Sequence design for magnetic resonance spectroscopic imaging of prostate cancer at 3 T



Magnetic resonance spectroscopic imaging (MRSI) has proven to be a powerful tool for the metabolic characterization of prostate cancer in patients before and following therapy. The metabolites that are of particular interest are citrate and choline because an increased choline-to-citrate ratio can be used as a marker for cancer. High-field systems offer the advantage of improved spectral resolution as well as increased magnetization. Initial attempts at extending MRSI methods to 3 T have been confounded by the J-modulation of the citrate resonances. A new pulse sequence is presented that controls the J-modulation of citrate at 3 T such that citrate is upright, with high amplitude, at a practical echo time. The design of short (14 ms) spectral–spatial refocusing pulses and trains of nonselective refocusing pulses are described. Phantom studies and simulations showed that upright citrate with negligible sidebands is observed at an echo time of 85 ms. Studies in a human subject verified that this behavior is reproduced in vivo and demonstrated that the water and lipid suppression of the new pulse sequence are sufficient for application in prostate cancer patients. Magn Reson Med 53:1033–1039, 2005. © 2005 Wiley-Liss, Inc.

Magnetic resonance spectroscopic imaging (MRSI) has proven to be a powerful tool for the metabolic characterization of prostate cancer in patients before and following therapy (1). The metabolites that are of particular interest are citrate and choline, because an increased choline-to-citrate ratio can be used as a marker for cancer. Although most of the work so far has been done using 1.5-T whole-body scanners, higher field systems (≥3T) offer the promise of higher spectral resolution, as well as increased magnetization that can be used to achieve higher spatial resolution and/or higher SNR. Thus, MRSI at 3 T has the potential to significantly improve the metabolic assessment of prostate cancer's.

However, initial attempts at extending MRSI methods that have worked at 1.5 to 3 T have been confounded by the J-modulation of the citrate resonances. The modulation at 3 T is different than that at 1.5 T because the protons of the citrate molecule behave as a strongly coupled AB spin system, and thus the relative phase and amplitude of the different spectral lines depend on both field strength and echo time. Investigations into the optimal echo time and RF pulse spacing for the PRESS pulse sequence at 3 T using 2D -J-acquisitions (2) have demonstrated that citrate is maximally upright at a TE of 260 ms (3, 4) vs. 120 ms at 1.5 T (5). This long echo time is impractical due to SNR losses with T2 decay, especially for metabolites with shorter T2. Prior studies have suggested that achieving maximally inverted citrate at ≈90 ms is a better approach (3, 6). Inverted lines, however, are undesirable because of the potential for cancelation with neighboring spectral components within inhomogeneous voxels, reduced amplitude, and difficulties in automated spectral processing. Attempts to use J-refocused PRESS techniques (7, 8) to acquire upright citrate at 3 T were only partially successful, resulting in a TE of 180 ms (4). While better than 260 ms, this TE is still longer than desired given the T2 decay.

In this paper we present the design of a new pulse sequence that controls the J-modulation such that citrate is upright, with high amplitude, at a practical echo time. The development of pulse-sequence components such as the short (14 ms) spectral–spatial refocusing pulses and trains of nonselective refocusing pulses are described. Phantom experiments and simulations demonstrating the modulation of citrate at various echo times are shown. In vivo spectra are presented, showing that using the new sequence results in upright citrate, along with the other metabolites of interest as well as robust water and lipid suppression.


The protons attached to the CH2 groups of the citrate molecule (OOC–CH2–C(OH)(COO)–CH2–COO) behave as a strongly coupled AB spin system. Thus, with pulse sequences conventionally used for MRSI localization, such as the PRESS sequence (9), the four peaks that comprise the citrate spectrum go through a complicated modulation in amplitude and phase as a function of echo time. This behavior is not readily described in classical terms and requires a quantum mechanical description. The most convenient expression for calculating citrate behavior is obtained through the density matrix formalism. These matrix expressions are implemented as part of a numerical simulation under Methods. The density matrix, ρ, holds all of the information about the spin system at any point in time, with the evolution of the system as a function of time governed by the Hamiltonian H:

equation image(1)

The Hamiltonian for an AB spin system can be written as

equation image(2)

where I and S are the angular momentum operators for spins A and B, respectively, ωA and ωB are the chemical shifts of spins A and B, respectively, and the subscripts denote the direction along which the operators act. The first bracketed expression in Eq. [2] is the conventional Zeeman Hamiltonian, and the second term is a dot product between the operators for spins A and B, governing the coupling behavior.

With the conventional PRESS sequence and echo times typical of in vivo application (70–140 ms), there is enough evolution time that J-coupling significantly changes the citrate spectrum from its equilibrium state (Fig. 1). Such long echo times are required so that T2 decay can attenuate the abundant signal from macromolecules, which would otherwise overwhelm the signals of interest. However, with the PRESS sequence at 3 T, the best condition that can be found (at a practical echo time) is maximally inverted citrate as seen at TE = 90 ms in Fig. 1. To cope with this problem, we investigated modifications to the number of RF pulses and the pulse timings as a means of controlling J-modulation.

