- Email: [email protected]
A hybrid technique for spectroscopic imaging with reduced truncation artifact
Magnetic Resonance Imaging, Vol. 17, No. 3, pp. 435– 443, 1999 © 1999 Elsevier Science Inc. All rights reserved. Printed in the USA. 0730-725X/99 $–see front matter
PII S0730-725X(98)00187-8
● Original Contribution
A HYBRID TECHNIQUE FOR SPECTROSCOPIC IMAGING WITH REDUCED TRUNCATION ARTIFACT GREGORY METZGER, SHANTANU SARKAR, XIAODONG ZHANG, KEITH HEBERLEIN, MAQBOOL PATEL, AND XIAOPING HU Department of Radiology and Center for Magnetic Resonance Research, University of Minnesota Medical School, Minneapolis, MN 55345, USA Traditionally, Fourier spectroscopic imaging is associated with a small k-space coverage which leads to truncation artifacts such as “bleeding” and ringing in the resultant image. Because substantial truncation artifacts mainly arise from regions having intense signals, such as the subcutaneous lipid in the head, effective reduction of truncation artifacts can be achieved by obtaining an extended k-space coverage for these regions. In this paper, a hybrid technique which employs phase-encoded spectroscopic imaging (SI) to cover the central portion of the k-space and echo-planar spectroscopic imaging (EPSI) to measure the peripheral portion of the k-space is developed. EPSI, despite its inherently low SNR characteristics, provides a sufficient SNR for outer high-spatial frequency components of the aforementioned high signal regions and supplies an extended k-space coverage of these regions for the reduction of truncation artifacts. The data processing includes steps designed to remove inconsistency between the two types of data and a previously described technique for selectively retaining only outer k-space information for the high signal regions during the reconstruction. Experimental studies, in both phantoms and normal volunteers, demonstrate that the hybrid technique provides significant reduction in truncation artifacts. © 1999 Elsevier Science Inc. Keywords: MRS; Spectroscopic imaging; Echo-planar; Truncation artifact.
INTRODUCTION In phase-encoded spectroscopic imaging (SI)1 the spatial frequency space (also known as k-space) is sampled discretely with the sampling density determined by the desired field of view (FOV) and spatial resolution. Due to low signal-to-noise ratio (SNR) of the metabolites and the limited measurement time, only a small central portion of the k-space can be covered in a phase-encoded SI experiment. For example, a two-dimensional CSI experiment may only cover the k-space with a 16 3 16 or 32 3 32 matrix. With such limited sampling in k-space, truncation artifacts arise, leading to substantial intervoxel signal contamination or “bleeding.” To reduce truncation artifacts in SI, various methods have been developed. One common approach is to apodize in the k-space or smooth in the spatial domain,
which can be achieved either through post-processing or modified data acquisition.2,3 The filtering approach, however, is inadequate for coping with signal leakage from areas having high signal. Other techniques have utilized structural information from anatomic images to augment the reconstruction of spectroscopic images for the reduction of truncation artifacts.4 –9 For example, SLIM5 assumes that the measured volume consists of a number of homogeneous compartments that can be identified in an anatomic image and solves for compartmental spectra directly from the measured data. Anatomic information is also used in data extrapolation techniques.4,6 –9 While these techniques reduce the truncation artifact to a large extent, their effectiveness is limited by inherent assumptions in data modeling. In proton SI of the head, truncation artifacts are particularly problematic since the intense extracranial lipid
RECEIVED 7/31/98; ACCEPTED 8/30/98. Address correspondence to Dr. Xiaoping Hu, Center for MR Research, 2021 6th Street, SE, Minneapolis, MN 55455. Email: [email protected]
Work supported by the National Institutes of Health (grant NS34756). A preliminary version of this work was presented at the Sixth Annual Scientific Meeting of ISMRM in Sydney, Australia. 435
436
Magnetic Resonance Imaging ● Volume 17, Number 3, 1999
signal can detrimentally contaminate the spectroscopic signal in the brain. Several approaches for lipid signal elimination have been developed to address this particular problem. One approach is the use of spatially selective excitation techniques to excite the spins in a region well within the brain and away from the lipid;10 this approach, however, limits the extent of the available FOV, making it impossible to study the entire cerebral cortex. A recent method based on spatial saturation utilizes 8 pulses to saturate the spins in the scalp in a octagonal fashion.11 Although the saturation technique can be quite effective, it is somewhat cumbersome to use and may be limited by the RF deposition at higher fields, and it is also virtually impossible, particularly for interleaved multislice studies, to completely saturate the complex shape of the extracranial lipid with eight pulses. Taking advantage of the relatively short T1 of the lipids, others have used an inversion pulse with a proper delay to null the lipid signal;12–15 this approach can be fairly robust, but relies on the assumption that the lipid resonances have a single relaxation time, and the inversion pulse also introduces a significant reduction of signal arising from the metabolite resonances.13 Realizing that substantial signal contamination due to truncation arises mainly from areas having strong signals, a strategy for reducing contamination from these regions based on variable k-space coverage was developed.16 With this strategy, a larger region of the k-space was covered, with the low k-space points sampled using a conventional method and the high k-space data sampled using a fast approach. Hence, an extended region of the k-space was covered with a slight increase in measurement time. The rapidly sampled high k-space data, being low in SNR, contained sufficient information for the high intensity areas only. In order to reconstruct data sampled in this unconventional manner an alternative reconstruction algorithm was also developed.16 The technique was demonstrated to be very robust with onedimensional (spatial) implementation16 where the high spatial frequency components were sampled with no data averaging while the low k-space data were collected with many averages. The variable averaging approach was augmented by a variable TR approach where the high k-space was sampled with a reduced repetition time.17 Although the principle of variable k-space coverage is valid in general, the variable averaging and variable TR approaches are not sufficient in practice for multidimensional experiment where, as is typically the case, only one average is needed. To achieve variable coverage in 2-D spectroscopic imaging, a hybrid technique combining phase-encoding SI and echo-planar spectroscopic imaging18 –20 is developed in this work. Echo-planar spectroscopic imaging (EPSI) was previously described18 –20 and successfully
demonstrated recently.21,22 With its rapid sampling capability, EPSI can be used to cover a large matrix in k-space efficiently. However, this capability cannot be translated into a larger k-space coverage due to data averaging needed. In other words, it is not feasible to acquire a large matrix using EPSI alone for truncation artifact reduction because the required data averaging makes the acquisition time impractical. However, EPSI provides an efficient means for sampling the high kspace of the intense lipid signal. By combining high spatial-frequency data with those provided by high SNR phase-encoded SI, a hybrid data set can be formed which when properly reconstructed results in reduction of truncation errors. This technique is experimentally verified in both phantoms and normal volunteers. MATERIALS AND METHODS Acquisition All experiments were conducted on a Siemens 1.5 Tesla Vision imaging system (Siemens Medical Systems, Iselin, New Jersey). This system is equipped with actively shielded body gradients capable of achieving 25 mT/m with sinusoidal ramping in 300 ms. Data were acquired with a standard Siemens quadrature head coil. Two-dimensional phase-encoded spectroscopic imaging data were collected using a spin-echo sequence. In this sequence, selective 90° and 180° pulses were used to generate a spin-echo in a slice while phase-encoding steps were applied along the two in-plane dimensions. For water suppression, the spin-echo excitation was preceded by chemically selective water suppression pulses.23 The second half of the echo was sampled for a duration of 819.2 ms with 512 data points corresponding to a spectral bandwidth of 625 Hz. Experimental data presented in this paper were acquired with a TR/TE of 1.5 s/135 ms, a slice thickness of 1 cm, and a field of view of 24 3 24 cm2. The EPSI sequence used parameters identical to the phase-encoded SI sequence except that one of the spatial dimensions is encoded with an alternating read-out gradient as illustrated in Fig. 1. Each half-cycle of the oscillating read-out gradient consists of sinusoidal ramps (300 ms each) and a 200 ms plateau. Individual echoes were sampled evenly in time with 200 points which were interpolated to 128 equidistant kspace points during processing. A total of 512 echo pairs, arising from the positive and negative lobes of the readout gradient, were collected in 819.2 ms. In the following, echoes corresponding to the positive gradient lobes are referred to as odd echoes and those corresponding to the negative lobes even echoes. The hybrid technique relies on the acquisition and combination of an EPSI data set and an SI data set. For phantom and in vivo studies presented in this paper, a
A Hybrid spectroscopic imaging technique ● G. METZGER
ET AL.
