selective irradiation

Transcription

1 Proc. Natl. Acad. Sci. USA Vol. 81, pp , November 1984 Medical Sciences In vivo nuclear magnetic resonance chemical shift imaging by selective irradiation (NMR techniques/in vivo biochemistry/chemical imaging) P. A. BOTTOMLEY, T. H. FOSTER, AND W. M. LEUE General Electric Corporate Research and Development Center, P.O. Box 8, Schenectady, NY Communicated by Charles P. Bean, July 12, 1984 ABSTRACT NMR images of preselected chemically shifted species can be obtained by selective irradiation of the remainder of the NMR chemical shift spectrum prior to application of a conventional NMR imaging sequence. The chemicalselective irradiation consists of narrow-bandwidth ir/2 or saturation radio-frequency pulses applied in the absence of imaging gradients. The technique permits substantial reductions in scan and reconstruction times over standard three- and four-dimensional Fourier transform chemical-shift-imaging methods, when images of few spectral peaks are desired. It is also suitable for the elimination of chemical shift artifacts in conventional high-field NMR imaging. In vivo applications of the technique to the head and limbs in a 1.5-T magnetic field yield 1H H20 and -CH2- images, with little detectable -CH2- in muscle and brain. Advances in NMR techniques have yielded two promising new tools of medical consequence. First, NMR images display the spatial distribution of relatively mobile magnetic nuclei such as 1H in the body (1). Second, discrimination of the chemical environment of a particular nuclear species is practical via spatially localized in vivo chemical shift spectroscopy, which typically provides profiles of the concentrations of 'H-, 31P-, or 13C-containing biochemicals (2-10). The compatibility of these two technologies is frustrated by the substantially higher static-magnetic-field (>1 T) and field-homogeneity (l ppm across the region of interest) requirements of chemical shift spectroscopy (10). In addition, the magnetic field gradients used for spatial localization in conventional NMR imaging cannot usually be applied synchronously with the acquisition of chemical shift data because the spatial and chemical shift information is in general inseparable. Thus, although the instrumental problems associated with performing high-field body imaging and spectroscopy on the same instrument have been substantially solved (9, 10), spatial localization of chemical shift spectra in clinical applications typically involves the restricted spatial selectivity provided by small flat NMR detection coils attached to the sample surface (2-7, 9, 10). While providing improved signal-tonoise ratios over whole-volume coils, the intrinsically inhomogeneous radio-frequency (rf) sensitivity profiles of surface coils result in surface tissue contamination of spectra and lack of penetration in deep organ studies (11). Several techniques have been proposed to enable spatial resolution of spectra using whole-volume rf coils, the most popular of which employs spatial encoding magnetic field gradients and three- or four-dimensional (3-D or 4-D) Fourier transformation (FT) to reconstruct complete images of chemical shift spectra as a function of position (12-15). The latter techniques are similar to conventional FT zeugmatography or spin-warp imaging methods except that the readout The publication costs of this article were defrayed in part by page charge payment. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C solely to indicate this fact magnetic field gradient that is normally applied during data acquisition is removed and replaced by a gradient pulse applied in the same direction after the rf excitation pulse but prior to data acquisition. To regain all of the spatial information along the direction of the readout gradient lost by this operation, the entire spectroscopic imaging pulse sequence must be repeated at least n, times with nx different readout gradient pulse amplitudes, where n, is the desired spatial resolution in the direction of the readout gradient (Gx). Therefore, an n,-fold increase in the minimum image scan time results, and the FT reconstruction is increased by an extra dimension beyond that required for conventional imaging, corresponding to the frequency domain of the chemical shift information. We