Incorporating lactate/lipid discrimination into a spectroscopic imaging sequence

Iiicorporating Lactatekipid Discrimination into a Spectroscopic Imaging Sequence Elfar Adalsteinsson, Daniel M. Spielman, Graham A. Wright, John M. Pauly, Craig H. Meyer, Albert Macovski A s;pectroscopic imaging sequence incorporating a two-shot lactate editing method was used in two human brain studies to image lactate and NAA. The subtractive editing method allows separate images of lactate, NAA, and lipids to be collected during a single study with no SNR penalty. The sequence uses a spectral-spatial excitation for slice selection and water suppression, and employs inversion recovery and an echo time of 136 ms for additional lipid suppression. Key words: spectroscopic imaging; lactate; lactatellipid discrimination; &coupling.

INlrRODUCTlON

sequence where both coupled and uncoupled spins can be observed. Spectral-spatial RF pulses (12) excite simultaneously a range of chemical shifts and a spatial slice. These pulses allow an efficient design of a spectroscopic imaging sequence that incorporates a lactate editing method (13-16) which produces both edited spectra for lactate, and non-edited spectra for uncoupled spins. In this communication we discuss the design of a spectroscopic imaging sequence that allows imaging of lactate without sacrificing the signal from NAA, and present two anecdotal in vivo results that demonstrate the utility of this method.

Detection of in vivo lactate is complicated by the presence of signals from both water and fat. Water can be suppressed based on its chemical shift difference; however, lipids are co-resonant with lactate. For proton spectroscopic imaging of the brain, a number of lipid suppression techniques have been proposed. The excited volume of interest can be limited to an area localized wholly within the brain (1,2). The relatively short TI and T2 relaxation times of lipids have also been exploited by using inversion recovery methods and late echoes (3). While these methods do suppress lipids, they do not definitively distinguish lipids from lactate. Residual lipids from abnormal pathology may be falsely identified as lactate, as can subcutaneous fat close to the region of intlerest through sidelobes in the spatial impulse response. \larious zero and double quantum filtering techniques have been be used to pass only signals from coupled spins (4-7). However, all signals except lactate are suppressed, and the data from uncoupled peaks such as Nacetyl aspartate (NAA), creatine, and choline are lost. Significant interest has been devoted to research on those metabolites (8-11) and their relation to disease, suggesting that relevant information may be lost if they are not obs,erved. Thus, it is desirable to incorporate efficient lactatel lipid discrimination into a spectroscopic imaging pulse

METHODS

The J-coupling effect between the methyl and methine protons in the lactate molecule has been used in a subtraction-based scheme (13-16), where two excitations are needed to obtain a lipid-free lactate spectrum. The two sequences used here, are

();,

1

- 2/ -

=1x

1 -21

111

and

(;),

1 -

-

1

=kdx

-

-*

121

2J

The 1-index refers to spins that resonate around the the CH3 doublet of lactate at 1.3 ppm, and k refers to spins close to the CH quartet at 4.2 ppm. The x and y indices refer to the phase of the RF excitation. Neglecting relaxation, the spin operator calculations in (17) can be used to analyze the effects of sequences 111 and [2] on lactate, assuming it is a weakly coupled spin system. With an initial magnetization of Ikz + I,=, excitation by sequence [I]leads to a final magnetization of Ikz + I,x at echo time 111.The k-component of the spin is unexcited, but the I-component is purely transverse magnetization. On the other hand, applying sequence [2] to the same initial magnetization results in final magnetization of - Ikz - I,x at echo time 1lJ. Again the k-component has longitudinal magnetization only, but the l-component has transverse magnetization with a phase difference of T with respect to the previous case. Spectra of the difference between the signals from [I] and [21 will contain the coupled lactate doublet, while signals from other spins (NAA, lipids) will cancel. The sum of the two spectra will show uncoupled spins (NAA),in addition to spins that are coupled to a partner within the passbands of the ( T L ! )and ~ ~ nIx pulses (lipids). This method is illustrated schematically in Fig. 1.