437
Fig. 1. Sequence diagram for the EPSI sequence. The phase-encoded SI sequence (not shown here) is identical to the EPSI sequence except for the use of a phase-encoding gradient along the second spatial dimension. In this study, TR/TE are l.5 s/135 ms.
128 3 128 3 512 EPSI sequence and a 24 3 24 3 512 phase-encoded SI sequence were employed using parameters given above. These sequences take approximately 3.2 and 14.4 min, respectively. Despite the use of identical acquisition parameters in the two sequences, there are inconsistencies between data sets acquired with them due to inherent differences in k-space traversal and the non-ideal performance of the MR hardware. Several steps designed to remove these inconsistencies, as described in the next subsection, were taken during preprocessing of the EPSI data. Preprocessing of EPSI Data Temporal shift. Along the echo-planar encoding direction, the k-space coverage is achieved in a timedependent fashion, leading to a k-dependent temporal shift. This shift can be visualized in Fig. 2, which schematically illustrates the evolution of k with respect to time. Because such a shift is absent in the phase-encoded SI data, the time shift in the EPSI data must be removed to ensure data matching. This removal is described below assuming an EPSI data set in a single spatial dimension with no loss of generality. Spatially encoded FIDs are formed from EPSI data by taking profiles along time in the k-t space (Fig. 2). In this work, the odd echoes and the even echoes are treated separately to avoid inconsistencies due to the alternating gradients. Each FID is shifted such that it is temporally aligned with the time for k 5 0. This shift is achieved in the spectral frequency domain by the application of a linear phase modulation proportional to the time shift desired, say t(k). Specifically,
Fig. 2. A schematic representation of the traversal of the k-space during data acquisition in EPSI. Only 2 of the 512 cycles of the oscillating readout gradient are shown here. Profiles along time on the positive gradient lobe are the spatially encoded FIDs used in this study. The locations on the trajectory of one such FID are indicated by open circles and the associated temporal shift, t(k), necessary to align the FID temporally is also indicated.
Magnetic Resonance Imaging ● Volume 17, Number 3, 1999
438
s˜ ~t,k!
IFT ™™™ 3
S˜ ~ v ,k!
FT s˜ ~ v ,k! 5 S˜ ~ v ,k!e 22pt~k!vi ™™™ 3 s~t,k!
where s˜ (t,k) is the original FID, S˜ ( v,k) is the spectral frequency representation of s˜ (t,k), S(v,k) is the phasemodulated version of S˜ (v,k), and s(t,k) is the FID properly shifted. K-space interpolation. Uniform ADC (analog-to-digital converter) triggering in time during the sinusoidal gradient ramping results in non-equidistant k-space sampling. In order to match the k-space locations of EPSI data to those of the phase-encoded SI data, the EPSI data are resampled using sinc-interpolation as described previously for EPI.24 In addition, interpolating to an equidistant grid allows the use of the fast Fourier transform for spatial reconstruction along the echo-planer encoding direction. The interpolation requires the knowledge of the actual location of the measured data in the k-space. Ideally, the location of these points can be derived by integrating the programmed read-out gradient waveform. In practice, the actual gradients generated by the hardware usually deviate from that specified in the sequence. Therefore, a map of the k-space trajectory along the echo-planar encoding direction is experimentally obtained by using self-encoding.24 To reduce the influence of field inhomogeneity, the k-space trajectory is mapped twice using the two possible gradient polarities, and the final trajectory is derived from the subtraction of the two measurements. The mapped trajectory is used in the sinc-interpolation. In addition to facilitate the k-space interpolation, the measured trajectory also revealed a gradient hardware delay associated with the execution of the alternating gradient waveform. The delay, which was determined to be 4.8 ms, is compensated for in the measured data by shifting the EPSI data before further processing. Removal of temporal mismatches between phase-encoded SI and the EPSI. Although the timing of both sequences were matched exactly, there still exists a slight temporal discrepancy between the EPSI data processed with the steps described above and the SI data. This discrepancy, which is identical for all spatially encoded FIDs, arises from additional hardware delays in the EPSI sequence and gives rise to deleterious mismatches between the two data sets. This temporal difference is determined from the two data sets and removed. First, the FIDs in the phase-encoded SI data are sinc-interpolated by a factor of 32, and correlation between FIDs in the two data sets are calculated, at lags determined by the interpolated phase-encoded SI data, for each k-space point in a central 12 3 12 region. Subsequently the lag
Fig. 3. Schematic representation of the k-space coverage by the hybrid technique. The central region (C) is covered by phase-encoded SI data and the outer region (O) is covered with echo-planar encoding data.
corresponding to the maximum correlation is taken as the temporal shift for each k-space point. Shifts for the 12 3 12 region in the k-space are averaged, giving rise to a shift, typically 1.2 ms, which is applied globally to the EPSI data set. Reconstruction The first step in the reconstruction is to form a combined 128 3 128 data set from the phase-encoded SI and EPSI data such that the former occupies the central 24 3 24 matrix and the latter covers the rest of the matrix (see Fig. 3). This combined data set is low in SNR in the region filled by the EPSI data. As a result, Fourier transformation cannot be directly applied to the combined data set for spatial reconstruction because it will result in unacceptable SNR in the image space. To avoid this SNR problem, selective reconstruction16 is used. The essence of the algorithm is that it reconstructs the spectroscopic images with variable spatial resolution such that the high spatial frequency information is only retained for the areas with high signal such as the subcutaneous lipid. Briefly, this is achieved with the following steps. An initial spectroscopic image is obtained by inversely Fourier transforming the combined data set. This initial (complex) image is subsequently multiplied by a spatial mask reflecting the spatial extent of the lipid regions (determined by thresholding the spectroscopic image integrated over the lipid frequency range), and the
A Hybrid spectroscopic imaging technique ● G. METZGER
ET AL.
