MULTINUCLEAR IN VIVO CHEMICAL SHIFT (METABOLIC) IMAGING IN HUMANS

The possible clinical applications of nuclear magnetic resonance spectroscopy (MRS) are extensive and range over many disciplines. In the past few years, many studies have suggested the utility of in vivo spectroscopy for human diseases such as cancer, epilepsy, Alzheimer's disease, multiple sclerosis, and stroke. Since ³¹P, ¹⁹F and ¹H MRS provide different kinds of metabolic information, multiple nuclei examinations may be clinically more useful than those of a single nucleus. The aim of our research is to provide methods that will enable measurements of more than one nucleus in a single session, and to improve on the localization methods, making examinations shorter and clinically more robust.

THREE-DIMENSIONAL PROTON LOCALIZED SPECTROSCOPY IN THE HUMAN BRAIN. GONEN, in collaboration with ARIAS-MENDOZA,* STOYANOVA,* GOELMAN^a

Localized, multivoxel ¹H spectroscopy of the human brain in vivo is currently achieved by suppressing the subcutaneous fat at the rim of a selectively excited thin axial slice, followed by 16×16 chemical-shift-imaging (CSI) to localized spectra in the plane of the slice. If spectra from a larger volume are desired, acquisition of several such slices must be interleaved. Unfortunately, due to the long T₂'s of brain metabolites (~500 ms), the recycle time (TR) of each slice is ~1 s. Thus, when N (typically 4) slices are interleaved, each is examined only once every NxTR, much longer than the T₁s of the metabolites.

To improve the efficiency of acquiring spectra, we developed three-dimensional (3D) coverage of the volume-of-interest (VOI) that excites and observes the spins in the entire volume at each acquisition. Therefore, N^1/2 signal-to-noise ratio (SNR) gain per given measurement time is expected over the current method of sequential interleaving N two-dimensional slices. A hybrid of CSI and transverse Hadamard spectroscopic imaging (HSI) was used to obtain 3D, multivoxel arrays of voxels in a Siemens SP63 imager in ~25 min. Water suppression was followed by outer-volume-suppression of the fat signals and selective-excitation of an axial, "slab-like" VOI (Figure 1a). The spatially-selective HSI-encoding radio frequency (RF) pulses incorporate naturally into the PRESS double spin-echo sequence used to excite the VOI selectively. This VOI accommodates few partitions along its short axis (hence, the use of HSI) and many partitions along the other long axes (hence, the use of CSI) in the hybrid.

FIGURE 1. a) Gradient and RF waveforms. Note the 90° of the PRESS is also a 4th order HSI encoding pulse along the Z direction. Bottom: ¹H spectra from 1 ml voxels from the slice marked by the arrow in the image; b) from 3D 2D-CSI/1D-HSI hybrid, c) from the overlapping 2D interleaved slice. Both procedures were 28 min.

3D hybrid spectroscopy was performed on a 42-year-old female volunteer. A 16×16×4 localization matrix in an 8×8×4 cm PRESS VOI yielded 1 ml voxels. The location of the box in the brain and the spectra from slice 3 (richest in cerebral features) are shown in Figure 1b. The MRS took 25 min. and the entire session 1 hour. For comparison, a 27 min. 4 slice interleaved MRS was performed immediately afterwards over the same volume of brain without moving the volunteer or changing the imager's settings (Figure 1c). The analysis showed that, as expected, the SNR in 1b is ~2-fold better than in 1c for each of the ¹H spectral lines. This is because the entire examination time is available to all the slices in the hybrid, versus only N^-1 of it, for the interleaved.

