Implementation of 3 T Lactate-Edited 3D 1H MR Spectroscopic Imaging with Flyback Echo-Planar Readout for Gliomas Patients
- First Online:
- Cite this article as:
- Park, I., Chen, A.P., Zierhut, M.L. et al. Ann Biomed Eng (2011) 39: 193. doi:10.1007/s10439-010-0128-x
The purpose of this study was to implement a new lactate-edited 3D 1H magnetic resonance spectroscopic imaging (MRSI) sequence at 3 T and demonstrate the feasibility of using this sequence for measuring lactate in patients with gliomas. A 3D PRESS MRSI sequence incorporating shortened, high bandwidth 180° pulses, new dual BASING lactate-editing pulses, high bandwidth very selective suppression (VSS) pulses and a flyback echo-planar readout was implemented at 3 T. Over-prescription factor of PRESS voxels was optimized using phantom to minimize chemical shift artifacts. The lactate-edited flyback sequence was compared with lactate-edited MRSI using conventional elliptical k-space sampling in a phantom and volunteers, and then applied to patients with gliomas. The results demonstrated the feasibility of detecting lactate within a short scan time of 9.5 min in both phantoms and patients. Over-prescription of voxels gave less chemical shift artifacts allowing detection of lactate on the majority of the selected volume. The normalized SNR of brain metabolites using the flyback encoding were comparable to the SNR of brain metabolites using conventional phase encoding MRSI. The specialized lactate-edited 3D MRSI sequence was able to detect lactate in brain tumor patients at 3 T. The implementation of this technique means that brain lactate can be evaluated in a routine clinical setting to study its potential as a marker for prognosis and response to therapy.
Keywords3D magnetic resonance spectroscopic imagingIn vivo lactate detectionBrain tumorGlioblastoma multiforme
Lactate is a metabolic marker that is observed in many brain pathologies.8,11 Neoplastic processes that are present in tumors have low oxygen supply and depend on non-oxidative glycolysis for energy production.31 This means that lactate can be considered as an indicator of anaerobic glycolysis and reduced cellular oxygenation, which is of interest for evaluating response to radiation or other therapies. The potential for identifying regions of metabolic stress and ischemic area in brain means that in vivo measurements of lactate are of interest in patients with a number of different brain pathologies. Previous studies performed at 1.5 Tesla (1.5 T) have shown that the presence of lactate and lipid peaks in 1H magnetic resonance spectroscopic imaging (MRSI) data is associated with a diagnosis of high-grade tumor.15,27 Elevated signals from lactate and lipid were associated with short survival in patients with glioblastoma multiforme (GBM) who were evaluated either prior to surgery or radiation treatment at 1.5 T.2,26 In these cases, increased lactate was interpreted as an indication of increased tumor metabolism and growth.1,24 The detection of lactate is thus of interest for evaluating prognosis and response to therapy in brain tumor patients.
1H magnetic resonance spectroscopy (MRS) of lactate demonstrates two resonances: a doublet at 1.3 ppm from methyl protons (CH3) and a quartet at 4.1 ppm from methine protons (CH). The methyl protons and the methine protons are weakly coupled to each other with a J-coupling constant of 6.93 Hz. For proton in vivo spectroscopy, the methyl doublet has been the target for lactate detection because the methine peaks are close to the water resonance and are not usually visible because of their relatively low signal intensity. Despite this advantage, the methyl doublet can be difficult to quantify because of lipid peaks, which overlap in the range of 0.9–1.3 ppm. In order to overcome this problem, a number of techniques have been developed to measure and separate the lactate doublet from lipid resonances.6,7,29
Lactate editing combined with point resolved spectroscopy (PRESS) localized 3D MRSI has been applied to glioma patients at 1.5 T for non-invasive detection of lactate and other brain metabolites.2,15,26 These studies demonstrated the detection of lactate as well as Cho, Cr, NAA and lipid, and suggested that in vivo measurement of lactate as well as other MR-derived parameters may help in diagnosis and proper therapy selection for glioma patients. Although the increased signal strength at higher field is expected to enhance the sensitivity of brain metabolites including lactate, the detection of brain lactate using 1H MRS based on J-difference editing at 3 Tesla (3 T) scanner has not been reported. Several studies have reported their unsuccessful attempts at measuring lactate in brain tumor patients using single voxel PRESS-localized MRS at 3 T.12,14 The poor lactate detection that was observed at 3 T in these studies was due to chemical shift mis-registration artifact caused by the limited bandwidth of refocusing pulses used for the localization of spectroscopic data.
