THREE DIMENSIONALIN VIVO METABOLIC IMAGING
IN THE HUMAN BRAIN

The possible clinical applications of in vivo localized nuclear magnetic resonance (NMR) spectroscopy (MRS) are extensive. In the past few years, many studies have been published suggesting the utility of human in vivo multivoxel localized proton spectroscopy (¹H MRS) for cancer, epilepsy, Alzheimer's disease, multiple sclerosis, AIDS and stroke. The aim of our research is to provide advanced diagnostic MRS techniques that will: 1) measure more than one nucleus in one session, 2) improve localization methods to make examinations shorter and clinically more robust, 3) collect the maximum amount of spectral information in a given time, and 4) translate these advanced methodologies to clinical research and application.

THREE-DIMENSIONAL HADAMARD PROTON LOCALIZED SPECTROSCOPY IN THE HUMAN BRAIN. GONEN, VISWANATHAN, in collaboration with GOELMAN^a

In vivo ¹H MRS of human brain is currently performed by suppressing the subcutaneous fat at the rim of a thin, selectively excited slice followed by two-dimensional (2D) chemical shift imaging (CSI). Using Hadamard spectroscopic imaging (HSI) instead of CSI offers improved localization, i.e., the ability to more precisely identify the region(s) in the object from which a signal comes from. However, until recently, transverse (T-HSI) or longitudinal HSI (L-HSI) could not be used to localize more than 8 partitions (i.e., ~8 cm field-of-view [FOV] in the brain) due to high pulse power requirements. Fortunately, time-shifting of the adiabatic inversions, proposed recently by Goelman, allows us to use 16 L-HSI partitions and perform a complete 3D-16^th order L-HSI to cover an FOV that previously could only be imaged with CSI.

A schematic diagram of the 3D-HSI pulse sequence used is shown in Figure 1. It comprised a chemical-shift-selective-suppression (CHESS) pulse to suppress the water signal, followed by 2D-16^th order, 10 millisecond (ms) long, 20% time shifted, adiabatic LHSI pulses in two directions, in the axial plane. The volume-of-interest (VOI) was selectively excited by a double spin-echo with the THSI 90° pulse along the short, inferior-superior, axis of the brain.

FIGURE 2. Left: Axial brain slice superimposed with the real part, 4-1 ppm region, of a 10x10 spectral sub-matrix of a 16x16 2D LHSI array. Note the exceptional isolation of lipid contamination near the skull. Right: NAA metabolic map of the spectra from the left, (10x10 cm VOI) imposed on the image. Note correspondence with the gross anatomy (ventricles) in the slice.

The sequence in Figure 1 was performed on a 32-year-old male volunteer. The VOI was interactively image-guided and all the shaped pulses were generated with our software. The 10x10 ¹H-MRS spectra matrix from a 10x1x1 cm slice within the VOI are shown in Figure 2, superimposed over the magnetic resonance imaging (MRI) for anatomical reference. With a recycle-time of 1.5 s, the experiment took ~30 minutes to produce superior signals from 1 milliliter (ml) voxels. The correspondence of the spectra to the gross anatomy is also exceptional, as depicted by the Nacetylaspartate (NAA) metabolic map shown in Figure 2. Additionally, the point-spread-function of LHSI successfully rejected lipid signals as close as 1 cm from the skull (Figure 2). To our knowledge, this is the first time 16^th order HSI has been used for multivoxel ¹H-MRS.

The most common neuro-imaging abnormality in Neurofibromatosis Type 1 (NF1) is high signal intensity lesions on T₂ weighted MRI, also and referred to as "unidentified bright objects" (UBO; Figure 3a). Unlike tumors, UBOs do not enhance with contrast or exhibit mass effect. Because of their diffuse, multi-focal nature, studying their metabolism requires 3D MRS around the entire lesion. Unfortunately, 3D ¹H MRS is currently done by interleaving four 2D slices.

Since interleaving is inefficient, obtaining good voxel signal-to-noise ratios (SNRs) requires ~40 minutes, an obstacle in pediatric applications since MRS is aborted if the sedated child awakes. The obstacle is overcome with our 1D-4^th order HSI/2D-CSI hybrid for full 3D volume localization. It covers 1/4 to 1/2 liter VOI, producing 4 axial HSI slices of 16x16 voxels each over 6_Zx6_Xx6_Y or 6x8x8 cm VOIs, yielding hundreds of sub 2 cm³ voxels in under 30 minutes.