SIMULTANEOUS THREE-DIMENSIONAL CHEMICAL SHIFT IMAGING OF ¹H-DECOUPLED ¹⁹F AND ³¹P FOR 5-FLUOROURACIL CHEMOTHERAPY. GONEN, in collaboration with LI,* MURPHY-BOESCH,* HE,* NEGENDANK,* T. BROWN*

In vivo ¹⁹F MRS studies of the chemotherapeutic agent 5-fluorouracil (5FU) suggest a correlation between 5FU uptake and retention in tumors and response to the therapy (which is less than 50%). The metabolic pathway of 5FU may be influenced by the energy status of the organ and/or tumor, which can be observed non-invasively by ³¹P MRS. The relationship between energy metabolite levels (from ³¹P MRS) and 5FU metabolism (from ¹⁹F MRS) and the tumor's response should, therefore, be investigated to determine if this information can predict the success of that treatment. Multivoxel techniques must be used to ensure differentiation between tumor and normal tissue. Unfortunately, observation of both nuclei requires over two hours because MRS of multiple nuclei is currently performed sequentially and the sensitivity of ³¹P MRS is low, as are the levels of ¹⁹F metabolites. Such a lengthy session can rarely be tolerated clinically.

FIGURE 2. a) Gradient and RF waveforms in the ¹H-decoupled simultaneous 3D CSI acquisition of ¹⁹F and ³¹P. b)¹⁹F (1600 Hz, 10 Hz filter) and ³¹P (400 Hz, 5 Hz Filter) spectra on a common vertical scale (per nucleus). The spectra in each matrix (obtained simultaneously) correspond to the 3×3 cm voxel in the dashed area of the image.

By exploiting the modularity of 3D CSI, we have simultaneously acquired ³¹P and ¹⁹F, thereby substantially reducing the time required. Because ¹⁹F is pulsed after ³¹P (between the phase encoding gradient lobes, S₁ and S₂, cf. Figure 2a), it is possible to impose the same field-of-view (FOV), L19_F = L31_P , and voxel size on both nuclei simultaneously if:

Unlike other techniques, CSI localization is independent of the value of the nutation angle, ₁ and ₂, which allows the use of surface-coils with inhomogeneous B₁s. Further, since the ¹⁹F in 5FU and its principal end-catabolite, -fluoro--alanine (FBAL) and some ³¹P metabolites are J-coupled to protons, significant additional sensitivity can be extracted by bi-level ¹H-decoupling to collapse J-multiplets and generate nuclear Overhauser enhancement (NOE). However, proximity of resonances (6%) made isolating the ¹⁹F receiver from the ¹H-decoupler difficult. This problem has only recently been solved by Murphy-Boesch et al.

The ¹H-decoupled simultaneous ¹⁹F and ³¹P experiment of Figure 2a was done at 1.5 Tesla magnetic field in a Siemens Magnetom 63SP imager equipped with a second RF transceiver and a third ¹H-decoupler channel. The 64 MHz ¹H coil used for magnetic resonance imaging (MRI) and decoupling was a 30×17 cm "figure-8" resonator with dual 15 cm diameter 59.8/25.6 MHz ¹⁹F/³¹P surface coils within its concave side. The sample was a 1 liter spherical flask (~12 cm diameter) filled with urine from a patient who was receiving continuous 5FU infusion. The urine was spiked with 5FU (to demonstrate the coverage of the broad, ~1200 Hz, ¹⁹F spectral-width) and with trimethylphosphate (to demonstrate the decoupling effects on the ³¹P). 8×8×8 3D CSI with 24×24×24 cm FOV (for both nuclei) resulted in overlapping 3×3×3 cm3 ¹⁹F and ³¹P voxels. Four phase-cycled averages were performed over 34 min. (TR=1 s).

¹⁹F and ³¹P spectra from the central slice of the 3D CSI matrices obtained from the phantom are shown in Figure 2b and illustrate the sensitivity profiles of the ¹⁹F and ³¹P coils. To evaluate the advantages, the decoupler was switched off, and the experiment was repeated. These results are also shown in Figure 2b. They clearly illustrate the dramatic SNR improvement for both the J-coupled ¹⁹F metabolites 5FU and FBAL. The same decoupling transmission also affects the ³¹P, collapsing the trimethylphosphate multiplet. NOE clearly enhances the signal of the P_i, which is not J-coupled to protons.

³¹P SIGNAL ENHANCEMENT BY POLARIZATION TRANSFER FROM ¹H IN CHEMICAL SHIFT IMAGING OF THE HUMAN BRAIN. GONEN, MOHEBBI-GILLANI, in collaboration with STOYANOVA,* T. BROWN*

³¹P MRS on human subjects is limited by the low SNR and spectral line-overlap. This is particularly problematic in the phosphomonoester (PME) and phosphodiester (PDE) regions where metabolite variations may predict response to cancer therapy. Currently, these problems are partially addressed by proton decoupling to collapse the ³¹P-¹H J-coupled multiplets and generate NOE. Nevertheless, further improvement of the SNR is clearly desirable, but not at the expense of longer measurement or higher fields.