MRM 30:124-130 (1993) From the Magnetic Resonance Systems Research Laboratory, and the Radiological Sciences Laboratory, Department of Diagnostic Radiology (D.LI.S.), Stanford University, Stanford, California. Address correspondence to: Elfar Adalsteinsson, 120 Durand, Stanford University, Stanford, CA 94305-4055. Received December 21, 1992; revised March 8, 1993; accepted March 8, 1999. This work was supported by the General Electric Company, the Lucas Foundation. and the National Institutes of Health Grants CA48269 and CA44665. 0740-3194/93 $3.00 Copyright 0 1993 by Williams 8 Wilkins All rights of reproduction in any form reserved.

124

Incorporating Lactate/Lipid Discrimination

125

from which spectroscopic images can be calculated. A sequence diagram with the two different rf excitations and gradient waveforms is shown in Fig. 2. In the modified sequence, the signal from uncoupled spins is obtained from the sum of the excitations described by [ I ] and [Z]. It has the same SNR as if acquired with the original spectroscopic sequence without the spectral editing scheme, using two averages. Thus, compared with a conventional scan with averaging, there is no time or SNR penalty associated with incorporating the editing technique into the pulse sequence.

Lactate r---------

m

Seq. 1

7

'

:

.

? Seq. 2

Uncoupled spins and lipids I--------I

II

Seq. 1 Seq. 2

'\

I I I

I

't,

Ok

Chemical shift

RESULTS

0 1

To demonstrate the use of the method described above for in vivo imaging of lactate and NAA, we present the reFIG. 1. Schematic illustration of the editing method. The dashed sults of two patient studies. One patient had a brain tulines indicate ideal spectral profiles of the (7r/2),y,7rb, and T ~ , / ~mor (Oligodendroglioma) and the other had suffered a pulses. The triangles represent the coupled lactate, and excited right parietal stroke. components of uncoupled spins (and lipids) at chemical shift w,. Both studies were carried out on a 1.5 T whole body Resonances at chemical shift okare not excited. Signa magnet with shielded gradients and a quadrature head coil. In vivo shimming preceded each study (20). The range of peaks that are available with this method The echo time was 136 ms, the optimized inversion time was 175 ms, and the repetition time was 2 s, which re, ~ qxpulses. depends on the passbands of the ( ~ / 2 ) and ) ~ ~ sults in a scan time of 17 min for a 16 X 16 point acquiUncoupled spins within the passband of the ( ~ d . 2and sition. The nominal voxel size was 1.5 X 1.5 X 1.5 cm, or Tlxpulses, as well as spins that are coupled to a partner approximately 3.4 CC. No smoothing was applied on the within this passband, see the same spin echo in both k-space data. excitations, while uncoupled spins outside this passband For each study we show the edited lactate image, the are not excited and will not produce transverse magneNAA image from the sum spectra, and a T2-weighted tization. In this study, the passbands were chosen to be high resolution water image corresponding to the specequal for the two pulses, with half amplitude bandwidth troscopic slice. An unedited image from sequence [I] of 3.9 ppm, centered on the CH3 doublet of lactate. The (processed in the same way as the edited lactate image) stopband starts at 2 3 . 1 ppm from the lactate doublet. The that contains both lactate and lipids is also presented. peaks from creatine and choline are in the transition The spectroscopic images were interpolated onto a 256 X bands of these pulses and will not be observed, while 256 grid from the original 16 X 1 6 data set, and overlayed NAP, is readily imaged. Water suppression is achieved by onto-an edge detected version of the water image as a placing the water peak in the stopband of the excitation visual aid (3) in identifying anatomical landmarks. pulse. A lipid-suppressed single and multisection spectroscopic imaging sequence is described in ref. (3). It can be modified to efficiently incorporate the lactate editing method described above. The ( ~ / 2 ) , -pulse ~ is implemented as a spectral spatial excitation, which provides the needed water suppression along with the slice selection. The pulse is designed to satisfy the previously menRf2 tioned passband/stopband criteria, and the slice thickness is approximately 1.5 cm. A 4-ms sinc pulse was used Sljce both as an inversion pulse and as the ~ hspin-echo , ~ ~selection , gradient pulse, although to reduce sensitivity to B, inhomogeneity, a hyperbolic secant pulse could be used as an inversion pulse. The rrIx spin echo pulse was designed with the Shinnar-Le Roux algorithm (18),using the same spectral criteria as for the ( ~ / 2 excitation ) ~ ~ pulse. The rela0 50 100 150 200 250ms tively short TI relaxation times of lipids ( T,,llp 300 ms) relative to NAA and lactate (Tl,lac T1,NAA = 1500 ms FIG. 2. Sequence diagram. RF, and RF2 are the two excitations (19)) were used to improve lipid suppression by inverneeded for the editing method; both excitations use the same set of sion recovery prior to excitation, where the inversion gradients. The inversion time is chosen to minimize the lipid signal time, TI, was tuned to minimize the lipid signal at an at an echo time of 136 ms. The spectral-spatial pulse and the echo time of 136 ms (3). With phase encoding along x and spectrally selective spin echo pulse are 16 ms long, other RF y, this sequence generates a three dimensional data set pulses are 4 m s long.

q

.

