[1] 13w In Vivo High-Resolution Volume-Selected Proton Spectroscopy and TI Measurements in the Dog Brain

[1] 115w Successful in vivo NMR spectroscopy requires a combination of techniques to address the problems of volume selection, water suppression, and resolution. All this needs to be done in the very heterogeneous environment found in living organisms. Previously published techniques are used to obtain 'H spectra from a dog brain, observing metabolites with concentrations below 1 mM. Measurements of spin-lattice relaxation times (T,) are also presented. The 'H relaxation times are long ( T , > 1 .O s) yielding information about the fluidity of the molecular environment. Comments are made concerning the achievable linewidth in vivo and the deficiencies that phase-encoding spectroscopic methods may have in obtaining high-resolution 'H spectra. o 1989 Academic Press, Inc.

[1] 116w All pulses were applied using a slotted resonator to give a homogeneous B, field, but the signal was received on a surface coil having a diameter of 4.8 cm. Using 4.5 kW of pulse power, the a12 pulse time was typically 120 ,US. This probe arrangement has two major advantages: (i) improved signal-to-noise arising from close coupling of the surface coil to the region of interest and (ii) reduced water proton error signal resulting from the limited received volume. This results in reduced dependence on the add-subtract component in the volume selection procedure and thus increased sensitivity because of higher gains possible in the receiver loop. The probe was ar-0 To whom correspondence should be addressed.

[2] 123w ranged such that the B1 fields from the two coils were perpendicular so as to reduce coupling. Complete geometric decoupling however is not possible and further decoupling was achieved using the circuit shown in Fig. 1. During transmission on the slotted resonator, the voltage induced in the surface coil is sufficient for the diodes to conduct. This shifts the resonant frequency of the surface coil by 5 MHz thus reducing any induced rf field. During reception the diodes are effectively open-circuited. It should be noted that this probe arrangement and the results obtained herein are contrary to the published opinion (2) that 3D localization methods cannot be successfully employed to obtain water-suppressed 'H spectra from subjects that extend outside the main transmitter coil.

[3] 112w High-resolution NMR spectroscopy requires a homogeneity in the volume of interest of less than 1 part variation in lo7. This is difficult to achieve over large volumes, especially in vivo where there may be magnetic susceptibility changes within the sample. NIMBLE ( 3 ) provides a method of adjusting the homogeneity in the voxel being investigated. It makes use of the VOSY pulse sequence ( 4 ) to perform volume-selected shimming and assumes that inhomogeneities can be removed from a small volume with the x, y , and z linear shim gradients. Using this technique, linewidths as low as 6 Hz at half-height can be achieved in viva Volume-Selected and Water-Suppressed Spectra

[4] 21w The in vivo spectrum of a voxel in the dog's brain was obtained using the pulse sequence described in Fig. 2.

[5] 83w The water resonance was suppressed in a manner similar to that described previously ( 5 ) using a combination of SUBMERGE (6) and a refocused read pulse cluster ( 7, 8 ) . SUBMERGE employs a series of narrowband sinc pulses separated by an application of one of the linear field gradients, acting as a homospoil. Suppression was improved using pulsed gradients to adjust the shimming during the application of SUBMERGE. The rationale for this is as follows. The optimum resolution in the

[6] 149w Figure 3 shows typical spectra obtained at the three stages of the experiment on a dog's brain. The field homogeneity was adjusted observing the total signal received by the surface coil, resulting in a spectrum showing H 2 0 and lipid with a linewidth at half-height for the water proton resonance of 30 Hz (Fig. 3A). A proton density image was obtained using a standard spin-echo method and coordinates for selected voxels were determined. Using the water signal, NIMBLE was used to optimize the homogeneity of the voxel and a 'H spectrum was obtained using SPACE (Fig. 3B). Following volume-selected shimming the linewidth of this peak was typically between 5 and 10 Hz. The pulse sequence shown in Fig. 2 was used to obtain the watersuppressed, volume-selected spectrum (Fig. 3C). This spectrum is an average of 128 acquisitions and took 14 min to collect from a 1 6-cm3 volume.