Several studies have investigated the signal cancellation of J-coupled resonance due to the chemical shift difference between J-coupled spin partners in PRESS MRSI sequence in the context of lactate and GABA.9,10 This artifact is produced because spatially selective RF pulses cause a relative shift in the location of the selected volume for J-coupled resonances, thereby leading to net signal loss when the final signal is contributed from different regions. Kelley et al.10 demonstrated in phantoms that BASING pulses incorporated into non-editing PRESS sequence reduced the artifact for the detection of lactate methyl resonance in single voxel and MRSI data. For our editing scheme, this artifact may happen in the second cycle with BASING editing off, which is basically regular J-evolution. In the present study, we sought to eliminate this artifact by using higher-bandwidth RF pulses since the spatial offset due to chemical shift difference is inversely proportional to RF bandwidth. In addition, we over-prescribed PRESS-localized volume and applied high bandwidth saturation pulses in order to further minimize this artifact.
The clinical use of lactate-edited MRSI based on J-difference editing has been limited by the requirement of two successive acquisitions per phase encoding step and the acquisition time which has typically been 20 min.15 Flyback echo-planar spectroscopic imaging has been used to allow the acquisition of MRSI data in a shorter scan time. Cunningham et al.3 demonstrated the feasibility and potential of MRSI data acquisition at 3 T with high spatial resolution and large coverage in a short scan time with flyback echo-planar readout gradient waveforms.
The purpose of this study was to implement a lactate-edited 3D PRESS 1H MRSI sequence at 3 T with new high bandwidth 180° pulses, new BASING pulses and a flyback echo-planar readout gradient in order to allow a clinically suitable scan time of 10 min and demonstrate the feasibility of using this sequence for the detection of brain lactate as well as Cho, Cr, NAA, and lipid in patients. We addressed the effect of chemical shift artifact on lactate signal with different PRESS over-prescription (over-PRESS) factors and compared the metabolite SNR and its ratio of the lactate-edited MRSI data between a flyback echo-planar readout gradient method and conventional MRSI with elliptical k-space sampling. The method was then applied to patients with gliomas in order to determine whether it could detect lactate within lesions.
Materials and Methods
Sequence Development and Implementation
Comparison of BASING, VSS, and PRESS pulse parameters between the 1.5 T and the newly implemented 3 T sequence
Peak amplitude (mG)
Inversion BW (Hz)
Transition BW (Hz)
90° pulse BW (Hz)
180° pulse BW (Hz)
Estimation of Chemical Shift Artifacts with Over-PRESS Factors
All spectroscopic imaging data were acquired using a 3 T Signa HDx (v12x) MR scanner (GE Healthcare, Milwaukee, WI) and an eight-channel phased array head coil. The spectroscopic imaging data were acquired with PRESS volume localization with an over-PRESS factor between 1.2 and 1.7. The over-PRESS factor was achieved by reducing the amplitude of the slice selection gradients. Chemical shift-selective saturation (CHESS) pulses were used for water suppression. VSS pulses of width 40 mm were placed on all sides of the prescribed volume to define the edges of the selected volume, and six additional graphic VSS bands were used in order to further improve the suppression of subcutaneous lipids for volunteer and patient scan. High-order shimming was performed prior to the MRSI data acquisition to optimize the magnetic field homogeneity. In order to estimate the sensitivity profile of each coil element, proton density weighted gradient-echo (GRE) images were acquired using the manufacture-provided parallel imaging (ASSET) calibration sequence (TR/TE = 150/20 ms).