describe and demonstrate an alternative technique for directly generating volume NMR images of a preselected chemical species. The method employs selective irradiation of all other species in the absence of imaging gradients immediately prior to application of any conventional NMR imaging sequence. Since the spatial imaging sequence is intact and chemical information is selected a priori, there is no increase in the minimum scan time for imaging nor in the Fourier reconstruction time. If signal averaging is required to image chemical species such as those of 31P or '3C in vivo, the maximum scan-time advantage over the 3-D or 4-D FT techniques reduces to nx/n for a single chemical image, where N is the number of averages required for each set of imaging gradient values. If multiple chemical images of m different chemically shifted peaks are desired, a further m- fold reduction in the scan-time advantage ensues. However, for in vivo 1.5-T 1H NMR, where typically N = 1, m = 2, and nx = 256 (10), the time advantage is 128-fold. The technique is equivalently applicable to the elimination of any chemical shift artifacts present in high-field NMR images (10, 16). Chemical shift artifacts occur when the readout gradient is too weak to confine the complete chemical shift spectrum within a single picture point-that is, when Gx < 2 nx w8/yax), where 8, y, and ax are, respectively, the frequency bandwidth of the chemical shift spectrum, the nuclear gyromagnetic ratio, and the spatial extent of the sample in the direction of Gx (16). Artifacts are manifested as rings in multiple-angle-projection imaging (16) and as ghost images in conventional spin-warp imaging (10). Increasing Gx to remove the artifacts increases the bandwidth of the NMR imaging signal, thereby reducing the signal-to-noise ratio. Therefore, selective irradiation of undesirable chemical species prior to imaging is an advantageous alternative means of removing the source of artifacts in chemical shift imaging whereby no readjustment of gradient strength is necessary. Previous nonimaging in vivo applications of selective irradiation to NMR spectroscopy include the saturation transfer method for studying 31P reaction-rate kinetics (17) and, more Abbreviations: FT, Fourier transformation; 2-, 3-, and 4-D, two-, three-, and four-dimensional.

2 recently, solvent (H20) suppression in 'H spectroscopy of rat brain (7). It is 'H-selective saturation NMR chemical imaging of the human brain and limbs that is exemplified here. METHODS All experiments were performed on a 1.5-T, 1-m-bore Oxford Instruments superconducting magnet and a broad-band NMR imaging research system operating at a 62.5-MHz 1H NMR frequency (10). The selective-irradiation chemical-imaging timing sequence is shown in Fig. 1: the rf and magnetic field gradient pulses applied during intervals to through t3 correspond (i) to image slice selection with a ir/2 rf pulse, (ii) spatial encoding in one direction within the slice, (iii) application of a ir rf pulse to generate a spin-echo NMR signal, and (iv) acquisition of the spin-echo data in the presence of a second orthogonal gradient within the slice, respectively, as in a conventional 2-D FT planar spin-warp imaging sequence (18). Selective irradiation of undesired chemical species is performed in interval t4, immediately preceding the imaging sequence, using a long-duration, amplitude-modulated rf pulse in the absence of imaging gradients. Hydrogen in the body is principally associated with the two discrete chemical groups H20 and -CH2-, often associated with lipid (1, 4, 8, 10, 14). The separation of the H20 and -CH2- chemical shift peaks is about 3.5 ppm (10, 19). Selective-irradiation chemical imaging is possible only if the spectral peaks are resolvable from the entire imaging volume dur- r- Gz I Medical Sciences: Bottornley et al I..N >/Programmable % Amplitude v: - -.1I Proc. NatL. Acad. Sci. USA 81 (1984) 6857 ing interval t4. In the absence of applied imaging gradients, we achieve better than 0.5 ppm full-width half-maximum linewidth on the H20 resonance from the entire head or limbs as demonstrated in Fig. 2, using variable superconducting and resistive magnetic-field-gradient shim coils (10). However, application of the pulsed imaging gradients in the Fig. 1 sequence can produce