1 A

-

-

AA-

126

Adalsteinsson et al.

FIG. 3. Brain tumor, Oligodendroglioma. (a) A slice from a &-weighted axial spin echo scan corresponding to the spectroscopic slice. Spectroscopic images based on peaks from (b) lactate, (c) NAA, and (d) lipids. (e) Unedited image, acquired with sequence [l], that contains both lactate and lipids and was created with the same spectral integration width as the edited lactate image. The spectroscopic images are interpolated onto a 256 x 256 grid from the original 16 x 16 data set, and overlayed onto an edge detected version of the water image (a).

127

Incorporating LactatelLipid Discrimination

Figure 3 presents the results for the brain tumor patient. The Tz-weighted water image in Fig. 3a shows abnormality in the left frontal lobe, corresponding to an increased lactate signal and a decreased level of NAA in the same location in Figs. 3b and 3c. The lactate signal is hardly visible in the unedited image, Fig. 3e, while in the edited image, Fig. 3b, the subcutaneous fat ring is canceled, and the lactate is clearly seen. The NAA image has lipid artifacts from subcutaneous fat since the chemical shift difference between the NAA peak and the much stronger lipid signal is not large enough to separate the two signals when the image is formed from the sum spectra. The magnitude of sum- and difference spectra from three voxels are shown in Figs. 4a and 4b. The spectra show that the lipid signal, already significantly reduced due to the inversion pulse and a late echo, is reduced down to the noise level in the difference spectra. NAA is easily observed in the sum spectrum, and the lactate doublet is clear in the difference spectrum from a voxel within the tumor. The residual signal in the 3- to 5-ppm region is from water that is excited though sidelobes in the excitation profile of the spectral-spatial pulse. Figure 5 presents the results for the stroke patient. The T,-weighted spin-echo image shows asymmetry between the left and right halves of the brain at this particular

3 c

L

6

4

2

0

-2

-4

-6 ppm

6

4

2

0

-2

-4

-6pprn

FIG. 4. Brain tumor, Oligodendroglioma. The magnitude of (a) sum- and (b) difference spectra from the three voxels labeled in Fig. 3a. Lipids and NAA are observed in the s u m spectra, but only lactate is seen in the difference spectra.

slice. The NAA signal is significantly decreased in the right half, Fig. 5c, and Fig. 5b shows a marked increase in lactate in two spots on the right hand side. In the unedited image, Fig. Se, the lactate can not be seen. Figures 6a and 6b show the magnitude of sum- and difference spectra from three voxels in these images. Signals from lactate and NAA are clearly observed, although the lactate signal is weaker than in the previous study. DISCUSSION