439
Fig. 4. A, gray scale representation of the k-space data (real part) for the first time point of the hybrid data. B, Corresponding k-space data from the EPSI sequence alone. The central 24 3 24 region (enclosed by four brackets in A) in the hybrid data contains the phase-encoded SI data whereas the rest consists of EPSI data. The absence of discontinuity in the hybrid data and the close correspondence between the hybrid data and the EPSI data indicate that a high level of data consistency is achieved. K-space profiles along the echo-planar encoding direction showing the match of the real part (C), and the imaginary part (D) between the phase-encoded SI data (broken curve) and the EPSI (solid curve) data. These profiles further demonstrate the quality of the match. Note that outside the central 24 points in the k-space, only EPSI data exist.
resultant image is Fourier transformed to form an intermediate k-space data set. Replacing the central k-space portion of the intermediate data set with the measured phase-encoded SI data produces the final k-space data set which is subsequently inversely Fourier transformed to generate the final spectroscopic image. RESULTS The quality of data matching between the phaseencoded SI data and the preprocessed EPSI data, which is a crucial step in the hybrid technique, is presented here for phantom data. In Fig. 4, gray scale images corresponding to the k-space data of the first time point are shown. The absence of discontinuity in the hybrid data (panel C) demonstrates that the two data sets are wellmatched. The quality of matching is further illustrated by k-space profiles along the EPSI direction shown in Fig. 4 where a close agreement between the real and imaginary profiles are seen, indicating a good agreement between the two types of data.
Figure 5A illustrates the structural image of the phantom which consists of a ring of cooking coil surrounding an inner tube containing an acetate/lactate solution. In Fig. 5B, a spatial-spectral slice of the spectroscopic images reconstructed from the hybrid data set with selective reconstruction is shown. For comparison, a corresponding slice from the reconstruction of the phase-encoded SI data (24 3 24) zero-padded to 128 3 128 is presented in Fig. 5C. The effectiveness of the expanded k-space coverage provided by the hybrid data set is evidenced in Fig. 5B by the elimination of the lipid signal contamination which is severe in Fig. 5C. The spatial extent of the reduction of truncation artifact can be more readily visualized in Figs. 5D and 5E where the images of the cooking oil are shown. Figure 5F presents a 1-D spatial profile at 220 Hz, the approximate location of the lipid resonance. Outside the true origin of the lipid signal, the profile from the hybrid data set (broken line) shows negligible amplitude while that from the zero-padded data (solid line) exhibits substantial artifactual signal.
440
Magnetic Resonance Imaging ● Volume 17, Number 3, 1999
Fig. 5. Results from the phantom study. A, the structural image of the phantom. B, a spatial-spectral slice of the resultant spectroscopic image of the hybrid technique; the spatial location of the slice is marked as a column in (A). C, the spatial-spectral slice, corresponding to the slice shown in (B), derived from Fourier transforming the phase-encoded SI data with zero-padding. Images created from integrating over the frequency range of cooking oil’s methyl resonance are shown for (D) the hybrid technique and (E) reconstruction of the phase-encoded SI data with zero-padding. F, one-dimensional spatial profile taken at 220 Hz from the hybrid technique (broken curve) and zero-padded FT (solid curve) of the phantom. Note that the two sides of the oil are of different heights due to B0 inhomogeneity. G, representative spectra of the phantom from the zero-padded FT (broken curve) and hybrid technique (solid curve). The hybrid reconstruction exhibits the virtual elimination of the lipid signal contamination surrounding the lactate resonances.
A Hybrid spectroscopic imaging technique ● G. METZGER
The average lipid signal contamination in the hybrid result is approximately 1/10 that in the zero-padded result. A representative spectrum from the interior of the phantom (Fig. 5G) shows that in the hybrid result, the lactate peaks are not distorted by the presence of intense artifactual lipid resonances. Note that the lactate peaks are inverted at the TE of 135 ms and the lipid contamination in the zero-padded result is also inverted due to the sinc point spread function of ringing. Similar results were obtained from in vivo studies. Figure 6A illustrates the anatomic image of one subject studied. Spatial-spectral slices of the hybrid result and zero-padded phase-encoded SI result are shown in Fig. 6B and 6C, respectively while lipid images are presented in Fig. 6D and 6E. The contamination from the lipid signal is greatly reduced in the hybrid result. In Fig. 6F, a representative spectrum in a voxel near the lipid is shown for both the hybrid technique (solid) and the zero-padded FT (dashed). The distortion of the NAA resonance by the lipid signal contamination that is seen in the zero-padded result is virtually removed in the hybrid result. DISCUSSION A hybrid technique that combines conventional phase-encoded SI with echo-planar SI for truncation artifact reduction is described in this paper. The hybrid technique is experimentally demonstrated to be effective in reducing signal contamination from areas having high signal such as the subcutaneous lipid in proton spectroscopic imaging of the head. This reduction is achieved by an expanded coverage of the k-space (by a factor of 5 in each dimension as implemented) using the EPSI acquisition with a slight increase in measurement time (3.2 min in the present study), which is acceptable for most practical experiments where the entire study often takes 1–2 h. The amount of truncation artifact reduction is determined by the size of the extended matrix and the data matching. In our implementation using an extended matrix size of 128 3 128, the reduction is approximately a factor of 10. In studies such as short TE studies where further reduction is desired, the present method can be used in conjunction with existing approaches listed in the introduction. It should be noted that an increased k-space coverage cannot be realized by simply using the echo-planar data acquisition alone because the required data averaging would make measurement time prohibitively long. An alternative approach to the present hybrid approach is to acquire the low k-space and the high k-space separately with EPSI sequences having different k-space coverage and data averaging. In that case, data matching is still needed because inconsistency between EPSI data ac-
ET AL.
441
quired with different spatial encoding will be significant. In addition, as demonstrated by others, when averaging is needed for the low k-space data, it is often more efficient to use the phase-encoded SI.26 For simplicity, gradient echoes during the positive gradient lobes are processed separately from those during the negative gradient lobes in the acquired EPSI data. The preprocessing steps described in this paper are identical for the two sets of echoes except the time shift. Alternatively, if a broader spectral bandwidth is desired an interlaced Fourier transform approach can be applied to incorporate both echoes in the spectral processing, doubling the spectral bandwidth.27 The preprocessing steps are necessary for removing inconsistencies between data obtained with the two sequences. For a predetermined experimental protocol, the preprocessing steps can be optimized with phantom experiments and standardized for subsequent studies with minimal user interaction. On an SGI Octane workstation, these steps take approximately 5 min using software written in a PV Wave environment (Visual Numerics) and can be further optimized if coded in a standard programming language such as C. Despite the careful steps taken to remove data inconsistency, the hybrid reconstruction can still be degraded by errors such as eddy current effects arising from the oscillating gradients and the difference in the analog filters in the two sequences. Consequently, the amount of truncation artifact reduction achieved experimentally can be further increased by fully accounting for these errors. Selective reconstruction was used to improve the SNR of the metabolite resonances by providing a low spatial resolution for the metabolites while retaining the high spatial resolution for the high signal areas such as the subcutaneous lipid. This approach is justified because the SNR of the metabolite signals does not permit their high spatial resolution mapping. Relative to the EPSI preprocessing steps, the computational load of selective reconstruction is negligible. CONCLUSION In this work, a hybrid technique that combines phaseencoded spectroscopic imaging with echo-planar spectroscopic imaging is described and experimentally verified. Experimental results in both phantoms and normal volunteers demonstrate that the technique is effective in reducing truncation artifacts with a minimal increase in experimental time. For proton spectroscopic imaging studies of the head the hybrid technique significantly reduces contamination of the extracranial lipid signal, making it possible to accurately detect metabolites in the brain that are near the lipid resonances. This technique is expected to be of practical utility in studies where signal
442
Magnetic Resonance Imaging ● Volume 17, Number 3, 1999
Fig. 6. Results from a normal volunteer. a, anatomic image of the axial slice studied. Spatial-spectral slices of the spectroscopic imaging data obtained with hybrid technique (B) and the Fourier transformation of the phase-encoded SI data with zero-padding (C). The spatial location of the spatial-spectral slice is shown as a horizontal bar in panel (A). Lipid images, obtained by integrating over the frequency range of the lipid resonances, derived with the hybrid technique (D) and phase-encoded SI with zero-padding (E). F, a representative spectrum (with 8-Hz line broadening) is displayed for both the hybrid technique (solid) and the zero-padded FT (dashed). The spectral plot illustrates the NAA peak free of residual lipid signal when the hybrid technique was used.