Figure 1.

The effect of J-modulation on the appearance of the citrate spectrum at 3 T with the conventional PRESS sequence. The four peaks undergo a complicated modulation in amplitude and phase as a function of echo time. In this simulation, the time between the initial 90° pulse and the first 180° pulse was set to 10 ms.

It has long been known (10) that a train of closely spaced refocusing pulses could inhibit J-modulation, even in strongly coupled systems such as that described by Eq. [2]. The reason for this, as given analytically in (10), is that the frequency of the echo train modulation depends on a factor Rtcp, where tcp is the time between refocusing pulses. The variable R is given by

equation image(3)

where J is the coupling constant and δ is the chemical shift difference between the spins A and B, with both values in hertz. Although the dependence on Rtcp is quite complicated, the modulation can be shown to become negligible for Rtcp ≪ 1 (11, 12). With approximate numbers for citrate at 3 T (δ = 16 Hz, J = 15 Hz (13)), the limit on tcp for negligible modulation is tcp ≪ 46 ms.

The goal of the pulse-sequence design was to incorporate the J-refocusing properties of a train of refocusing pulses into a pulse sequence robust enough for prostate MRSI. For water and lipid suppression, the use of spectral–spatial RF (SSRF) refocusing pulses within the PRESS sequence (14) has proven to be a reliable approach (15). A significant advantage of SSRF pulses is that the short duration of the subpulses (e.g., 1 ms) provides high bandwidth in the spatial dimension (e.g., 8 kHz). This limits the chemical-shift misregistration that becomes problematic at high field. The new sequence design that is proposed for citrate detection in the prostate is shown in Fig. 2. The interrelation of the pulse timings is given in Table 1. All of the spacings are set as short as possible, except for the interpulse spacing within the refocusing train, which is used to control the echo time.

Figure 2.

Diagram of the new pulse sequence. The spatial selection is accomplished in the same way as the conventional PRESS sequence, with a spatially selective 90° pulse and two spectral–spatial 180° pulses selecting the three dimensions of the excited/refocused volume. The train of hard 180 s with an MLEV phase cycle provides J-refocusing. The bipolar crusher gradients surrounding the hard 180 s inhibit unwanted FIDs that could come from the transition regions of the 180° pulses. The bottom trace shows the phase of the SSRF pulses.

Table 1. Interrelation of timing parameters for the sequence design shown in Fig. 2.
Timing intervalComputationValue used (ms)
  1. Note. For the timings that are set as short as possible (t12 and ttl) the dominant factors in determining the timing are listed, where t90 is the duration of the initial 90° pulse, tss is the duration of the spectral–spatial pulse, and tc is the duration of the crusher gradient. A long t23 is used so that formation of the spin echo is sufficiently far from the last crusher. The right column gives the values used in the experiments.

t12equation image13.0
ttlequation image11.9
tcp(TE − 2t23 − 2ttl)/(N − 2))4.2
ttrttlt12 + t23tcp12.7


In order to limit the degree of J-modulation during each RF pulse, new dual-band spectral–spatial pulses (15) were designed to be as short as possible. The challenge was to retain adequate tolerance to main-field inhomogeneity, which is a particular concern for prostate MRSI because of the nearby air–tissue interface. The B-polynomial for the spectral dimension (one of the inputs to the Shinnar–Le Roux (SLR) transform (16)) was designed using the complex-valued version of the Remez exchange algorithm, as implemented in MATLAB (The Mathworks, Inc., Natick, MA). To reduce the peak RF amplitude, nonlinear phase was created across the spectral passbands using the “root flipping” method (17, 18) as done in Ref. (15). This B-polynomial, together with the B-polynomial for the spatial dimension (time-bandwidth = 7.2), was passed through the inverse 2-D SLR transform (19) to generate the pulse shown in Fig. 3. The gradient waveform was designed to meet the constraint that the peak amplitude of the pulse be within the RF amplifier limits for body-coil transmission (16 μT). In order to keep the amplitude down, a fairly long (1 ms) sublobe duration was used, and the length of each gradient ramp was limited to 200 μs, giving a longer plateau. This caused the maximum gradient amplitude to be 30 mT/m (see Fig. 3b), giving a minimum voxel thickness of 4.8 mm, which was adequate for this application. The bandwidth of the pulse in the spatial dimension is 6.1 kHz, giving a nominal chemical-shift misregistration of 0.5 mm for citrate relative to choline, assuming a 40-mm profile. However, due to the use of the 2-D design method (19), the chemical-shift misregistration over the passband (choline to citrate) is zero. The spectral response of the pulse accommodates a +28/−30 Hz tolerance to main field inhomogeneity (these shifts result in 5% attenuation of citrate and choline, respectively). The bandwidth in the spatial dimension of the pulse was 6.1 kHz.

