MAGNETIC RESONANCE IN MEDICINE 14,40 1-408 ( 1990)

Single-Shot Localized Echo-Planar Imaging (STEAM-EPI) at 4.7 Tesla

ROBERT TURNER, *,I MARKUS VON KIENLIN, * CHRIT T . w. MOONEN, * AND PETER C. M. VAN Z I J L ~

*Biomedical Engineering and Instrumentation Branch and -f National Cancer Institute, National Institutes of Health, Bethesda, Maryland 20892

Received November 3, 1989; revised February 6, 1990

The resolution and homogeneity limitations of echo-planar imaging (EPI) are over- come by zoom imaging of an easily shimmed localized volume. Use of the stimulated echo enables single-shot localization. In vivo 0.5-mm resolution EPI images of selected regions of a cat brain at 4.7 T are presented. 0 1990 Academic Press, Inc.

INTRODUCTION

Echo-planar imaging (EPI) is one of the fastest possible magnetic resonance im- aging (MRI) methods, allowing the acquisition of the whole time-domain data set in a single scan ( 1 ) . It has important potential for clinical and functional studies, being particularly appropriate for imaging moving organs and for obtaining brain diffusion images free from motion artifact (2). Since all the potential signal is utilized for image formation, EPI is an imaging scheme which is optimal in terms of signal-to-noise. When averaging is necessary to achieve sufficient signal-to-noise ratio, the misleading and widespread motion artifacts of conventional in vivo imaging, which arise from discontinuities in the time-domain data before Fourier transformation, cannot ap- pear. This remains true even if the averaging is performed using the raw time data, although in this case motion and instrumental instabilities may cause some spurious changes in image contrast (which was not observed in the present study).

However, EPI has not yet become a widespread tool in clinical research on com- mercial machines, mainly because of its stringent hardware requirements. The gradi- ent switching times available for many state-of-the-art instruments today still limit acquisition times to a minimum of 50 ms for 64 X 64 images. Since the data are acquired as a series of gradient-recalled echoes, for which neither chemical shift varia- tions nor field inhomogeneity effects are refocused, an excellent Bo -homogeneity, corresponding to a linewidth of perhaps 15 Hz over an entire brain, is prerequisite for EPI imaging. Since the effect of susceptibility variations is proportional to field strength, while the pixel width in hertz depends only on total sampling time (which is constrained in other ways) this goal becomes increasingly hard to achieve for im- aging at high field, desirable though this may be from the point of view of signal-to- noise ratio. Even if such homogeneity were achieved, the spatial resolution in EPI

$ To whom correspondence should be addressed.

40 I 0740-3 194190 $3.00 Copyright 0 1990 by Academic Press, Inc. All rights of reproduction in any form reserved.

402 COMMUNICATIONS

: slice selection aradients : gradient crushers 0 : echo planar imaging gradients

FIG I . Diagram of the imaging sequence used, STEAM-EPI. The three 7r/2 pulses are all frequency- selective in the presence of mutually orthogonal gradients, defining a single VOI of rectangular section. Volume selection is immediately followed by MBEST gradient switching and echo-planar data acquisition.

would still be limited by the array size, which is in turn determined by fundamental hardware constraints. Using a whole-body imaging system, an in-plane resolution of 1 mm is typically the best that is achievable in a single shot by EPI.

One way around both the homogeneity difficulty and the limitation on resolution is inner volume imaging. “Zooming” the volume of interest (VOI) is not only a means of obtaining high resolution images of restricted areas within the sample with- out wraparound artifact (.?, 4 ) , but also makes such areas accessible to EPI because of the better homogeneity generally obtainable in a small volume. Combining the stimulated-echo acquisition mode (STEAM) localization technique ( 4 , 5 ) with EPI ( 6 ) , we obtain images of selected volumes of interest in cat brain. The method and the results are presented, and further possible improvements are discussed.

THE STEAM-EPI SEQUENCE

The STEAM sequence consists of three 90” rf pulses which generate a stimulated echo (Fig. 1 ). When each of these pulses is frequency selective in the presence of mutually orthogonal field gradients, the stimulated echo originates from a VOI which is the intersection of the three selected planes. Unwanted transverse magnetization is dispersed by a gradient pulse in the second TE/2 period (a “TE crusher” (7)). A gradient of equal duration and magnitude is placed in the first TE period to establish a properly refocused stimulated echo of the spins of interest. In the middle period, called the TM period, the spin magnetization lies along the z-axis. A gradient pulse in this period (“TM crusher”) is used to disperse all possible nonstimulated echoes. An excellent degree of localization is thereby assured in a single acquisition ( 7). In the STEAM-EPI sequence, one of the selective pulses determines the image slice thickness, and the two others determine the size of the VOI in the readout and in the phase-encode directions. Give adequate hardware, the imaging parameters can be chosen in such a manner that the field of view (FOV) is as small as the selected VOI,