A Hybrid spectroscopic imaging technique ● G. METZGER
contamination arises from regions having high signal intensity. REFERENCES 1. Brown, T.R.; Kincaid, B.M.; Ugurbil, K. NMR chemical shift imaging in three dimensions. Biophysics 79:3523– 3526; 1982. 2. Brooker, H.R.; Mareci, T.H.; Mao, J.T. Selective Fourier transform localization. Magn. Reson. Med. 5:417– 433; 1987. 3. Garwood, M.; Robitalle, P.M.; Ugurbil, K. Fourier series windows on and off resonance using multiple coils and longitudinal modulation. J. Magn. Reson. 75:244 –260; 1987. 4. Hu, X.; Stillman, A.E. Technique for reduction of truncation artifact in chemical shift images. IEEE Trans. Med. Imaging 10:290 –294; 1991. 5. Hu, X.; Levin, D.N.; Lauterbur, P.C.; Spraggins, T.A. SLIM: Spectral localization by imaging. Magn. Reson. Med. 8:314 –322; 1988. 6. Patel, M.A.S.; Hu, X. A robust algorithm for reduction of truncation artifact in chemical shift images. IEEE Trans. Med. Imaging 12:812– 818; 1993. 7. Patel, M.S.; Hu, X. Selective data extrapolation for chemical shift imaging. In: Book of Abstracts: Second Annual Meeting of the Society for Magnetic Resonance in Medicine, vol 1. San Francisco, CA: SMRM, 1994: p. 1168. 8. Plevritis, S.K., Resolution Improvements for Magnetic Resonance Spectroscopic Images, Ph. D. thesis, Stanford University, 1992. 9. Liang, Z.P.; Lauterbur, P.C. A generalized series approach to MR spectroscopic imaging. IEEE Trans Med Imaging 10:132–137; 1991. 10. Luyten, P.R.; Marien, A.J.H.; Heindel, W.; van Gerwen, P.H.J.; Herholz, K.; den Hollander, J.A.; Friedmann, G.; Heiss, W.-D. Metabolic imaging of patients with intracranial tumors: H-1 MR spectroscopic imaging and PET. Radiology 176:791–799; 1990. 11. Duijn, J.H.; Matson, G.B.; Maudsley, A.A.; Weiner, M.W. 3D Phase encoding 1H spectroscopic imaging of human brain. Magn. Reson. Imaging 10:315–319; 1992. 12. Hetherington, H.P.; Pan, J.W.; Mason, G.F.; Ponder, S.L.; Twieg, D.B.; Deutsch, G.; Mountz, J.; Pohost, G.M. 2D 1H Spectrosocopic imaging of the human brain at 4.l T. Magn. Reson. Med. 32:530 –534; 1994. 13. Spielman, D.M.; Pauly, J.M.; Macovski, A.; Glover, G.H.; Enzmann, D.R. Lipid-suppressed single- and multisection proton spectroscopic imaging of the human brain. J. Magn. Reson. Imaging 2:253–262; 1992.
ET AL.
443
14. Hetherington, H.P.; Mason, G.F.; Pan, J.W.; Ponder, S.L.; Vaughan, J.T.; Twieg, D.B.; Pohost, G.M. Evaluation of cerebral gray and white matter metabolite differences by spectrscopic imaging at 4.1T. Magn. Reson. Med. 32:565– 571; 1994. 15. Hetherington, H.P.; Pan, J.W.; Mason, G.F.; Adams, D.; Vaughn, M.J.; Twieg, D.B.; Pohost, G.M. Quantitative 1H spectroscopic imaging of human brain at 4.1 T using image segmentation. Magn. Reson. Med. 36:21–29; 1996. 16. Hu, X.; Patel, M.; Ugurbil, K. A new strategy for spectroscopic imaging. J. Magn. Res. (Series B) 103:30 –38; 1994. 17. Hu, X.; Patel, M.S.; Chen, W.; Ugurbil, K. Reduction of truncation artifact in CSI with variable TR. J. Magn. Reson. (Series B) 106:292–296; 1995. 18. Mansfield, P. Spatial mapping of chemical shift in NMR. Magn. Reson. Med. 1:370 –386; 1984. 19. Matsui, S.; Sekihara, K.; Kohno, H. High-speed spatially resolved NMR spectroscopy using phase-modulated spinecho trains: Expansion of the spectral bandwidth by combined use of delayed spin-echo trains. J. Magn. Reson. 64:167–171; 1985. 20. Matsui, S.; Sekihara, K.; Kohno, H. Spatially resolved NMR spectroscopy using phase-modulated spin-echo trains. J. Magn. Reson. 67:476 – 490; 1986. 21. Posse, S.; DeCarli, C.; Le Bihan, D. Three-dimensional echo-planar MR spectroscopic imaging at short echo times in the human brain. Radiology 192:733–738; 1994. 22. Posse, S.; Tedeschi, G.; Risinger, R.; Ogg, R.; Le Bihan, D. High speed 1H spectroscopic imaging in human brain by echo planar spatial-spectral encoding. Magn. Reson. Med. 33:34 – 40; 1995. 23. Ogg, R. J.; Kingsley, P.; Taylor, J. WET: A T1 and B1 insensitive water suppression method for in vivo localized 1 H NMR spectroscopy. J. Magn. Reson. (Series B) 104:1– 10; 1994. 24. Bruder, H.; Fischer, H.; Reinfelder, H.E.; Schmitt, F. Image reconstruction for echo planar imaging with nonequivalent k-space sampling. Magn. Reson. Med. 23:311– 323; 1992. 25. Takahashi, A.; Peters, T. Compensation of multi-dimensional selective excitation pulses using measured k-space trajectories. Magn. Reson. Med. 34:446 – 456; 1995. 26. Pohmann, R.; von Kienlin, M. Theoretical evaluation and comparison of fast chemical shift imaging methods. In: Book of Abstracts: Fifth Annual Meeting, vol 1. Vancouver: SMRM, 1997: p. 101. 27. Metzger, G.; Hu, X. Application of interlaced fourier transform to echo-planar spectroscopic imaging. J. Magn. Reson. 125:166 –170; 1997.