Figure 3.

Design of a short spectral–spatial refocusing pulse. The RF (a) and gradient (b) waveforms are shown. The net duration of the pulse is 14 ms, with each sublobe 1 ms long. The pulse was designed for body-coil transmission, with a peak amplitude of 16 μT. The spectral response of the pulse (c) shows a passband (≥95%) for the metabolites of interest ranging from −162 Hz (3.43 ppm) to −297 Hz (2.37 ppm), with the frequencies given relative to water. With two of these pulses applied in succession (see Fig. 2) the spectral profile achieved will be the square of the profile shown, giving partial (100-fold) suppression of water in the attenuated passband and 120 dB (million-fold) suppression of lipids in the stop band beginning at −390 Hz (1.65 ppm). The 2D spatial vs. spectral profile (d) is plotted at the minimum slice thickness of 4.8 mm (FWHM).

These new SSRF refocusing pulses were implemented in the modified version of the PRESS sequence shown in Fig. 2. A train of eight nonselective (hard) refocusing pulses was used, with an eight-step MLEV (20) phase cycle to provide some tolerance to RF inhomogeneity and mis-set transmit gain. For comparison with experimental results, the J-modulation of citrate within this pulse sequence was simulated using the density matrix method expressed in Eq. [1]. With Eq. [2] written in matrix form, and a matrix expression for the effect of an RF pulse (21), the behavior of Eq. [1] was investigated in MATLAB. Each spectral–spatial pulse shown in Fig. 2 was represented by a series of hard pulses separated by the sublobe duration (1 ms), with each hard pulse having the same flip angle as the corresponding subpulse and free evolution between the hard pulses.

The pulse sequence was implemented on a whole-body GE Signa 3-T scanner (General Electric Healthcare Technologies, Waukesha, WI). The quality of the spatial localization was investigated by acquiring projection images through the excited/refocused volume. Images were acquired with and without outer volume suppression pulses (22, 23), both with and without the hard-pulse re focusing train. To approximate the conditions used in vivo, an inflatable endorectal coil (Medrad Inc., Indianola, PA) was placed adjacent to the excited volume and connected to the scanner, but the body coil was used for both RF transmission and reception to eliminate coil reception profile inhomogeneities from these measurements. The phantom was spherical and contained distilled water doped with gadolinium (T1 ≈ 1 s). The parameters for this acquisition were TR = 1.7 s, TE = 85 ms, 256 × 128 encoding, 20-cm field of view. To test the performance of the individual spectral–spatial refocusing pulses, a spin-echo profile measurement was made in a uniform rectangular phantom as in Ref. (24). By misadjusting the magnetic field shim along the length of the phantom, the spectral selectivity of the pulse was resolved.

To test the sequence in spectroscopic imaging mode, data were acquired from a spherical phantom containing the metabolites of interest (choline, creatine, and citrate) in solution. Voxels of 5.4 mm per side (0.157 cm3) were resolved by phase encoding (chemical shift imaging (CSI)), with 12×8×8 encoding steps. An endorectal coil (described above) was used for signal reception, and the whole-body birdcage coil was used for RF transmission. The echo time was varied from 75 to 105 ms to investigate the J-modulation. The TR was 1.3 s.

In vivo spectra were acquired from a human subject, scanned under a protocol approved by the University of California Institutional Review Board. The subject was 62 years of age and had biopsy-proven cancer in the left mid-gland with a Gleason score of 2 + 2. The same voxel resolution, phase-encode coverage, and coils as described above were used. The dimensions of the excited/refocused volume (42.6 × 25.3 × 24.0 mm3) were chosen to exclude periprostatic lipids. The 85-ms echo time was chosen based on the best results from the phantom experiments. The time-averaged RF power deposition was estimated from the reflected power monitor built into the scanner. The TR was manually varied until the power deposition was within safety limits (2 W/kg), giving a minimum TR of 1.5 s for this subject. The raw data were saved for offline reconstruction using custom software (25). Line broadening (3 Hz Gaussian) was applied in the time domain.


The experimental characterization of the spatial localization achieved with the new pulse sequence is shown in Fig. 4. The spectral vs. spatial profile shows agreement with the simulation in Fig. 3. The spatial profile of the excited/refocused volume is made considerably sharper when the saturation pulses are applied (see Fig. 4b and d). No change in the selected volume was observed when the nonselective refocusing train was turned off.

Figure 4.