Additional sensitivity and resolution could be obtained by heteronuclear polarization-transfer (PT) used extensively in high-resolution NMR. SNR gain of 1_H / 31_P ~2.4-fold can theoretically be realized for ³¹P J-coupled to ¹Hs and an additional [T₁(³¹P)/T₁ (¹H)]^0.5 is achievable from the shorter T₁ of protons. PT from ¹H to ¹³C, where the direct ¹³C-H bond provides for J-coupling >150 Hz, has been reported recently. However, because PMEs and PDEs have no direct ³¹P-¹H bonding, their J-couplings are much smaller (5-7 Hz), and require proportionally longer delays to generate PT.

Because any pulse(s) sequence that results in net transverse magnetization can be used in CSI (Figure 3a), we inserted refocused insensitive nucleus enhancement by polarization-transfer (RINEPT) from ¹H to ³¹P into 3D CSI (Figure 3b). RINEPT permits decoupling of the ³¹P signal. Because ¹H-decoupling saturates all the protons, including those that source the observed ³¹P magnetization, the effective ¹H recovery period, T_REC, is shorter than the cycle TR in Figure 3. A 1.5 s TR was chosen as optimal for the ³¹P and ¹H T₁ ratios in the brain and the 0.26 s ¹H-decoupling. The delays ₁, ₂= 22,11 ms were chosen for the ~6 Hz H-P J-coupling constants of PMEs and PDEs. The experiments were done in our imager using our home-built 64/25 MHz, ¹H/³¹P dual-tuned, four-ring, birdcage head-coil for imaging and CSI.

FIGURE 3. Top: Gradient and RF waveforms in a) 3D CSI and b) RINEPT; c) Ernst-angle versus d) RINEPT excitation spectra corresponding to the highlighted column of 3×3 cm voxels on the image from a slice of 3D CSI in a volunteer's brain. The spectral width (900 Hz) and vertical scale are common.

8×8×8 (27 ml voxels) RINEPT and direct-excitation Ernst-angle (_E) 3D CSI were performed on several volunteers. For comparison, both were performed in the same session without change to the volunteer's position or instrumental parameters. Two spectra arrays from a slice of the CSI matrices from one such experiment are shown in Figure 3c,d together with the image for anatomical reference. Comparing Figure 3c with 3d shows that all ³¹P spectral lines not J-coupled to protons are edited out from the RINEPT spectrum, in particular, the broad phospholipid background in the PME and PDE region.

Analysis on the (entire) 3D data sets reveals gain of ~1 in the PME region and ~1.8 in the PDE region. The enhancement is due entirely to the shorter T₁ of the protons. The full RINEPT ×2.4 sensitivity advantage is not realized because additional, homonuclear H-H J-couplings of similar magnitude (~6 Hz) competes with the H-P transfer. Addition of further pulses to remove this loss mechanism is currently under study.

DUAL NUCLEAR MAGNETIC RESONANCE SPECTROSCOPY: ¹H TWO-DIMENSIONAL CHEMICAL SHIFT IMAGING INTERLEAVED WITH ¹H-DECOUPLED ³¹P THREE-DIMENSIONAL CHEMICAL SHIFT IMAGING OF THE HUMAN BRAIN. GONEN, JAVAID, in collaboration with ARIAS-MENDOZA,* T. BROWN*

Since in vivo ¹H and ³¹P MRS provide complementary information, they are likely to be more useful if acquired together rather than individually. For example, differences in ³¹P MRS, low phosphodiesters, and high phosphoethanolamine may be used to evaluate tumors and aid in diagnosis, planning, and monitoring treatment. ¹H MRS of tumors has shown levels of neuronal loss, possible correlation between grade and choline level, and differentiation between active tumor versus necrotic regions. However, long measurement time makes multiple nuclei MRS in humans difficult as ³¹P averaging and localization add 30-50 min. to the MRI examination. Consequently, even when both ¹H and ³¹P MRS may be beneficial, neither performing them sequentially, nor recalling a patient under the same conditions at a later time, is feasible.