PII S0730-725X(98)00187-8
● Original Contribution
A HYBRID TECHNIQUE FOR SPECTROSCOPIC IMAGING WITH REDUCED TRUNCATION ARTIFACT GREGORY METZGER, SHANTANU SARKAR, XIAODONG ZHANG, KEITH HEBERLEIN, MAQBOOL PATEL, AND XIAOPING HU Department of Radiology and Center for Magnetic Resonance Research, University of Minnesota Medical School, Minneapolis, MN 55345, USA Traditionally, Fourier spectroscopic imaging is associated with a small k-space coverage which leads to truncation artifacts such as “bleeding” and ringing in the resultant image. Because substantial truncation artifacts mainly arise from regions having intense signals, such as the subcutaneous lipid in the head, effective reduction of truncation artifacts can be achieved by obtaining an extended k-space coverage for these regions. In this paper, a hybrid technique which employs phase-encoded spectroscopic imaging (SI) to cover the central portion of the k-space and echo-planar spectroscopic imaging (EPSI) to measure the peripheral portion of the k-space is developed. EPSI, despite its inherently low SNR characteristics, provides a sufficient SNR for outer high-spatial frequency components of the aforementioned high signal regions and supplies an extended k-space coverage of these regions for the reduction of truncation artifacts. The data processing includes steps designed to remove inconsistency between the two types of data and a previously described technique for selectively retaining only outer k-space information for the high signal regions during the reconstruction. Experimental studies, in both phantoms and normal volunteers, demonstrate that the hybrid technique provides significant reduction in truncation artifacts. © 1999 Elsevier Science Inc. Keywords: MRS; Spectroscopic imaging; Echo-planar; Truncation artifact.
INTRODUCTION In phase-encoded spectroscopic imaging (SI)1 the spatial frequency space (also known as k-space) is sampled discretely with the sampling density determined by the desired field of view (FOV) and spatial resolution. Due to low signal-to-noise ratio (SNR) of the metabolites and the limited measurement time, only a small central portion of the k-space can be covered in a phase-encoded SI experiment. For example, a two-dimensional CSI experiment may only cover the k-space with a 16 3 16 or 32 3 32 matrix. With such limited sampling in k-space, truncation artifacts arise, leading to substantial intervoxel signal contamination or “bleeding.” To reduce truncation artifacts in SI, various methods have been developed. One common approach is to apodize in the k-space or smooth in the spatial domain,
which can be achieved either through post-processing or modified data acquisition.2,3 The filtering approach, however, is inadequate for coping with signal leakage from areas having high signal. Other techniques have utilized structural information from anatomic images to augment the reconstruction of spectroscopic images for the reduction of truncation artifacts.4 –9 For example, SLIM5 assumes that the measured volume consists of a number of homogeneous compartments that can be identified in an anatomic image and solves for compartmental spectra directly from the measured data. Anatomic information is also used in data extrapolation techniques.4,6 –9 While these techniques reduce the truncation artifact to a large extent, their effectiveness is limited by inherent assumptions in data modeling. In proton SI of the head, truncation artifacts are particularly problematic since the intense extracranial lipid
RECEIVED 7/31/98; ACCEPTED 8/30/98. Address correspondence to Dr. Xiaoping Hu, Center for MR Research, 2021 6th Street, SE, Minneapolis, MN 55455. Email: [email protected]
Work supported by the National Institutes of Health (grant NS34756). A preliminary version of this work was presented at the Sixth Annual Scientific Meeting of ISMRM in Sydney, Australia. 435
436
Magnetic Resonance Imaging ● Volume 17, Number 3, 1999
signal can detrimentally contaminate the spectroscopic signal in the brain. Several approaches for lipid signal elimination have been developed to address this particular problem. One approach is the use of spatially selective excitation techniques to excite the spins in a region well within the brain and away from the lipid;10 this approach, however, limits the extent of the available FOV, making it impossible to study the entire cerebral cortex. A recent method based on spatial saturation utilizes 8 pulses to saturate the spins in the scalp in a octagonal fashion.11 Although the saturation technique can be quite effective, it is somewhat cumbersome to use and may be limited by the RF deposition at higher fields, and it is also virtually impossible, particularly for interleaved multislice studies, to completely saturate the complex shape of the extracranial lipid with eight pulses. Taking advantage of the relatively short T1 of the lipids, others have used an inversion pulse with a proper delay to null the lipid signal;12–15 this approach can be fairly robust, but relies on the assumption that the lipid resonances have a single relaxation time, and the inversion pulse also introduces a significant reduction of signal arising from the metabolite resonances.13 Realizing that substantial signal contamination due to truncation arises mainly from areas having strong signals, a strategy for reducing contamination from these regions based on variable k-space coverage was developed.16 With this strategy, a larger region of the k-space was covered, with the low k-space points sampled using a conventional method and the high k-space data sampled using a fast approach. Hence, an extended region of the k-space was covered with a slight increase in measurement time. The rapidly sampled high k-space data, being low in SNR, contained sufficient information for the high intensity areas only. In order to reconstruct data sampled in this unconventional manner an alternative reconstruction algorithm was also developed.16 The technique was demonstrated to be very robust with onedimensional (spatial) implementation16 where the high spatial frequency components were sampled with no data averaging while the low k-space data were collected with many averages. The variable averaging approach was augmented by a variable TR approach where the high k-space was sampled with a reduced repetition time.17 Although the principle of variable k-space coverage is valid in general, the variable averaging and variable TR approaches are not sufficient in practice for multidimensional experiment where, as is typically the case, only one average is needed. To achieve variable coverage in 2-D spectroscopic imaging, a hybrid technique combining phase-encoding SI and echo-planar spectroscopic imaging18 –20 is developed in this work. Echo-planar spectroscopic imaging (EPSI) was previously described18 –20 and successfully
demonstrated recently.21,22 With its rapid sampling capability, EPSI can be used to cover a large matrix in k-space efficiently. However, this capability cannot be translated into a larger k-space coverage due to data averaging needed. In other words, it is not feasible to acquire a large matrix using EPSI alone for truncation artifact reduction because the required data averaging makes the acquisition time impractical. However, EPSI provides an efficient means for sampling the high kspace of the intense lipid signal. By combining high spatial-frequency data with those provided by high SNR phase-encoded SI, a hybrid data set can be formed which when properly reconstructed results in reduction of truncation errors. This technique is experimentally verified in both phantoms and normal volunteers. MATERIALS AND METHODS Acquisition All experiments were conducted on a Siemens 1.5 Tesla Vision imaging system (Siemens Medical Systems, Iselin, New Jersey). This system is equipped with actively shielded body gradients capable of achieving 25 mT/m with sinusoidal ramping in 300 ms. Data were acquired with a standard Siemens quadrature head coil. Two-dimensional phase-encoded spectroscopic imaging data were collected using a spin-echo sequence. In this sequence, selective 90° and 180° pulses were used to generate a spin-echo in a slice while phase-encoding steps were applied along the two in-plane dimensions. For water suppression, the spin-echo excitation was preceded by chemically selective water suppression pulses.23 The second half of the echo was sampled for a duration of 819.2 ms with 512 data points corresponding to a spectral bandwidth of 625 Hz. Experimental data presented in this paper were acquired with a TR/TE of 1.5 s/135 ms, a slice thickness of 1 cm, and a field of view of 24 3 24 cm2. The EPSI sequence used parameters identical to the phase-encoded SI sequence except that one of the spatial dimensions is encoded with an alternating read-out gradient as illustrated in Fig. 1. Each half-cycle of the oscillating read-out gradient consists of sinusoidal ramps (300 ms each) and a 200 ms plateau. Individual echoes were sampled evenly in time with 200 points which were interpolated to 128 equidistant kspace points during processing. A total of 512 echo pairs, arising from the positive and negative lobes of the readout gradient, were collected in 819.2 ms. In the following, echoes corresponding to the positive gradient lobes are referred to as odd echoes and those corresponding to the negative lobes even echoes. The hybrid technique relies on the acquisition and combination of an EPSI data set and an SI data set. For phantom and in vivo studies presented in this paper, a
A Hybrid spectroscopic imaging technique ● G. METZGER
ET AL.