Six children with NF1, ranging in age from 10 months to 11 years old, have been studied with this technique. The 3D coverage allowed us to distinguish spectra of UBO from "normal brain" of a 3-year-old NF1 patient (Figure 3a). The UBO is clear on the image, and spectra from that region and its surroundings exhibit 2>=Cho:Cr>=1.3 (P<0.05). Comparison with a similar region in a normal 7-month-old infant (Figure 3b) shows that in healthy brain Cho:Cr=~1.0±0.1. When a 10-month-old female NF patient was studied, her MRI did not indicate an abnormality (Figure 3c). However, the MRS from her thalamus is analogous to that of the UBO (Figure 3a), rather than to that of the normal age-match (Figure 3b). This indicates that the disease's metabolic processes have already begun, but have not yet altered her anatomy.

THREE-DIMENSIONAL LOCALIZED ¹H MRS IN MULTIPLE SCLEROSIS. GONEN, LI, VISWANATHAN, in collaboration with MANNON,^c KOLSON,^c GROSSMAN^c

Contrast-enhanced and T₂-weighted MRI are central in the evaluation of multiple sclerosis (MS) due to their sensitivity and ability to gage the accumulation (burden) of the disease. However, microscopic lesions may be more pervasive than those seen on MRI, i.e., in normal appearing white matter (NAWM) of the patients. Augmenting MRI, ¹H-MRS is used to assess the irreversibility of lesions, discern edema versus. demyelination, and evaluate NAWM. Unfortunately, the transient, multifocal nature of MS, combined with the extent of the white matter, from the brainstem to the subcortex, make the previous single-voxel and the more recent 2D multivoxel methods ill-suited to evaluate NAWM. Due to their limited VOI, these methods must be image-guided onto visible pathologies, a serious shortcoming for the investigation of NAWM. The obstacle is overcome with a 3D (1D-8^th order HSI with 2D 16x16 CSI) hybrid to cover, at <1 cm³ resolution in a clinically feasible ~45 minutes and with excellent SNR, almost all of the white matter in the brain. This makes a priori knowledge of precise location(s) of MS pathologies unnecessary.

NAA deficits in NAWM manifest in severalwhite matter regions of a 29-year-old relapsing-remitting (RR) female MS patient from a cohort of 12 in our long-term, serial ¹H-MRS study. Her MRI and MRS were compared with a matched control. As shown in Figure 4, corresponding axial images from the 3^rd, 4^th and 5^th slices in the corpus callosum and thalamus are superimposed with NAA metabolic maps representing 8_LRx10_AP cm VOIs, zero filled to 64x64.

FIGURE 4. Top, a--c: Three slices from the brain of a 29-year-old female RR MS patient, superimposed with the 8LRx10AP cm NAA metabolic maps. Bottom: Three approximately corresponding slices from an age and sex matched normal control and equal-size NAA maps. All maps are on the same quantitative scale. Note the patient's 20 to 30% NAA deficits (marked with arrows) relative to control.

FIGURE 5.Top, left, a: 1H MRS from 5 mM NAA in a 3 liter phantom using the sequence on the right. Insert: schematic structure of NAA and its 1H peak assignments. Center, b: 1H MRS from a human head using the sequence on the right without the 180°s or f alternation. Bottom, c: same as b, with the phase-cycling of Table 1. Note near complete elimination of the lipid's signal. Right:  The WBNAA sequence. 180°1 and 180°2 were applied in the order of the steps cycle in Table1. The FIDs were either added or subtracted from the average according to f in Table 1. The GX, GY and GZgradient pulses are crushers only.

QUANTITATING TOTAL BRAIN N-ACETYL ASPARTATE AND ITS DECLINE WITH AGE IN MULTIPLE SCLEROSIS USING NON-ECHO ¹H MRS. GONEN, VISWANATHAN, SWAMINATHAN, in collaboration with BABB,^§ CATALAA,^c MANNON,^c KOLSON,^c GROSSMAN^c

The amino acid, NAA (Figure 5a), is thought to be present almost exclusively in neurons and their dendritic extensions. Thus, it is a sought indicator of neuronal injury of any source. Its CH₃ group yields a prominent peak in the ¹H-MRS of the brain. However, current ¹HMRS is restricted to small VOIs completely within brain; this is done to protect the small NAA peak from the overwhelming signals of subcutaneous lipids at similar chemical shifts (Figure 5b). Since these VOIs are much smaller than the brain, they must be image-guided onto the pathology of interest subjecting ¹HMRS to: 1) the assumption that relevant metabolic changes occur only there; and 2) repositioning errors in longitudinal studies. Both problems would be circumvented if the amount or concentration of NAA in the whole brain (WBNAA) were available.