To address the time problem, we exploited the shorter T₁'s of proton metabolites (0.5-1.5 s versus 2-6 s for ³¹P) to interleave a proton acquisition in each TR of the ³¹P (Figure 4a). The ³¹P TR is increased to 1.5 s to accommodate a proton recovery time, T_REC(¹H), of 0.5 s.

FIGURE 4. a) Timings of a TR cycle in the heteronuclear interleaved experiment. TR(³¹P) = 1.5 s, T_REC(¹H)=0.5 s and 500 ms acquisition for both. During 256 ms of the ³¹P FID acquisition, broadband, ±140 Hz, WALTZ-4 decoupling was applied at _metabolites 1 ppm upfield from water. The decoupler was switched to _water, during T_REC(¹H) to maintain NOE, then "hopped" to 300 MHz to prevent interference with the ¹H-receiver during proton detection. b) Schematic representation of the ¹H and ³¹P VOIs. The outer, ³¹P 48×46×24 cm3 VOI is partitioned into a 16×16×8 localization grid to yield 27 ml voxels. Within it, a 24×24×1.5 cm3 proton 2D slice is partitioned into a 16×16×1 matrix to yield 3.4 ml ¹H voxels. Coplanar with this slice, an image-guided 7.5×7.5×1.5 cm₃ proton box located within the brain is defined by PRESS.

³¹P sensitivity was maintained by an appropriate increase of its nutation angle, ¹H-decoupling, and maintaining 15%-60% NOE by transmitting low-level RF at _water, during T_REC. The decoupler was switched to 300 MHz during ¹H acquisition to prevent saturating the proton-receiver. To suppress unwanted subcutaneous lipids signals from the ¹H spectrum, an axial PRESS (TE=135 ms) "box" located entirely within the brain was excited and a 16×16 2D proton CSI performed in its plane (Figure 4b). Off resonance ³¹P RF irradiation was applied during the second half of ¹H acquisition to collapse the phospholipid broad background signal.

The experiments were done on a 1.5 T Siemens Magnetom 63SP full body imager equipped with a second RF cabinet and a home-built decoupler. A quadrature, dual-tuned ¹H/³¹P (63/25 MHz), four ring, birdcage coil was used for the MRI and MRS. The fields-of-view were 24×24×1.5 and 48×48×24 cm, partitioned into 16×16 and 16×16×8 phase-encodes for 1.5×1.5×1.5 and 3×3×3 cm³ voxel for ¹H and ³¹P, respectively.

The experiment of Figure 4 was performed on several volunteers and results from a 40-year-old, healthy female are presented. ¹H MRS from within the 7.5×7.5×1.5 cm³ PRESS box, together with an overlapping spectrum of the coplanar ³¹P are shown in Figure 5. The corresponding axial MRI (with ¹H VOI indicated) is given for anatomical reference. At TR(³¹P)=1.5 s the MRS took 48 min. Together with ~30 min. for subject loading, coil tuning, shimming, and imaging, the entire procedure required less than 90 min.

FIGURE 5. A 24×24 cm axial image of the volunteer head with ¹H and ³¹P spectra from the same highlighted region. Both MRS were obtained during the same 48 min. MRS session.

Because the entire MRS time is available to both nuclei, the ¹H and ³¹P SNRs are improved 80% and 12%, respectively, compared with sequential measurements of the same duration. Since the ¹H box is contained within the larger ³¹P VOI, precise placement relative to the ³¹P matrix is unnecessary; the ³¹P 3D CSI data can be voxel-shifted in all directions and always properly aligned with the protons in post-processing rather than while the patient is lying in the magnet.

GONEN, O., HU, J., STOYANOVA, R., LEIGH, J.S., GOELMAN, G., BROWN, T.R. Hybrid three-dimensional (1DHadamard, 2D-chemical shift imaging) phosphorus localized spectroscopy of a human brain. Magn. Reson. Med. 33: 300-308, 1995.

*Fox Chase researcher

^a G. Goelman: University of Pennsylvania, Philadelphia, PA 19104