437
Fig. 1. Sequence diagram for the EPSI sequence. The phase-encoded SI sequence (not shown here) is identical to the EPSI sequence except for the use of a phase-encoding gradient along the second spatial dimension. In this study, TR/TE are l.5 s/135 ms.
128 3 128 3 512 EPSI sequence and a 24 3 24 3 512 phase-encoded SI sequence were employed using parameters given above. These sequences take approximately 3.2 and 14.4 min, respectively. Despite the use of identical acquisition parameters in the two sequences, there are inconsistencies between data sets acquired with them due to inherent differences in k-space traversal and the non-ideal performance of the MR hardware. Several steps designed to remove these inconsistencies, as described in the next subsection, were taken during preprocessing of the EPSI data. Preprocessing of EPSI Data Temporal shift. Along the echo-planar encoding direction, the k-space coverage is achieved in a timedependent fashion, leading to a k-dependent temporal shift. This shift can be visualized in Fig. 2, which schematically illustrates the evolution of k with respect to time. Because such a shift is absent in the phase-encoded SI data, the time shift in the EPSI data must be removed to ensure data matching. This removal is described below assuming an EPSI data set in a single spatial dimension with no loss of generality. Spatially encoded FIDs are formed from EPSI data by taking profiles along time in the k-t space (Fig. 2). In this work, the odd echoes and the even echoes are treated separately to avoid inconsistencies due to the alternating gradients. Each FID is shifted such that it is temporally aligned with the time for k 5 0. This shift is achieved in the spectral frequency domain by the application of a linear phase modulation proportional to the time shift desired, say t(k). Specifically,
Fig. 2. A schematic representation of the traversal of the k-space during data acquisition in EPSI. Only 2 of the 512 cycles of the oscillating readout gradient are shown here. Profiles along time on the positive gradient lobe are the spatially encoded FIDs used in this study. The locations on the trajectory of one such FID are indicated by open circles and the associated temporal shift, t(k), necessary to align the FID temporally is also indicated.
Magnetic Resonance Imaging ● Volume 17, Number 3, 1999
438
s˜ ~t,k!
IFT ™™™ 3
S˜ ~ v ,k!
FT s˜ ~ v ,k! 5 S˜ ~ v ,k!e 22pt~k!vi ™™™ 3 s~t,k!
where s˜ (t,k) is the original FID, S˜ ( v,k) is the spectral frequency representation of s˜ (t,k), S(v,k) is the phasemodulated version of S˜ (v,k), and s(t,k) is the FID properly shifted. K-space interpolation. Uniform ADC (analog-to-digital converter) triggering in time during the sinusoidal gradient ramping results in non-equidistant k-space sampling. In order to match the k-space locations of EPSI data to those of the phase-encoded SI data, the EPSI data are resampled using sinc-interpolation as described previously for EPI.24 In addition, interpolating to an equidistant grid allows the use of the fast Fourier transform for spatial reconstruction along the echo-planer encoding direction. The interpolation requires the knowledge of the actual location of the measured data in the k-space. Ideally, the location of these points can be derived by integrating the programmed read-out gradient waveform. In practice, the actual gradients generated by the hardware usually deviate from that specified in the sequence. Therefore, a map of the k-space trajectory along the echo-planar encoding direction is experimentally obtained by using self-encoding.24 To reduce the influence of field inhomogeneity, the k-space trajectory is mapped twice using the two possible gradient polarities, and the final trajectory is derived from the subtraction of the two measurements. The mapped trajectory is used in the sinc-interpolation. In addition to facilitate the k-space interpolation, the measured trajectory also revealed a gradient hardware delay associated with the execution of the alternating gradient waveform. The delay, which was determined to be 4.8 ms, is compensated for in the measured data by shifting the EPSI data before further processing. Removal of temporal mismatches between phase-encoded SI and the EPSI. Although the timing of both sequences were matched exactly, there still exists a slight temporal discrepancy between the EPSI data processed with the steps described above and the SI data. This discrepancy, which is identical for all spatially encoded FIDs, arises from additional hardware delays in the EPSI sequence and gives rise to deleterious mismatches between the two data sets. This temporal difference is determined from the two data sets and removed. First, the FIDs in the phase-encoded SI data are sinc-interpolated by a factor of 32, and correlation between FIDs in the two data sets are calculated, at lags determined by the interpolated phase-encoded SI data, for each k-space point in a central 12 3 12 region. Subsequently the lag
Fig. 3. Schematic representation of the k-space coverage by the hybrid technique. The central region (C) is covered by phase-encoded SI data and the outer region (O) is covered with echo-planar encoding data.
corresponding to the maximum correlation is taken as the temporal shift for each k-space point. Shifts for the 12 3 12 region in the k-space are averaged, giving rise to a shift, typically 1.2 ms, which is applied globally to the EPSI data set. Reconstruction The first step in the reconstruction is to form a combined 128 3 128 data set from the phase-encoded SI and EPSI data such that the former occupies the central 24 3 24 matrix and the latter covers the rest of the matrix (see Fig. 3). This combined data set is low in SNR in the region filled by the EPSI data. As a result, Fourier transformation cannot be directly applied to the combined data set for spatial reconstruction because it will result in unacceptable SNR in the image space. To avoid this SNR problem, selective reconstruction16 is used. The essence of the algorithm is that it reconstructs the spectroscopic images with variable spatial resolution such that the high spatial frequency information is only retained for the areas with high signal such as the subcutaneous lipid. Briefly, this is achieved with the following steps. An initial spectroscopic image is obtained by inversely Fourier transforming the combined data set. This initial (complex) image is subsequently multiplied by a spatial mask reflecting the spatial extent of the lipid regions (determined by thresholding the spectroscopic image integrated over the lipid frequency range), and the
A Hybrid spectroscopic imaging technique ● G. METZGER
ET AL.