Chemical Shift Artifacts with Different Over-PRESS Factors in Phantom
Comparison Between the Conventional Phase Encoding and Flyback Readout Gradient Methods
Patient Data Acquisition
A total of 10 patients (eight male, two female, median age 55, range 41–64) with GBM were included in this study. Some patients had more than one examination and a total of 34 scans from these patients were examined to demonstrate the feasibility of using the new lactate-edited 3D MRSI sequence for the detection of lactate in brain tumor at 3 T. All patients were scanned using 3 T lactate-edited PRESS 3D MRSI sequence with the flyback echo-planar readout gradient applied in SI dimension using an over-PRESS factor of 1.5 (TR/TE = 1104/144 ms, FOV = 16 × 16 × 16 cm, voxel size = 1 × 1 × 1 cm, total acquisition time = 9.5 min, 712 dwell points, and 988 Hz bandwidth). The MRI protocol included an axial T2-weighted fluid attenuated inversion recovery (FLAIR) sequence4 with 3-mm slice thickness and axial pre- and post-gadolinium-DTPA T1-weighted spoiled gradient recalled (SPGR) images with 1.5-mm thickness. All patients provided informed consent as approved by the committee on human research at our institution.
Data Processing and Quantification
The method used for spectral quantification has been described previously.19 In brief, the 3D MRSI data with the conventional phase encoding were reconstructed by processing each of the spectra from the eight-channel coil individually, apodizing with a 4 Hz Lorentzian filter in the time domain, zero filling to 1024 points, applying phase and frequency correction, and then removing residual water and baseline. The individual data were Fourier-transformed to produce a 3D spatial array of spectra and combined using in-house developed software that weights the data by their coil sensitivities from the low resolution proton density weighted images.16,25 The two cycles of data were then either summed or subtracted to produce a 3D array of summed or subtracted spectra.
The 3D MRSI data with the flyback echo-planar readout were reconstructed and combined in the same way as the data with the conventional phase encoding, but with the application of an additional linear phase correction in the SI dimension of spectral k-space to prevent spatial chemical shift artifacts.3 To reconstruct the data acquired with the flyback echo-planar readout gradient, only samples from the flat part (plateau) of the gradient waveforms were used.
The SNR of brain metabolites was calculated using the previously published method.20 The region of 100 spectral points without metabolite signal was selected to estimate the standard deviation (SD) of the noise. Sub-regions with a size of 20, 40, and 80 points were made and located at the beginning of the selected region. The location of the sub-region was incremented by 1 point until it reached the end of the selected region, resulting in a total of 81, 61, and 21 divisions, respectively. For each sub-region, the SD was computed, the minimum SD was taken for each division, and the minimum SD of the divisions was used for the final SD of noise. The SNR of peak height over the SD of noise was calculated for Cho, Cr, NAA, and lactate.
In order to quantify lactate level in patients, median lactate SNR was estimated for each examination. Lactate peaks with SNR equal to or greater than 4 were only included in calculating the median lactate SNR. The estimation of lactate SNR was confined to the lactate appearing in contrast enhancing lesion (CEL) or T2 hyperintense region (T2h). The lactate found in resection cavity, necrotic region, or cerebrospinal fluid was excluded from the analysis.
Estimation of Chemical Shift Artifacts with Over-PRESS Factors
The Effect of Chemical Shift Artifacts on Metabolite Signal in Phantom
Comparison between three over-PRESS factors
Comparison Between the Conventional Phase Encoding and Flyback Readout Gradient Methods
The median SNR values and metabolite ratios of brain metabolites for raw flyback, normalized flyback, and conventional MRSI methods
Phantom (n = 1)
Volunteer (NAWM, n = 2)
Patients (NAWM, n = 33)
Patient Data from the Lactate-Edited 3D MRSI with a Flyback Gradient
This study demonstrated the feasibility of detecting lactate, as well as Cho, Cr, NAA, and lipid in 9.5 min using a new 3 T lactate-edited PRESS 3D MRSI sequence with a flyback echo-planar readout gradient. Using this approach, we were able to detect lactate in patients with brain tumors (median normalized SNR = 17) using a nominal voxel size of 1 cc and a factor of two reduction in scan time compared to conventional elliptical phase encoding MRSI (19 min).
Implementation of higher-bandwidth volume selection pulses was required because of the increased chemical shift between lactate methyl (1.3 ppm) and methine (4.1 ppm) resonances at higher field. It should be noted that the bandwidth of slice selective 180° pulses in the current study (1500 Hz) was higher than those reported in the previous studies (ranging from 874 to 1385 Hz) which used a PRESS scheme for volume selection at 3 T.14,16,28 The use of higher-bandwidth pulses was important in minimizing chemical shift mis-registration because the chemical shift artifact is inversely proportional to the size of the RF bandwidth used for localization. The use of an over-PRESS factor and high bandwidth VSS pulses further minimized artifacts from chemical shift mis-registration.