[7] 150w A broad baseline "hump" was noted in all spectra. This broad background signal was not observed in previous applications of volume selection methods to obtain spectra from the brain (2). Those studies, however, employed methods that use long echo times as part of the volume selection procedure and thus this broad background signal may not have been observed. Using the techniques employed in this paper, there is a reduced T2 weighting as the magnetization is in the transverse plane for less than 30 ms. This broad resonance can be removed by data manipulation, either weighting the FID or performing spline baseline fitting of the transformed spectrum. Alternatively, deliberate use of a relatively short spin echo ( 10 ms) immediately prior to acquisition can be employed. The result of such an experiment is shown in Fig. 4, which also gives peak assignment on the basis of literature values ( I, 2).

[8] 14w Longitudinal relaxation times, T I 's, were measured by placing a composite a pulse

[9] 137w and a T~ time immediately after the SUBMERGE part of the pulse sequence. Figure 5 shows a series of spectra with different values for T ~. The intensity of each peak was measured for each T I time giving a curve similar to that shown for the CH3 of N-acetylaspartate in Fig. 6. The T I values for a number of resonances, calculated using a three-parameter fit ( 1 3 ) , are given in Table 1. Assum- T (ms) m 3 5 0 1150 650 450 250 I . I . I . I 4 3 2 1 PPm FIG. 5. Spectra obtained by placing a T -T , between the SUBMERGE and the SPACE parts of the pulse sequence shown in Fig. 2. 71 was vaned between 250 ms and 10 s (labeled m ).

[10] 90w ing that the dominant mechanism for spin-lattice relaxation is proton-proton dipole-dipole interactions it is possible to calculate the effective correlation time 7, using At a field strength of 2.35 T, r = 1.78 A, and T , = 1.02 s, the solutions for 7, are 3.65 X lo-'' and 1.093 X s. Although it is reasonable to assume the T , process to be FIG. 6. A plot of peak intensity against T I time in an inversion reCovery experiment forthe CH, resonance from N-acetylaspartate (peak I1 in Fig. 4).

[1] 88w It has recently been demonstrated that in vivo 'H spectra of brain can provide useful biochemical data ( I ) and that high-resolution 'H spectra can be obtained from the human brain (2). In this paper we demonstrate the use of a combination of previously published techniques to provide in vivo water-suppressed proton spectra of dog brain with linewidths from some metabolites of 6 Hz. Because of our probe arrangement the methods used are easily modified to obtain an accurate determination of longitudinal relaxation times, T I .

[2] 111w final spectrum is achieved by deliberate adjustment of the homogeneity in the voxel of interest using NIMBLE. However, this usually results in a large broadening of the resonances outside the voxel of interest, a point noted in the report on NIMBLE (3). The H20 resonance from the whole sample may well be 500 Hz wide. Although this feature is advantageous in aiding the add-subtract aspect of the volume selection method, the consequence is that water suppression may be compromised, making it impossible to excite the water using a frequency-selective pulse. On the other hand, it is relatively easy to shim the total volume observed by the surface coil to 40-50 Hz.

[3] 84w If SUBMERGE is applied in the presence of the shim settings for the whole sample then the water suppression will be greatly improved. For this reason the gradients were calibrated so that shimming changes introduced using the Bruker system control microprocessor (SCM) could be compensated by applying small pulsed field gradients during the SUBMERGE part of the pulse sequence. It must be emphasized that unless NIMBLE is used we find it impossible to achieve the degree of resolution shown in the spectra discussed below.

[4] 174w As noted in Fig. 2, the read pulse was a 1-27-5-7-5.4-27-7 pulse cluster (7) followed by a ~-7 ~-segment to refocus the phase roll as discussed previously (8). The free precession 7 during the pulse cluster was 2 ms giving maximum excitation at 250 Hz from resonance, and rR was set to 5.1 ms. The 1-5.4-5.4-1 pulse was chosen because of a superior excitation profile which excites signals within 100 Hz of the water resonance. SPACE (9,10) was used to select a voxel, with coordinates obtained from a standard spin-echo proton image. Preemphasis and Bo compensation ( I I ) were adjusted using the same gradient episodes as those used in the final experiment. The sinc pulses used were defined between +3n, and excitation profiles for 3-and 6-ms sinc pulses, determined as discussed previously ( 1 2 ) , had excitation bandwidths of 2.5 and 1.2 kHz, respectively. Using a gradient strength of 1000 Hz/cm, this gives a voxel size of 16 cm for a 3-ms sinc and 1.7 cm for a 6-ms sinc.

