half-cosine window function and no zero-filling. Integration of the side peak of the H2 triplet (from 3.10 ppm to 3.05 ppm) was performed and this data was interpolated by a factor of eight to generate images of the spatial variation of the signal.

The MEGA-PRESS-IVS method was demonstrated in vivo, on five healthy volunteers (age 31 ± 4 years; three female). The study was approved by the local Institutional Review Board and informed consent was obtained. In the MEGA-PRESS (control) experiment, a 3 × 3 × 3 cm3 volume was prescribed centrally in the posterior white matter. In the MEGA-PRESS-IVS experiment, a 4.5 × 4.5 × 3 cm3 volume was trimmed to 3 × 3 × 3 cm3 by four saturation pulses of thickness 15 mm. A total of 512 transients were acquired (using the head coil) for both the control and the IVS experiment in an experiment time of 17 min.


Appropriate choice of the IVS parameters, such as the degree of voxel enlargement and the appropriate overlap of the saturation pulses, relies upon the accuracy of the four-compartment model described above. To support the model, and to demonstrate that the spatial effects are significant at 3T, MRSI of the MEGA-PRESS excited voxel was performed. Images of the integral of the downfield peak of the H2 triplet are shown in Fig. 3: Fig. 3a shows the result with editing pulses at 7.6 ppm; Fig. 3c shows the result with editing pulses at 1.9 ppm; and Fig. 3e shows the difference image. In each case, a schematic diagram of the predicted image is shown, containing the predicted spectrum for each of the four compartments. This experiment appears to suggest that a 20% to 25% loss in signal occurs due to spatial effects; signal from two compartments (those corresponding to Case II and Case IV) is lost in the difference spectrum. Given that the signal loss is 20% to 25%, the improvement that can be achieved by IVS is 25% to 33%.

Figure 3.

Experimental demonstration of the regional modulation patterns predicted in Fig. 2. 2D PRESS-MRSI data was collected using 28 phase-encoding gradients in each dimension with a TR = 2.0 in a total acquisition time of 26 min. The total field of view = 80 mm × 80 mm and the PRESS-excited region = 60 mm × 60 mm. Images represent the integral (from 3.10 ppm to 3.05 ppm) of the left-hand peak of the H4 triplet, with the editing pulses off (a) and with the editing pulses on (c). The difference between these (corresponding to the MEGA-PRESS experiment) is shown in (e). Corresponding schematics, based on what would be predicted by the four-compartment model, are shown in (b), (d), and (e); the triplet lineshape in each compartment is also shown. The spatial compartments corresponding to Cases I–IV have been labeled on the schematics.

Figure 4 compares the MEGA-PRESS and MEGA-PRESS-IVS spectra of the GABA phantom. The sum spectra (which contain the uncoupled information) have signals of very similar intensity, demonstrating that the same volume has been excited in both cases. There is a 28% improvement in the integral of the edited difference signal, which is consistent with the expected SNR improvement based on the MRSI experiment.

Figure 4.

MEGA-PRESS-IVS (and control MEGA-PRESS) difference spectra of the GABA phantom. MEGA-PRESS-IVS results in a 28% improvement in the edited GABA signal. The sum spectra demonstrate that this arises from improved coupling behavior, rather than a volume increase.

Figure 5 shows spectra acquired in one volunteer, comparing those acquired with (in black) and without IVS (in gray). The two spectra were processed in an identical manner, with eight-fold zero-filling, 5-Hz Lorentzian line broadening, identical phase parameters, and no baseline correction. This example is typical of the group, in which an average signal integral increase of 24 ± 6% resulted from IVS.

Figure 5.

MEGA-PRESS-IVS (and control MEGA-PRESS) spectrum acquired in a healthy 22-year-old volunteer. a: Shows the volume from which signal is acquired, overlaid on a sagittal and coronal T1-weighted image. b: Compares the MEGA-PRESS-IVS difference (in black) and MEGA-PRESS (in gray) spectra. The sum spectra are included to demonstrate that the same volume was excited in each case and are plotted on a different vertical scale from the difference spectra.