439
Fig. 4. A, gray scale representation of the k-space data (real part) for the first time point of the hybrid data. B, Corresponding k-space data from the EPSI sequence alone. The central 24 3 24 region (enclosed by four brackets in A) in the hybrid data contains the phase-encoded SI data whereas the rest consists of EPSI data. The absence of discontinuity in the hybrid data and the close correspondence between the hybrid data and the EPSI data indicate that a high level of data consistency is achieved. K-space profiles along the echo-planar encoding direction showing the match of the real part (C), and the imaginary part (D) between the phase-encoded SI data (broken curve) and the EPSI (solid curve) data. These profiles further demonstrate the quality of the match. Note that outside the central 24 points in the k-space, only EPSI data exist.
resultant image is Fourier transformed to form an intermediate k-space data set. Replacing the central k-space portion of the intermediate data set with the measured phase-encoded SI data produces the final k-space data set which is subsequently inversely Fourier transformed to generate the final spectroscopic image. RESULTS The quality of data matching between the phaseencoded SI data and the preprocessed EPSI data, which is a crucial step in the hybrid technique, is presented here for phantom data. In Fig. 4, gray scale images corresponding to the k-space data of the first time point are shown. The absence of discontinuity in the hybrid data (panel C) demonstrates that the two data sets are wellmatched. The quality of matching is further illustrated by k-space profiles along the EPSI direction shown in Fig. 4 where a close agreement between the real and imaginary profiles are seen, indicating a good agreement between the two types of data.
Figure 5A illustrates the structural image of the phantom which consists of a ring of cooking coil surrounding an inner tube containing an acetate/lactate solution. In Fig. 5B, a spatial-spectral slice of the spectroscopic images reconstructed from the hybrid data set with selective reconstruction is shown. For comparison, a corresponding slice from the reconstruction of the phase-encoded SI data (24 3 24) zero-padded to 128 3 128 is presented in Fig. 5C. The effectiveness of the expanded k-space coverage provided by the hybrid data set is evidenced in Fig. 5B by the elimination of the lipid signal contamination which is severe in Fig. 5C. The spatial extent of the reduction of truncation artifact can be more readily visualized in Figs. 5D and 5E where the images of the cooking oil are shown. Figure 5F presents a 1-D spatial profile at 220 Hz, the approximate location of the lipid resonance. Outside the true origin of the lipid signal, the profile from the hybrid data set (broken line) shows negligible amplitude while that from the zero-padded data (solid line) exhibits substantial artifactual signal.
440
Magnetic Resonance Imaging ● Volume 17, Number 3, 1999
Fig. 5. Results from the phantom study. A, the structural image of the phantom. B, a spatial-spectral slice of the resultant spectroscopic image of the hybrid technique; the spatial location of the slice is marked as a column in (A). C, the spatial-spectral slice, corresponding to the slice shown in (B), derived from Fourier transforming the phase-encoded SI data with zero-padding. Images created from integrating over the frequency range of cooking oil’s methyl resonance are shown for (D) the hybrid technique and (E) reconstruction of the phase-encoded SI data with zero-padding. F, one-dimensional spatial profile taken at 220 Hz from the hybrid technique (broken curve) and zero-padded FT (solid curve) of the phantom. Note that the two sides of the oil are of different heights due to B0 inhomogeneity. G, representative spectra of the phantom from the zero-padded FT (broken curve) and hybrid technique (solid curve). The hybrid reconstruction exhibits the virtual elimination of the lipid signal contamination surrounding the lactate resonances.
A Hybrid spectroscopic imaging technique ● G. METZGER
The average lipid signal contamination in the hybrid result is approximately 1/10 that in the zero-padded result. A representative spectrum from the interior of the phantom (Fig. 5G) shows that in the hybrid result, the lactate peaks are not distorted by the presence of intense artifactual lipid resonances. Note that the lactate peaks are inverted at the TE of 135 ms and the lipid contamination in the zero-padded result is also inverted due to the sinc point spread function of ringing. Similar results were obtained from in vivo studies. Figure 6A illustrates the anatomic image of one subject studied. Spatial-spectral slices of the hybrid result and zero-padded phase-encoded SI result are shown in Fig. 6B and 6C, respectively while lipid images are presented in Fig. 6D and 6E. The contamination from the lipid signal is greatly reduced in the hybrid result. In Fig. 6F, a representative spectrum in a voxel near the lipid is shown for both the hybrid technique (solid) and the zero-padded FT (dashed). The distortion of the NAA resonance by the lipid signal contamination that is seen in the zero-padded result is virtually removed in the hybrid result. DISCUSSION A hybrid technique that combines conventional phase-encoded SI with echo-planar SI for truncation artifact reduction is described in this paper. The hybrid technique is experimentally demonstrated to be effective in reducing signal contamination from areas having high signal such as the subcutaneous lipid in proton spectroscopic imaging of the head. This reduction is achieved by an expanded coverage of the k-space (by a factor of 5 in each dimension as implemented) using the EPSI acquisition with a slight increase in measurement time (3.2 min in the present study), which is acceptable for most practical experiments where the entire study often takes 1–2 h. The amount of truncation artifact reduction is determined by the size of the extended matrix and the data matching. In our implementation using an extended matrix size of 128 3 128, the reduction is approximately a factor of 10. In studies such as short TE studies where further reduction is desired, the present method can be used in conjunction with existing approaches listed in the introduction. It should be noted that an increased k-space coverage cannot be realized by simply using the echo-planar data acquisition alone because the required data averaging would make measurement time prohibitively long. An alternative approach to the present hybrid approach is to acquire the low k-space and the high k-space separately with EPSI sequences having different k-space coverage and data averaging. In that case, data matching is still needed because inconsistency between EPSI data ac-
ET AL.
441
quired with different spatial encoding will be significant. In addition, as demonstrated by others, when averaging is needed for the low k-space data, it is often more efficient to use the phase-encoded SI.26 For simplicity, gradient echoes during the positive gradient lobes are processed separately from those during the negative gradient lobes in the acquired EPSI data. The preprocessing steps described in this paper are identical for the two sets of echoes except the time shift. Alternatively, if a broader spectral bandwidth is desired an interlaced Fourier transform approach can be applied to incorporate both echoes in the spectral processing, doubling the spectral bandwidth.27 The preprocessing steps are necessary for removing inconsistencies between data obtained with the two sequences. For a predetermined experimental protocol, the preprocessing steps can be optimized with phantom experiments and standardized for subsequent studies with minimal user interaction. On an SGI Octane workstation, these steps take approximately 5 min using software written in a PV Wave environment (Visual Numerics) and can be further optimized if coded in a standard programming language such as C. Despite the careful steps taken to remove data inconsistency, the hybrid reconstruction can still be degraded by errors such as eddy current effects arising from the oscillating gradients and the difference in the analog filters in the two sequences. Consequently, the amount of truncation artifact reduction achieved experimentally can be further increased by fully accounting for these errors. Selective reconstruction was used to improve the SNR of the metabolite resonances by providing a low spatial resolution for the metabolites while retaining the high spatial resolution for the high signal areas such as the subcutaneous lipid. This approach is justified because the SNR of the metabolite signals does not permit their high spatial resolution mapping. Relative to the EPSI preprocessing steps, the computational load of selective reconstruction is negligible. CONCLUSION In this work, a hybrid technique that combines phaseencoded spectroscopic imaging with echo-planar spectroscopic imaging is described and experimentally verified. Experimental results in both phantoms and normal volunteers demonstrate that the technique is effective in reducing truncation artifacts with a minimal increase in experimental time. For proton spectroscopic imaging studies of the head the hybrid technique significantly reduces contamination of the extracranial lipid signal, making it possible to accurately detect metabolites in the brain that are near the lipid resonances. This technique is expected to be of practical utility in studies where signal
442
Magnetic Resonance Imaging ● Volume 17, Number 3, 1999
Fig. 6. Results from a normal volunteer. a, anatomic image of the axial slice studied. Spatial-spectral slices of the spectroscopic imaging data obtained with hybrid technique (B) and the Fourier transformation of the phase-encoded SI data with zero-padding (C). The spatial location of the spatial-spectral slice is shown as a horizontal bar in panel (A). Lipid images, obtained by integrating over the frequency range of the lipid resonances, derived with the hybrid technique (D) and phase-encoded SI with zero-padding (E). F, a representative spectrum (with 8-Hz line broadening) is displayed for both the hybrid technique (solid) and the zero-padded FT (dashed). The spectral plot illustrates the NAA peak free of residual lipid signal when the hybrid technique was used.