The images presented demonstrate the feasibility of incorporating a lactate editing method into a spectroscopic imaging sequence. There are however a number of important issues that have to be considered when this method is used. Motion can easily cause problems with a subtractive editing method as the one described above. However, in studies of the human head, gross motion of the head can be easily limited over the 17-min scan time. It is a significant challenge to robustly isolate a weak lactate signal in the presence of lipids with orders of magnitude greater concentration. Small errors (e.g., Bo, B1inhomogeneity) yielding imperfect cancellation of lipids in the signal subtraction could overwhelm the lactate signal. However, if the intensity of lipids can be reduced to be on the same order of magnitude as the lactate signal, it is possible to use the subtraction based method in a robust manner. Inversion recovery is used to achieve additional lipid suppression and improve the quality of the subtraction. Furthermore, uncoupled spins with long relaxation times are better observed since the strong lipid signal is reduced. In head studies, inversion recovery can typically reduce the lipid signal by one order of magnitude while the loss of lactate signal due to the inversion is 20-30% (assuming TI = 180 ms, = 1500 ms (19)). The choice of echo time is limited by the J-coupling frequency to be an integer multiple of l/J = 136 ms. In this study we chose an echo time of 136 ms which will further reduce the lipid signal by 8O%, while the additional loss in lactate is about 10°/o (assuming Tz,lac= 12’00ms (19), T2,up= 85 ms. (21)). It is possible to collect the signal at an echo time of 272 ms, or to acquire a second echo with a spectrally selective spin echo pulse (timing of the 2nd echo is not constrained by J-coupling), but both of these options sacrifice more lactate signal for enhanced lipid suppression. The spectral profiles of the spin echo pulses in sequences [I] and [2] determine the quality of the lipid cancellation in the subtraction. One can show that for small errors in the flip angles, the lipid signal does not cancel, rather it is multiplied by a term that is proportional to 1 rn,,(w) - rnh,.,x(w)I . Here, mlx(w)and rnk,,lx(w) are the spin-echo profiles of the rlxand rk,Ix pulses in sequences 111 and [2], and w is the chemical shift. Figure 7 compares the spectral profiles of the pulses that were used in the implementation of sequences [I] and 121. It shows that their profiles match to within 2% over a chemical shift range of t 0 . 8 pprn around the center of excitation. If the nominal flip angle of the pulses

128

Adalsteinsson et al.

FIG. 5. Right parietal stroke. (a) A slice from a &-weighted axial spin echo scan correspondingto the spectroscopic slice. Spectroscopic images based on peaks from (b) lactate, (c) NAA, and (d) lipids. (e) Unedited image, acquired with sequence [l],that contains both lactate and lipids and was created with the same spectral integrationwidth as the edited lactate image. The spectroscopic images are interpolated onto a 256 x 256 grid from the original 16 x 16 data set, and overlayed onto an edge detected version of the water image (a).

Incorporating Lactate/Lipid Discrimination

129

I

0.1

R

n

-180"

0.08

ii

0.06 0.04

0.02

6

4

2

0

-2

-4

-6 ppm

0

-2

I

1 ' 6

4

2

0

-2

-4

-6 ppm

FIG 6. Right parietal stroke. The magnitude of (a) sum- and (b) difference spectra from the three voxeis labeled in Fig. 5a.

deviates by ?lo%, the profiles will match to within 2% over a 20.5 ppm range of chemical shift. Figure 7 also indicates what the effects of Bo inhomogeneity are on the subtraction. The cancellation of lipids will be severely affected if the shifts in main magnetic field due to inhomogeneity move the lipids out of the range of chemical shifts where the spectral profiles match closely. As described earlier, the sequence only passes signals from metabolites that are within a 1.97 ppm chemical shift range from the CH, doublet of lactate. This passband is wide enough to image NAA, but leaves out signals from creatine and choline. Reference (3) describes in detail the implementation and tradeoffs involved in using a spectral-spatial spin-echo pulse whose passband includes the range from choline to lactate, and a stopband of approximately 0.5 pprn around the frequency of water. The major limiting factors in obtaining sharper transition bands of the spectral-spatial excitation are gradient slew rate (1G in 600 ps), T2effects, and evolution of coupled spins during excitation. For a fixed duration pulse, a reduction in the transition width can be achieved by choosing a minimum phase design instead of a linear phase pulse, at the cost of nonlinear phase across the chemical shift axis. Quantitative measure of metabolites, calculated as ratios relative to creatine or choline, are possible with the extension described above. Furthermore, since the ex-

-1.5

-1

-0.5 ppm

FIG. 7. Difference of spin-echo magnetization profiles. The plot shows the absolute value of the difference of spin-echo profiles of ~ 7rkx,,xpulses that were implemented in the editing sethe w , and quence. These profiles are calculated under the assumption that each pulse is preceded and followed by gradient lobes of equal area. The profiles match to within 2% over a chemical shift range of -0.9 pprn when the nominal tlip angle of the pulses is 180". Ifthe flip angles vary from 160" to 200",the pulses match to within 2% over i0.6ppm.