This study shows that, at high fields, spatial effects may play a role in reducing the sensitivity of MEGA-PRESS measurements of GABA, and that a significant increase in signal can be achieved through the IVS method, averaging to 24% over five healthy volunteers, similar to the predicted 25% to 33% improvement. Although the improvement is not as dramatic as in the lactate case (14), due to the smaller chemical shift differences in the GABA spin system, it is important to maximize available signal, especially since the detection of low concentration metabolite signals in vivo by the MEGA-PRESS method results in spectra with low signal-to-noise ratios. Furthermore, as in vivo spectroscopy is developed at higher static field strengths, with increases in B1 field strength unlikely, spatial variation of modulation effects in the PRESS (or stimulated echo acquisition mode [STEAM]) experiments will become increasingly important.

It is important to consider the fact that, although the volume excited can be calibrated to be the same after IVS, it is unlikely that the location of the excited spins will be identical in both cases. The voxel in MEGA-PRESS-IVS suffers less from chemical shift displacement than in MEGA-PRESS because the voxel edges are principally defined by the high-bandwidth saturation pulses. That aside, the shape of the transition regions at the edges of the voxel will also be different for MEGA-PRESS and MEGA-PRESS-IVS. The voxel edge is likely to be more sharply defined by the IVS, for three reasons: the IVS pulses are excitation rather than refocusing pulses; the refocusing pulses must be short (and therefore of relatively poor profile) to keep the MEGA-PRESS TE to 68 ms, and third, the width of the saturation bands is narrower than the width of the voxel defined by the refocusing pulse.

It is interesting to note that the signal at 3.7 ppm, which arises from Glx, also increases in intensity as a result of the MEGA-IVS editing. This signal arises from spins that are coupled to signals at about 2 ppm, resulting in coediting. Because of the larger chemical shift difference between the coupled spins (∼220 Hz), spatial effects are more severe for Glx than for GABA. It is fortunate, although not surprising, that the IVS parameters chosen for GABA detection also result in an increase in the Glx signal. This result is worthy of further investigation in the future; it also suggests that the intensity of Glx signals is reduced in standard PRESS experiments performed at echo times of 68 ms.

The quantitative discussion of the possible percentage increases in acquired signal assumes a homogeneous distribution of metabolite within the volume prescribed. In the event that this is not the case, spatial effects can result in either more or less severe effects, depending on the precise location of a region of high concentration within the PRESS-excited voxel. This situation may add a further source of variation to quantitative measurements. In the case of GABA, this may be particularly true, since the need to prescribe a large volume, as dictated by signal-to-noise constraints, means that is it not possible to prescribe an anatomically uniform volume.

With all frequency-selective editing methods, the possibility of coediting signals from spin systems similar to the target arises; in the case of MEGA-PRESS editing for GABA, signals from macromolecules (11) and homocarnosine (16) are also acquired at 3.0 ppm. The macromolecular peak at 3.0 ppm is coupled to spins at 1.7 ppm. It has been suggested that applying the editing pulse at 1.9 ppm and 1.5 ppm will reduce contamination from the coedited macromolecules (11). The method has been applied to image-selected in vivo spectroscopy (ISIS)-based experiments in which longer, more frequency-selective editing pulses can be used. In the current MEGA-PRESS implementation with 14-ms editing pulses, this positioning of the editing pulses would result in significant loss of GABA signal, as well as suppression of the macromolecular contribution. This measurement of GABA without suppression of coediting resonances is often referred to as GABA+. It has also been shown that a preexcitation inversion pulse can be used to suppress the macromolecular contribution, taking advantage of the different T1 of macromolecules and more mobile metabolites; however this was not done in the current study. Coediting of compounds other than GABA should be similar between conventional MEGA-PRESS and MEGA-PRESS-IVS.

One possible alternative to the use of saturation pulses to overcome the loss of signal due to spatially varying coupling evolution would be to combine the MEGA-PRESS experiment with phase-encoding to isolate those regions in which coupling is well-behaved. However, it would be important not to reduce the GABA sensitivity significantly.

In this work we have discussed the effect of IVS on MEGA-PRESS, but it would be interesting in the future to explore its application to other editing methods, such as PRESS-based double-quantum editing, and to compare the relative sensitivity of the many different editing methods that have been proposed.