A Hybrid spectroscopic imaging technique ● G. METZGER
contamination arises from regions having high signal intensity. REFERENCES 1. Brown, T.R.; Kincaid, B.M.; Ugurbil, K. NMR chemical shift imaging in three dimensions. Biophysics 79:3523– 3526; 1982. 2. Brooker, H.R.; Mareci, T.H.; Mao, J.T. Selective Fourier transform localization. Magn. Reson. Med. 5:417– 433; 1987. 3. Garwood, M.; Robitalle, P.M.; Ugurbil, K. Fourier series windows on and off resonance using multiple coils and longitudinal modulation. J. Magn. Reson. 75:244 –260; 1987. 4. Hu, X.; Stillman, A.E. Technique for reduction of truncation artifact in chemical shift images. IEEE Trans. Med. Imaging 10:290 –294; 1991. 5. Hu, X.; Levin, D.N.; Lauterbur, P.C.; Spraggins, T.A. SLIM: Spectral localization by imaging. Magn. Reson. Med. 8:314 –322; 1988. 6. Patel, M.A.S.; Hu, X. A robust algorithm for reduction of truncation artifact in chemical shift images. IEEE Trans. Med. Imaging 12:812– 818; 1993. 7. Patel, M.S.; Hu, X. Selective data extrapolation for chemical shift imaging. In: Book of Abstracts: Second Annual Meeting of the Society for Magnetic Resonance in Medicine, vol 1. San Francisco, CA: SMRM, 1994: p. 1168. 8. Plevritis, S.K., Resolution Improvements for Magnetic Resonance Spectroscopic Images, Ph. D. thesis, Stanford University, 1992. 9. Liang, Z.P.; Lauterbur, P.C. A generalized series approach to MR spectroscopic imaging. IEEE Trans Med Imaging 10:132–137; 1991. 10. Luyten, P.R.; Marien, A.J.H.; Heindel, W.; van Gerwen, P.H.J.; Herholz, K.; den Hollander, J.A.; Friedmann, G.; Heiss, W.-D. Metabolic imaging of patients with intracranial tumors: H-1 MR spectroscopic imaging and PET. Radiology 176:791–799; 1990. 11. Duijn, J.H.; Matson, G.B.; Maudsley, A.A.; Weiner, M.W. 3D Phase encoding 1H spectroscopic imaging of human brain. Magn. Reson. Imaging 10:315–319; 1992. 12. Hetherington, H.P.; Pan, J.W.; Mason, G.F.; Ponder, S.L.; Twieg, D.B.; Deutsch, G.; Mountz, J.; Pohost, G.M. 2D 1H Spectrosocopic imaging of the human brain at 4.l T. Magn. Reson. Med. 32:530 –534; 1994. 13. Spielman, D.M.; Pauly, J.M.; Macovski, A.; Glover, G.H.; Enzmann, D.R. Lipid-suppressed single- and multisection proton spectroscopic imaging of the human brain. J. Magn. Reson. Imaging 2:253–262; 1992.
ET AL.
443
14. Hetherington, H.P.; Mason, G.F.; Pan, J.W.; Ponder, S.L.; Vaughan, J.T.; Twieg, D.B.; Pohost, G.M. Evaluation of cerebral gray and white matter metabolite differences by spectrscopic imaging at 4.1T. Magn. Reson. Med. 32:565– 571; 1994. 15. Hetherington, H.P.; Pan, J.W.; Mason, G.F.; Adams, D.; Vaughn, M.J.; Twieg, D.B.; Pohost, G.M. Quantitative 1H spectroscopic imaging of human brain at 4.1 T using image segmentation. Magn. Reson. Med. 36:21–29; 1996. 16. Hu, X.; Patel, M.; Ugurbil, K. A new strategy for spectroscopic imaging. J. Magn. Res. (Series B) 103:30 –38; 1994. 17. Hu, X.; Patel, M.S.; Chen, W.; Ugurbil, K. Reduction of truncation artifact in CSI with variable TR. J. Magn. Reson. (Series B) 106:292–296; 1995. 18. Mansfield, P. Spatial mapping of chemical shift in NMR. Magn. Reson. Med. 1:370 –386; 1984. 19. Matsui, S.; Sekihara, K.; Kohno, H. High-speed spatially resolved NMR spectroscopy using phase-modulated spinecho trains: Expansion of the spectral bandwidth by combined use of delayed spin-echo trains. J. Magn. Reson. 64:167–171; 1985. 20. Matsui, S.; Sekihara, K.; Kohno, H. Spatially resolved NMR spectroscopy using phase-modulated spin-echo trains. J. Magn. Reson. 67:476 – 490; 1986. 21. Posse, S.; DeCarli, C.; Le Bihan, D. Three-dimensional echo-planar MR spectroscopic imaging at short echo times in the human brain. Radiology 192:733–738; 1994. 22. Posse, S.; Tedeschi, G.; Risinger, R.; Ogg, R.; Le Bihan, D. High speed 1H spectroscopic imaging in human brain by echo planar spatial-spectral encoding. Magn. Reson. Med. 33:34 – 40; 1995. 23. Ogg, R. J.; Kingsley, P.; Taylor, J. WET: A T1 and B1 insensitive water suppression method for in vivo localized 1 H NMR spectroscopy. J. Magn. Reson. (Series B) 104:1– 10; 1994. 24. Bruder, H.; Fischer, H.; Reinfelder, H.E.; Schmitt, F. Image reconstruction for echo planar imaging with nonequivalent k-space sampling. Magn. Reson. Med. 23:311– 323; 1992. 25. Takahashi, A.; Peters, T. Compensation of multi-dimensional selective excitation pulses using measured k-space trajectories. Magn. Reson. Med. 34:446 – 456; 1995. 26. Pohmann, R.; von Kienlin, M. Theoretical evaluation and comparison of fast chemical shift imaging methods. In: Book of Abstracts: Fifth Annual Meeting, vol 1. Vancouver: SMRM, 1997: p. 101. 27. Metzger, G.; Hu, X. Application of interlaced fourier transform to echo-planar spectroscopic imaging. J. Magn. Reson. 125:166 –170; 1997.