transient line-broadening gradients induced by eddy currents in the magnet cryostat or structural metal, thereby slightly degrading the homogeneity during t4. Therefore, it is desirable to readjust the resistive shim gradient coils for optimum homogeneity during application of the image pulse sequence. Shimming the homogeneity by using the NMR signal from the selected slice recorded during t3 with G, turned off usually suffices. Selective irradiation of either H20 or -CH2- peaks in interval t4 is preferably accomplished with a sinc-function amplitude-modulated pulse of RF with a center frequency precisely tuned to the selected NMR peak. A pulse modulated by a sinc function in the time domain irradiates a rectangular profile of NMR frequencies of bandwidth inversely proportional to the pulse width. The amplitude and duration of the chemical-selective irradiation pulse are adjusted for maximal annihilation of the selected peak as observed in interval t3. Since the initial rf pulse in the imaging sequence is a ir/2 pulse in interval to, annihilation of selected peaks first occurs when the chemical-selective irradiation pulse is also nominally a r/2 pulse. An approximately 20-fold or more attenuation of the selected peak is typically achieved by this means without affecting the amplitude of the nonselected peak. Irradiation of the H20 peak effectively results in a -CH2- image and vice versa. Switching the operating frequency of the NMR spectrometer between the resonances of H20 and -CH2- has negligible effect on slice selection or nuclear-spin inversion in intervals t0 and t2 of the imaging sequence because the bandwidths of the 7T/2 and 7r rf pulses are much greater than the range of chemical shifts present in the sample. RESULTS Fig. 3 compares a series of 1H, -CH2-, and H20 chemicalselective irradiation images from the head and leg with conventional scans. Images consist of 256 x 256 independent RF s Chemical selective nl2 Slice Selective Head NMR Signal I ~ ~ I T IN+In T i t4 -to tj t2 t3 Time FIG. 1. Chemical-selective planar NMR imaging sequence. G, Gy, and G, are linear-magnetic-field imaging gradients defined by G, = abodax, G, = abo/ay, and Gz = dbo/az, where Bo is the main static NMR field directed along the z-axis of a Cartesian (x, y, z) coordinate system. rf transmitter pulses are represented by their envelopes. Intervals to through t3 represent a typical planar spin-warp spin-echo imaging sequence (10): chemical selection occurs in interval t4. The entire sequence is repeated with period TR using ny different values of GY to obtain an nx x ny point image. Spurious signals excited by the other pulses are removed by recording and subtracting two signals for each Gy value in the image sequence with the slice-selective pulse phase-alternated ppm FIG. 2. 'H NMR spectra from the entire human head and the leg recorded by using the head-volume imaging coil and a single rf excitation pulse. Peaks correspond to H20 and -CH2- groups.

3 6858 Medical Sciences: Bottomley et al. Proc. NatL Acad Sci. USA 81 (1984) A B C FIG. 3. 1H 62.5-MHz transverse NMR images representing the -CH2- (A) and H20 (B) distribution in the head and limbs obtained by selective irradiation of H20 and -CH2- resonances, respectively. (C) Conventional images recorded with the chemical-selective irradiation pulse turned off. The scans were obtained in 5 min from the lower leg (Left), to eye-level (Center), and to the base of the ventricles (Right). Array sizes are 256 x 256 points of 1 mm x 1 mm x 4 mm volume each. Spectrometer and display parameters are the same for each section. The white spot near the center is an instrumental artifact. picture points representing 1 mm x 1 mm x 4 mm spatial resolution. The scan times were 5 min, the pulse sequence repetition period (TR in Fig. 1) was 0.2 s, and the interval between ir/2 and ir rf pulses (T) was 8 ms. The spectrometer gain and display parameters were the same for each section. Fig. 3 A and B are -CH2- and H20 images recorded with a 40- ms chemical-selective irradiation pulse of amplitude about 1/10th that of the slice-selective ir/2 pulse in interval to. As the latter was of 3.5-ms duration, the power deposited in the sample by the chemical-selective pulse was only about 1/10 that deposited by the slice-selective pulse. The brain images (Fig. 3A) dramatically demonstrate our earlier observations of negligible detectable -CH2- signal from normal brain tissue relative to the H20 signal on the time-scale of the NMR experiment (10, 19). The same is apparently true of calf muscle: trace amounts of -CH2- in the