To summarize, spatial variations in GABA signal intensity can occur at 3T, which may result in decreased sensitivity of detection using the MEGA-PRESS experiment. The lost GABA signal can largely be recovered using the IVS methodology previously demonstrated for lactate (14). Since the baseline concentration of GABA is low and the signal is difficult to detect in vivo by MRS, it is important to use IVS methodology to ensure maximum efficiency of GABA detection.


We thank Dr. Michael Schär for guidance in programming the MEGA-PRESS sequence.


Product Operator Consideration of the Spatial Cases

The mechanism of the editing sequence as applied to GABA and its spatial dependence can be illustrated using the product operator approach for the three coupled spins that are involved in the editing sequence (the H2 [observed] is treated as a single spin I, and the two H3 protons [S and S′] selectively inverted by the editing pulses). (For clarity, the slice-selective refocusing pulses originating from the PRESS sequence will be referred to as the refocusing pulses throughout this paper, while the extra chemical-shift selective pulses added to the MEGA-PRESS experiment will be referred to as the inversion pulses). Some assumptions on the timing of the experiment (which are true for our implementation, but may not be in all cases) have been made in the following calculations: TE1 is assumed to be shorter than TE2; it is also assumed that the first editing pulse occurs after the first spin echo. The evolution of a spin I, analogous to the H2 spins of GABA, which gives rise to a triplet signal due to a coupling of size J to two spins S and S′, analogous to the H3 spins of GABA, will be considered, assuming that the 180° refocusing pulses and the inversion pulses are ideal.

Case I: Both Refocusing Pulses Are Applied to S Spins

Off Experiment: Editing Pulses Applied to Neither Spin

The double echo of the PRESS experiment refocuses evolution of the chemical shift offset, but coupling between the spins evolves for the whole duration of the sequence, TE = TE1 + TE2. Assuming that the excitation and refocusing pulses are applied with phase x, the product operators of spin I present at the start of detection are:

equation image(A1)

It is customary to perform the MEGA-PRESS experiment (for the detection of GABA) with an echo time of 1/2J, in which case these terms simplify to 4 IySzS′z.

On Experiment: Editing Pulses Applied to S and S′

Evolution of coupling between the spins is refocused at the start of detection, giving –Iy.

Case II: Only the First Refocusing Pulse Is Applied to S Spins

Off Experiment: Editing Pulses Applied to Neither Spin

Coupling evolves as expected during the first echo. The second refocusing pulse acts to refocus evolution of coupling (full refocusing occurs at a time TE1 into the acquisition). The product operators present at the beginning of acquisition are:

equation image(A2)

On Experiment: Editing Pulses Applied to S and S′

The first inversion pulse refocuses evolution of coupling over the initial period (TE1 + ½TE2). The second refocusing pulse has no effect on the evolution of coupling, since there are therefore no antiphase terms present when it is applied. Evolution of coupling during the following period TE2 is refocused by the second inversion pulse, so that the operator present at the start of detection is –Iy.

Case III: Only the Second Refocusing Pulse Is Applied to S Spins

Off Experiment: Editing Pulses Applied to Neither Spin

The first refocusing pulse acts to refocus evolution of coupling during the first echo, TE1. Coupling then evolves for the second echo, giving:

equation image(A3)

On Experiment: Editing Pulses Applied to S and S′

Evolution of the coupling is again refocused by the first refocusing pulse. It is then refocused again by the first inversion pulse after a further time ½TE2 – TE1 (this point is a time TE2 into the sequence). The second editing pulse refocuses evolution of coupling at a time TE1 after the beginning of detection, so the product operators present at the start of detection are:

equation image(A4)

Case IV: Neither Refocusing Pulse Is Applied to S Spins

Off Experiment: Editing Pulses Applied to Neither Spin

Evolution of coupling is refocused by both of the refocusing pulses, so that the term present at the beginning of detection is –Iy.

On Experiment: Editing Pulses Applied to S and S′

The product operators present at the start of acquisition are:

equation image(A5)

These results can be characterized by an apparent coupling evolution time, or the length of time that the operator –Iy would have to evolve under coupling alone to give the same result.