[5] 110w Following the SPACE volume selection method and prior to the read pulse cluster, all magnetization is along the 2 axis. We have noted that the water suppression can be further improved if the pulse segment 7-( delay)-?-(gradient) is inserted. The rationale is that any broad water error signal (outside the voxel) will be in an inhomogeneous B , field. Following the H, ?r pulse (which are in fact 8,e outside the B, homogeneous volume) error signal is rotated into the transverse plane and spoilt by the gradient. This procedure can result in an increase of up to 10 dB in receiver gain setting resulting in the appropriate increase in signal-to-noise.

[6] 82w Mongrel dogs ( 6-8 kg) were premedicated with acepromazine ( 100 pg/ kg) . Anesthesia was induced with 4% halothane in N 2 0 / 0 2 and maintained, after endotracheal intubation, with 1.5% halothane in 02. The electrocardiogram was monitored throughout the experiment and atropine ( 200-300 pg ) was administered subcutaneously whenever the pulse rate dropped below 100 beats per minute. In this way, anesthesia was maintained for up to 6 h, following which the dog was allowed to recover.

[7] 10w All experiments were approved by the University Animal Ethics Committee.

[8] 27w No. (see Fig. 4) I1 -CH3 of N-acetylaspartate 1.5 IV N-CH3 of creatine/phosphocreatine I .4 VI N-CH,-COO-of creatine/phosphocreatine 1 .O V N(CH,), of choline/phosphocholine 1 . 1

[9] 196w dominated by proton-proton dipolar interactions this may not be so for spin-spin relaxation. However, assuming a dipole-dipole mechanism the second solution for 7, would mean, using Eq. [ 21, that T2 would have a value of 0.1 1 ms, corresponding to a linewidth of 3 kHz, From these calculations it is apparent that the shorter 7, is the correct answer and that the upper limit for T2 is when T2 = T I . We may conclude that either ( a ) proton in vivo linewidths as low as 1 Hz are theoretically possible and on our system we are limited by magnetic field inhomogeneity, or (b) there is another mechanism dominating the T2 process, or (c) the linewidths observed have a significant contribution from chemical-shift line broadening wherein molecules in different compartments within our chosen voxel have slightly different chemical shifts. A possible dominating T2 process may well be slow molecular reorientation through regions of different magnetic field susceptibility. As regards the measured 7, value it is difficult to determine to what extent internal reorientation and overall molecular reorientation contribute; however, we can say that the metabolite molecules appear to be in a low-viscosity medium.

[10] 193w We have demonstrated, however, that linewidths as small as 6 Hz can be achieved. It is therefore essential that any spatial selection technique used for in vivo proton spectroscopy optimize Bo homogeneity. Experience in many laboratories studying in vivo NMR spectroscopy would indicate that it is difficult, if not impossible, to shim the Bo field over the total magnet d.s.v. such that the Bo homogeneity is of this order over all voxels of interest simultaneously. This suggests that optimum proton spectra can be obtained only if some form of volume-selected shimming is used. Therefore, any technique that uses an imaging type sequence and multidimensional Fourier transformation methods, methods which by their nature acquire data from all voxels simultaneously, will probably have compromised magnetic field homogeneity and thus poor resolution in some voxels. The fact that some success using such methods has been achieved using 31P spectroscopy ( 1 4 ) is not convincing evidence that such methods will work well under high-resolution conditions for in vivo 'H MRS as the Bo magnetic field inhomogeneities will be a factor of 3 less for 31P than for 'H in determining the achievable linewidth (in Hz).