Over-PRESS factors of 1.2, 1.5, and 1.7 were tested in this study in order to minimize the chemical shift mis-registration of two lactate resonances as well as other metabolites. Although the higher factor would have given more uniform lactate signal and metabolite ratios, an over-PRESS factor of 1.5 was used for all volunteer and patient scans because of the proximity of the excited volume to the subject’s skull and the potential of having increased contamination from subcutaneous lipid with an 1.7 over-PRESS factor.
Although, in theory, the chemical shift artifact should be eliminated with an 1.5 over-PRESS factor since the excited box (91 × 102 mm) is bigger than the prescribed box (80 × 80 mm) (Fig. 3b), the lactate spectra with an 1.5 over-PRESS factor were still not completely uniform. Imperfections in slice selection may have contributed to the artifact. The lactate signal in the left column was relatively smaller than those in the rest of the excited volume, however, the variation of lactate level over the entire excited volume was relatively small and showed significant improvement over the lactate spectra acquired with the over-PRESS factor of 1.2 (Fig. 5). The lactate SNRs in regions besides the left column were very uniform for the over-PRESS factor of 1.5 with the SD of 0.5 (Fig. 4). When interpreting lactate peaks in the left column, one should note the possibility of underestimating lactate in this region and may use other means to correct for the chemical shift artifacts. This can either be achieved using theoretical corrections that are determined from the size of the PRESS selected volume, the over-PRESS factor and the bandwidths of the selection pulses or to acquire empirical data with similar parameters from a uniform phantom and use the ratio of lactate intensities in different voxels to provide correction factors.
The trade-off in using a flyback echo-planar readout gradient is a decrease in sensitivity. The median signal reduction for the flyback echo-planar readout gradient compared to the conventional phase encoding in the normalized SNR of NAA was 17% for phantom and 22% for volunteers (Table 3), which are similar to previous findings described by Cunningham et al.3 The loss of data during rewind portion of the flyback trajectory and imperfections in the gradient trajectory contributed to the net signal loss. Lactate peaks with SNR smaller than 4 were difficult to be distinguished from noise, and therefore the SNR of 4 was used as a threshold for calculating the median SNR of lactate in patients. Although the conventional phase encoding method may provide higher sensitivity for lactate detection, especially when lactate concentration in tumors is small, the long scan time (20 min) has traditionally prevented it from being used in routine clinical practices. By integrating the flyback echo-planar readout gradient, we have made possible a factor of two reduction in scan time (9.5 min). As the use of 3 T scanner becomes more widespread due to its improved SNR associated with higher field strength, this technique is expected to be more appropriate for routine clinical uses and may assist in evaluating the role of lactate in the management of brain tumors.
In most cases, the highest lactate signal was observed in necrosis or resection cavity with a raw SNR as high as 18 in these regions. It is known that lactate produced in cancerous cells can accumulate in cystic or necrotic regions.13 Since lactate in these regions has been reported to be clinically irrelevant,2 we limited our estimation of lactate signal to CEL or T2h and excluded lactate in necrosis and resection cavity.
The current study was primarily aimed at demonstrating the practicability of detecting lactate signal using the new lactate-edited 3D MRSI sequence in patients with brain tumors at 3 T; hence the clinical interpretation of lactate was omitted. A study assessing the patterns of lactate prior to radiation and during treatment in GBM patients is currently underway in an attempt to predict treatment outcome.
We have developed and implemented a lactate-edited 3D MRSI sequence that incorporates specialized rf pulses and a flyback echo-planar readout gradient at 3 T. The results from the current work demonstrated the ability of this sequence to detect lactate in the presence of lipid signal from brain tumor patients in a clinically acceptable acquisition time of 9.5 min. While further studies are required, this robust technique should permit non-invasive measurement of brain lactate as well as Cho, Cr, NAA, and lipid in routine clinical settings at 3 T and may assist in investigating its potential as a marker for prognosis and response to therapy.
This study was partially supported by an academic-industry research grant ITL-BIO04-10148, which is funded by the UC Discovery Program in conjunction with GE Healthcare, and by NIH grants R01CA059880 and R01CA127612.
This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.