A sequence to obtain WBNAA must be non-T1 or T2 weighted and immune to the intense lipid signals. These issues are addressed by the sequence depicted in Figure 5; it comprises a four-step cycle of non-selective inversions and alternating receiver phases as shown in Table 1. Its components are: 1) Water suppression. A 25.6 ms long, 120 Hz CHESS and one 1331 pulse suppressed the water by more than 10³, as shown by the uncorrected flat baseline of the spectrum from a 3 liter 5 mM NAA phantom (Figure 5a). 2) Removal of T₁-weighting. A 10 s TR»T₁^NAA of 1.35±0.05 s, ensured that each step in the cycle starts from thermal equilibrium regardless of the precise T₁s. 3) Removal of T₂-weighting. Figure 5 is essentially a 90° (1331) excitation directly followed by detection. This avoids T₂-weighting and Jmodulation; the very minor "skew" of the CH₂ doublet, J=7Hz (Figure 5a). 4) Fat suppression. The overwhelming fat signals in (Figure 5b) were suppressed by ~10³ by the "add-subtract" cycle of Table 1. Since T₁s of lipids (<250 ms) are «T₁ ^NAA, an inversion time (TI) of 0.67· T₁^NAA ensures that the former are relaxed, while the latter cross the null, in steps 2 and 4. Combining the steps leaves just the sought equilibrium signal of the NAA (Figure 5c).

The intra- and inter-individual WBNAA variations were measured in a cohort of five 42±5-year-old normal females at 35 different times, using the sequence of Figure 5. The average individual WBNAA in that group was 10.63 mM with a 95% confidence interval of 10.43 to 10.82 mM, similar to previously reported local concentration.

We used this new method to compare WBNAA between RR MS patients, along with their age and sex-matched controls, to examine if consistent NAA deficits indicate neuronal/axonal losses sustained in the entire brain. Ten patients, 8 women and 2 men, ranging in age from 26 to 53, who have been suffering from RR MS for at least 6 years, were recruited. Fourteen normal controls, i.e., not having any known disease of the CNS, were also recruited. The amount of NAA in the brain was obtained in 15-20 minutes; conversion to WBNAA was accomplished by dividing the NAA amount by the brain volume from high-resolution MRI. All the results from all subjects are shown in Figure 6.

Mann-Whitney tests among the MS subjects showed that the WBNAA of the RR MS patients was significantly lower than that of the control subjects (P<0.0001); the difference was significantly greater among older subjects (P=0.03). Least squares regression and analysis of variance was used to characterize the WBNAA relationship with age. It revealed that the rate of WBNAA decline was significantly faster among MS patients than among controls (P=0.04).

Based on this observation, a second regression analysis was performed using the patient and control groups separately. For the patients, the linear prediction for the WBNAA,
, was

= 10.76 - 0.074 · AGE [mM],    [1]

where "AGE" is in years. The negative slope, -0.074, was significantly <0 (P=0.02). For controls was given by,

= 11.83 - 0.026 · AGE [mM]   [2]

The slope was also significantly less <0 (P=0.029), i.e., 25 to 52-year-old normal individuals also undergo a (slower) WBNAA decline. The s predicted by equations [1] and [2] for patients and controls as a function of age are shown in Figure 6, overlaid on the data.

FIGURE 6. WBNAA versus age data from all subjects (up to 5 measurements each), controls (m), and RR MS patients (l). Solid lines are the prediction for from equations [1] and [2] with their lower and upper 95% confidence bands (dashed-lines).

PUBLICATIONS

GONEN, O., MURPHY-BOESCH, J., LI, C-W., PADAVIC-SHALLER, K., NEGENDANK,W.G., BROWN, T.R. Simultaneous 3D CSI of ¹H-decoupled ¹⁹F and ³¹P for 5-FU chemotherapy. In High-power Gradient MR-imaging, Advances in MRI II, edited by M. Oudkerk and R.R. Edelman. Blackwell Science, Berlin--Vienna, pp. 271-277, 1997.

GONEN, O., VISWANATHAN,A.K., BABB, J., UDUPA, J., CATALAA, I., GROSSMAN, R.I. Total brain N-acetylaspartate concentration in normal, age-grouped females: quantitation with non-echo proton NMR Spectroscopy. Magn. Reson. Med. 40: 684-689, 1998.

GONEN, O.,WANG, Z.-J., VISWANATHAN, A.K., MOLLOY, P.T., ZIMMERMAN, R.A. A Hadamard/chemical shift imaging hybrid for pediatric brain 3D multivoxel proton NMR spectroscopy: Application to Neurofibromatosis Type I. AJNR, Am. J. Neuro. Radiol. (in press).

^§   Fox Chase researcher

^a   G. Goelman: Department of Nuclear Medicine, Hadassah Medical Center, The Hebrew University of Jerusalem, Israel

^b   R. Wang, P. Molly, R. Zimmerman: Departments of Radiology, Neurology and Neuro-Oncology, Children's Hospital of Philadelphia, Philadelphia, PA 19104

^c   R.I. Grossman, L.T. Mannon, D. Kolson, I. Catalaa: Department of Radiology, University of Pennsylvania Medical Center, Philadelphia, PA 19104

Illustrations or unpublished data in these reports should not be used without permission of the author.

Fox Chase Cancer Center

Scientific Report 1998