4 Medical Sciences: Bottornley et al FIG. 4. Difference image constructed from Fig. 3 C-B-A, using the upper-head section (Fig. 3 Right). The display gain is the same as for Fig. 3, but the background is offset to midrange. Artifacts arise from misregistration of constituent images. calf muscle are probably attributable to fascial planes, whereas most of the -CH2- signal derives from surface adipose and bone marrow. The H20 images (Fig. 3B) display signal from both adipose and nonadipose tissue components, reflecting the presence of water in both. However, the adipose water signal is significantly less than that of brain or muscle, consistent with the lower measured water content of adipose tissue (18 + 3% by weight compared to 79 ± 1% for brain and 75 ± 2% for muscle) (20). This is particularly evident in the eye-level scan of the head, where the brain signal is virtually the brightest signal apparent: the curiously stronger H20 signal from the retro-orbital regions may be attributable to shortened spin-lattice relaxation times. Both H20 head scans contain structural and contrast information partially obscured in the other images. The conventional, chemically unresolved images (Fig. 3C) exhibit some chemical shift artifact from the dual 1H components of adipose tissue, discernible as a ghost signal on the left periphery. No such artifact occurs in the brain tissue (10) or in Fig. 3 A and B. Total 'H NMR signal is essentially conserved in the images. Fig. 4 is a difference image of the upper head section constructed from Fig. 3 C-B-A with the same display gain parameter as in Fig. 3 but with the background (noise) display level readjusted to midrange to facilitate comparison of the difference signal with the noise level. Some edge artifacts are obviously due to changes in the orientation of the head between scans, but approximately two-thirds of the brain area in Fig. 4 has a signal amplitude - 5% of that in Fig. 3C. Thus, superposition of Fig. 3 A and B scans reconstructs Fig. 3C within experimental error. DISCUSSION Optimizing the homogeneity of the static magnetic field across the imaging plane during t4 is crucial to the successful implementation of the present technique. Regions of the sample that contribute NMR signal to the periphery of the chemical peaks in Fig. 2 or that are otherwise distributed in the baseline of the chemical shift spectrum are most susceptible to exclusion from chemical-selective irradiation. Artifacts so produced are manifested as unsuppressed H20 or -CH2- signals from these regions, minor evidence of which is exemplified in the images of the upper-head section, where some brain H20 signal appears at the top of Figure 3A and some -CH2- signal from the scalp is visible at the right of Fig. 3B. However, the conservation of the1h signal in Fig. 4 indicates that negligible signal components were irradiated by the chemical-selective pulses in both H20 and -CH2- images, Proc. Natl. Acad. Sci. USA 81 (1984) 6859 attesting to the sharp rectangular profile of these pulses in the frequency domain. Therefore, it is possible to accurately reconstruct either a H20 or a -CH2- image or a combined (H20/-CH2-) image by arithmetic addition or subtraction of the other two. Moreover, the ability to resolve the peaks in Fig. 2 and obtain substantially uniform and artifact-free chemical images indicates that inherent sample diamagnetism does not prohibit the achievement of adequate 'H spectral resolution, at least for heads and limbs. Thus, homogeneity problems are technically soluble with sufficient shim power. The uniformity of the chemical-selective rf magnetic field also affects the ability to suppress H20 or -CH2- resonances but to a lesser extent: it is a problem common to conventional spin-lattice relaxation time (T,) and spin-spin relaxation time (T2) NMR imaging studies (20). Additional difficulties are anticipated when the dynamic range of the chemical shift spectrum exceeds the ability of the chemical-selective pulse to suppress the strongest NMR signals. For