cited volume in the present method is a slice, it is possible to include an external standard of known concentration within the spatial field of view, and calculate absolute concentrations based on known concentration of the external standard (22). Alternatively, water as an internal reference has been used for quantitation (23,24), and could be applied in this case with a separate collection of the fully relaxed water signal. Finally, a multislice variation of the sequence is possible by adding slice selection gradients to the inversion pulse and the vh,Ix spin echo pulse, and by implement~ echo pulse as a spectral-spatial pulse. ing the T , spin In conclusion, a spectral-spatial excitation allows efficient incorporation of a subtraction based editing technique into a spectroscopic imaging sequence. With additional lipid reduction based on inversion recovery and an echo time of 136 ms, the two-shot sequence provides an edited lactate image without lipid artifacts when it is used to image human heads. In addition, data from uncoupled spins, such as NAA, are also available without SNR or time penalty compared to conventional scans. The promising results obtained, indicate that this method may be useful for regions outside the head where lipids are an even bigger problem.

ACKNOWLEDGMENTS The authors thank Dr. Dieter Enzmann for providing patients for this study.

REFERENCES 1. P. Bottomley, Ann. N. Y. Acad. Sci. 508, 333 (1987). 2. J. Frahm, H. Bruhn, M. Gyngell, K. Merboldt, W. Hanicke, R.

Sauter, Magn. Reson. Med. 9, 79 (1989).

130 3. D. M. Spielman, J. Pauly, A. Macovski, G. Glover, D. Enzmann, J. M a p . Reson. Imaging 2, 253 (1992). 4. C. Sotak, D. Freeman, J. M a p . Reson. 77, 382 (1988). 5. C. Sotak, D. Freeman, R. Hurd, J. Magn. Reson. 78, 355 (1988). 6. L. Trimble, J. Shen, A. Wilman, P. Allen, J. Magn. Reson. 86, 191 (1990). 7. R. Hurd, D. Freeman, Pro. Natl. Acad. Sci. U S A 86, 4402 (1989). 8. D. Sappey-Mariner, G. Calabrese, H. Hetherington, S. Fisher, R. Deicken, C. V. Dyke, G. Fein, M. Weiner, Magn. Reson. Med. 26, 313 (1992). 9. D. Menon, J. Ainsworth, I. Cox, R. Coker, J. Sargentoni, G.

Coutts, C. Baudouin, A. Kocsis, J. Harris, J. Comput. Assist. Tomogr. 16, 538 (1992). 10. J. Duijn, G. Matson, A. Maudsley, J. Hugg, M. Weiner, Radiology 183, 711 (1992). 11. H. Kugel, W. Heindel, R. Ernestus, J. Bunke, R. du Mesnil, G. Friedman, Radiology 183, 701 (1992). 12. C. H. Meyer, J. M. Pauly, A. Macovski, D. G. Nishimura, Magn. Reson. Med. 15,287 (1990).

Adalsteinsson et al. 13. S. R. Williams, D. G. Gadian, E. Proctor, J. M a p . Reson. 66, 562 (1986). 14. C. C. Hanstock, M. R. Bendall, H. P. Hetherington, D. P. Boisvert, P. S. Allen, J. Magn. Reson. 71,349 (1987). 15. P. Blondet, J. P. Albrand, M. von Kienlin, M. Decorps, J. M a p . Reson. 71, 342 (1987). 16. H. P. Hetherington, J. R. Hamm, J. W. Pan, D. L. Rothman, R. G. Shulman, J. Magn. Reson. 82, 86 (1989). 17. 0.W. Sorensen, G. W. Eich, R. R. Ernst, Prog. NMR Spectrosc. 16, 163 (1983). 18. J. Pauly, P. L. Roux, D. Nishimura, A. Macovski, IEEE Trans. Med. Imaging 10, 53 (1991). 19. J. Frahm, H. Bruhn, M. Gyngell, K. Merboldt, W. Hanicke, R. Sauter, Magn. Reson. Med. 11, 47 (1989). 20. P. Webb, A. Macovski, Magn. Reson. Med. 20, 113 (1991). 21. P. Bottomley, T. Foster, R. Argersinger, L. Pfeifer, Med. Phys. 11,425 (1984). 22. P. Narayana, L. Fotedar, E. Jackson, T. Bohan, I. Butler, J. Wolinsky, J. Magn. Reson. 83,44 (1989). 23. K. Thulborn, J. Ackerman, J. Magn. Reson. 53, 357 (1983). 24. P. Christiansen, 0.Henriksen, M. Stubgaard, P. Gideon, H. Larsson, Magn. Reson. Imaging 11,107 (1993).