Experimental characterization of the spatial localization achieved with the new pulse sequence. (a) The spectral vs. spatial profile for one of the 14-ms spectral–spatial refocusing pulses. (b) Projection through the superior/inferior dimension of the selected volume with outer volume suppression. (c) Same as (b) but without outer volume suppression. (d) Line plots through the left/right dimension of the images in (b) and (c) shown in upper and middle plots, respectively. For (d), bottom, the hard-pulse refocusing train was turned off. The difference in brightness between (b) and (c) is due to the reduced thickness of the volume in the projection dimension (through the page) when the outer volume suppression pulses are used.

The results from the phantom experiments, with a comparison to the simulation results, are shown in Fig. 5. Good agreement between simulation and experiment can be seen. However, there are slight discrepancies that are likely due to the approximate representation of the SSRF subpulses with hard pulses. This approximation was used because the actual duration of the subpulses (1 ms) was deemed negligible on the time scale of the J-coupling (15 Hz). The spectrum at TE = 85 ms shows an ideal situation, with high-amplitude, upright central peaks and suppression of the side wings. Note that J-modulation has not been completely removed (there are still minor variations with echo time). Instead, the J-modulation has been manipulated such that a useful spectral pattern is achieved at a practical echo time.

Figure 5.

Results from phantom experiment using the new pulse sequence. The citrate spectrum is shown at various echo times, with the corresponding simulation result for comparison. Each experimental spectrum is from a single 0.157-cm3 voxel, resolved from the full volume by phase encoding (CSI).

The spectra and corresponding anatomical image from the in vivo study are shown in Fig. 6. Note the upright citrate, possibly with small side bands buried in the noise floor, and the flat baseline. Splitting is visible between the two center citrate peaks. The other metabolites of interest are also visible within the spectral window. The second, attenuated passband of the spectral–spatial refocusing pulses gives the partially suppressed water shown in Fig. 6c, which is used for a phase and frequency reference for the reconstruction algorithm.

Figure 6.

Results from scan of a human subject using the new pulse sequence. The spectra shown in (b) are from 0.157-cm3 voxels, located in the region depicted on the T2-weighted image (a). The volume within the large white box was selected with the RF pulses, and the voxels within were resolved by phase encoding (12×8×8). The partially suppressed water can be seen in (c). TE was 85 ms.

The favorable behavior of citrate with this pulse sequence results from the short duration of each of the RF pulses, but mostly from the refocusing train added between the spatially selective pulses. This refocusing train is thus acting much like the RF decoupling train for which the MLEV phase cycle was originally developed (20). Inclusion of these extra refocusing pulses comes at the expense of higher RF power deposition. However, because the RF heating is dominated by the outer volume suppression pulses (22, 23), the addition of the refocusing train increases heating by only about 25% (0.5 W/kg).

The particular number of refocusing pulses and the interpulse timings used here were selected to provide easily measurable citrate. For this study, the behavior of coupled protons in molecules other than citrate, such as the polyamines, was not considered. The polyamine resonances cover a range overlapping with choline and creatine and can overwhelm the signal from these metabolites at shorter echo times, so attention will have to be focused on this in future development of the sequence. The refocusing train may provide high-amplitude peaks from the polyamines, in addition to citrate, which would be an unfortunate side effect. Increasing the number of pulses in the refocusing train might allow manipulation of the J-modulation of the polyamines and could allow slightly longer echo times to give more T2 attenuation of the polyamine signal. The sequence was designed to use other lengths of refocusing trains, such as 12 and 16 pulses, but these have not yet been thoroughly tested.

Using these particular pulse timings, the state of the spins has been manipulated such that the spectrum of the strongly coupled metabolite appears in an easily measurable form. This is in contrast to schemes such as LASER (26) and CPRESS (11), which are capable of almost completely quenching the J-modulation during the echo train. Although we have attempted to incorporate the features of these refocusing-train methods, the use of spectral–spatial 180° pulses precludes the closely spaced pulse trains that are the key to these methods. Instead, what we have achieved is closer in essence to JPRESS (4, 7, 8), in which a particular echo time must be chosen to get a pure in-phase signal. However, the new method allows a much shorter TE for citrate at 3 T and thus avoids substantial T2 losses for choline and other metabolites.


A new pulse sequence has been developed for MRSI of prostate cancer at 3 T. New spectral–spatial refocusing pulses were designed to be as short as possible (14 ms) so that J-modulation over the duration of the pulses would be minimized. A phase-cycled train of nonselective refocusing pulses was inserted inbetween these spatially selective refocusing pulses to manipulate the J-modulation of the citrate spectrum. Phantom studies and simulations showed that upright citrate with negligible sidebands is observed at an echo time of 85 ms. Studies in a human subject verified that this behavior is reproduced in vivo and demonstrated that the water and lipid suppression of the new sequence are sufficient for prostate cancer MRSI studies.


We thank Beverly Fein and Niles Bruce for assistance with the in vivo data acquisition. Supported by NIH grant RO1-CA05987.