example, because the adipose (principally -CH2-) T2 and T, values are approximately equal, in the limit TR << T,/T2, the adipose signal is approximately 8-fold that of muscle (mainly H20) wherein T2 << T, (21). Therefore, 95% suppression of the -CH2- peak would leave a -CH2- component with a significant amplitude of some 40% of that of muscle H2O. Inadequate suppression of intense retro-orbital fat -CH2- signal is another possible source of the corresponding bright regions observed in the eye-level H20 scan (Fig. 3B). Substantial improvement in peak suppression is conceivable by adjusting the chemical-selective pulse to provide for saturation of the NMR signal. A continuously swept saturating rf field applied to the H20 resonance in rat brain at 360 MHz has enabled visualization of millimolar concentrations of metabolites, including lactate, in vivo (7). Adaptation of this saturation technique to chemical shift imaging requires only extension of the t4 interval to occupy the entire timing gap between successive applications of the Fig. 1 sequence. Appropriate adjustment of the frequency excursions of the saturatingrf could possibly eliminate virtually all of the spectrum save the lactate resonance, thereby generating a metabolically useful image with the approximately 15-fold inherent improvement in sensitivity afforded by the 'H resonance over the 31P resonance. Other techniques employing multiple rf pulses (22-24) rather than continuous rf might also be inserted between imaging portions of the sequence: for example the WALTZ-8 decoupling sequence ends with a ir/2rf pulse, which could be replaced by the initial slice-selection ir/2 pulse of interval to (24). Care is necessary to avoid excessive rf power deposition in living samples (25). We thank W. A. Edelstein, J. F. Schenck, H. R. Hart, R. W. Redington, 0. M. Mueller, D. Vatis, and L. S. Smith for valuable contributions to the NMR imaging system. 1. Bottomley, P. A. (1982) Rev. Sci. Instrum. 53, Ross, B. D., Radda, G. K., Gadian, D. G., Rocker, G., Esiri, M. & Falconer-Smith, J. (1981) New Engl. J. Med. 304, Radda, G. K., Bore, P. J., Gadian, D. G., Ross, B. D., Styles, P., Taylor, D. J. & Morgan-Hughes, J. (1982) Nature (London) 295, Edwards, R. H. T., Dawson, M. J., Wilkie, D. R., Gordon, R. E. & Shaw, D. (1982) Lancet i, Neurohr, K. J., Barrett, E. E. & Shulman, R. G. (1983) Proc. Natl. Acad. Sci. USA 80, Prichard, J. W., Alger, J. R., Behar, K. L., Petroff, 0. A. C. & Shulman, R. G. (1983) Proc. Natl. Acad. Sci. USA 80, Behar, K. L., den Hollander, J. A., Stromski, M. E., Ogino, T., Shulman, R. G., Petroff, 0. A. C. & Prichard, J. W. (1983) Proc. Natl. Acad. Sci. USA 80,

5 6860 Medical Sciences: Bottomley et al. 8. Cady, E. B., Costello, A. M., Dawson, M. J., Deply, D. T., Hope, P. L., Reynolds, E. O., Tofts, P. S. & Wilkie, D. R. (1983) Lancet i, Bottomley, P. A., Hart, H. R., Edelstein, W. A., Schenck, J. F., Smith, L. S., Leue, W. M., Mueller, 0. M. & Redington, R. W. (1983) Lancet il, Bottomley, P. A., Hart, H. R., Edelstein, W. A., Schenck, J. F., Smith, L. S., Leue, W. M., Mueller, 0. M. & Redington, R. W. (1984) Radiology 150, Bottomley, P. A., Edelstein, W. A., Hart, H. R., Schenck, J. F. & Smith, L. S. (1984) Magn. Reson. Med., 1, Brown, T. R., Kincaid, B. M. & Ugurbil, K. (1982) Proc. Natl. Acad. Sci. USA 79, Maudsley, A. A., Hilal, H. K., Perman, W. H. & Simon, H. E. (1983) J. Magn. Reson. 51, Pykett, I. L. & Rosen, B. R. (1983) Radiology 149, Bottomley, P. A., Smith, L. S., Edelstein, W. A., Hart, H. R., Mueller, O., Leue, W. M., Darrow, R. & Redington, R. W. (1984) Magn. Reson. Med. 1, Proc. NatL Acad. Sci. USA 81 (1984) 16. Bottomley, P. A. (1981) J. Phys. E. 14, Nunnally, R. L. & Hollis, D. P. (1979) Biochemistry 18, Edelstein, W. A., Bottomley, P. A., Hart, H. R. & Smith, L. S. (1983) J. Comput. Assist. Tomogr. 7, Bottomley, P. A., Foster, T. H. & Darrow, R. D. (1984) J. Magn. Reson., 59, Bottomley, P. A., Foster, T. H., Argersinger, R. E. & Pfeifer, L. M. (1984) Med. Phys. 11, Bottomley, P. A. (1981) Experientia 37, Morris, G. A. & Freeman, R. (1978) J. Magn. Reson. 29, Levitt, M. H., Freeman, R. & Frenkiel, T. (1982) J. Magn. Reson. 50, Shaka, A. J., Keeler, J., Frenkiel, T. & Freeman, R. (1983) J. Magn. Reson. 59, Bottomley, P. A. & Edelstein, W. A. (1981) Med. Phys. 8,