COMMUNICATlONS 403

FIG. 2. (a) A spin-warp (STEAM) image of a cat brain, acquired using the Acustar 260-mm gradient coil set. The resolution during acquisition was 64 X 128 pixels, interpolated by zero filling to 128 X 128, for a FOV of 80 X 80 mm, yielding a pixel size of 1.25 X 0.63 mm. The slice thickness was 4 mm, the echo time TE = 80 ms. In total, 64 scans ( 1 per phase encoding step) were acquired. The data were sampled at 29 ps intervals, and the readout gradient was 10 mT m-'. (b) The single-shot 64 X 64-pixel EPI image corresponding to the preceding spin-warp image. The acquisition time was 122 ms, with TE = 140 ms. In this image, some distortions due to Bo inhomogeneities are clearly visible, especially in the upper right corner. The longer echo time leads to enhanced T2 contrast. However, the brain structures observed in the two images correspond. The data were again sampled at 29 ws intervals, and the oscillating gradient had a maximum value of 10 mT m -'. Use of much larger values of the oscillating gradient gave rise to unaccept- ably large artifacts arising from data acquired while the gradient was switching.

404 COMMUNICATIONS

COMMUNICATIONS 405

and the spatial resolution available in the readout portion of the sequence is thus optimized. Large, rapidly switchable, magnetic field gradients must be available.

The EPI gradient sequence (Fig. 1 ) modulates the stimulated echo into a series of gradient-recalled subechoes. A negative gradient lobe first dephases the signal, and then the sampling of the NMR signal starts in the presence of one constant (or blipped) gradient and one oscillating gradient. For the image reconstruction, the data acquired during alternate lobes of the oscillating gradient must be time reversed be- fore Fourier transformation. We normally use the MBEST (“modulus blipped echo- planar single-pulse technique” (2, 8)) variation of EPI, shown in Fig. 1, in which all of k-space is sampled, the image being formed from the modulus of the Fourier- transformed time data.

EXPERIMENTAL

Experiments were conducted on a General Electric CSI 4.7-T horizontal bore sys- tem equipped with a GE prototype shielded gradient coil set giving a maximum gradi- ent of 200 mT m-’, and later with an Acustar shielded gradient coil set capable of generating 50 mT m-’. Using the standard gradient amplifiers, each of these gradient coils could be switched in about 180 p s to its maximum gradient strength. Eddy currents were negligible, for the purposes of EPI. An anesthetized cat was placed in the magnet with a surface coil (id = 4.7 cm) placed on its head. Initially, shimming was performed using the stimulated echo with the slice selection gradients turned off. The total integrated signal modulus was used to indicate shim quality, the surface coil being used for both transmission and reception. The homogeneity was then ad- justed, using only the linear shims, for a horizontal slice through the brain of the cat, parallel to the surface coil plane. Both standard spin-warp and EPI images (2) were acquired. The small quantity of fat in the head of a cat, and its relatively short T2, meant that for the echo times used no chemically shifted fat image could be seen. Consequently, it was unnecessary to take precautions to eliminate such a signal by presaturation. From the images, the coordinates of a region of interest were estab- lished. Further local shimming was then performed.

The oscillating gradient for EPI imaging was generated by playing out a table of precalculated values during the data acquisition period, during which data were con- tinuously acquired at a uniform rate. Nonuniform data sampling, though desirable, is not easy to implement on the CSI system. Images were reconstructed in a few seconds using our own software and the NMR instrument hardware. Care was taken to ensure that the echoes remained equally spaced after time reversal of alternate echoes, in order to decrease the intensity of the ghost due to asymmetry between odd and even echoes.

FIG. 3. (a) The zoomed 64 X 128 spin-warp image of a 20 X 20 X 4-mm volume selected in midbrain, interpolated by zero filling to 128 X 128. The field of view is reduced to 30 X 30 mm, thus yielding a resolution of470 X 235 pm. Eight scans were accumulated for each phase-encoding step. In order to obtain an image contrast similar to the EPI images, TE = 140 ms. The data were sampled at 78 ps intervals, with a readout gradient of 10 mT m-’. (b) The zoomed 64 X 64 EPI image of the same selected volume as in Fig. 3. The FOV was 30 X 30 mm, giving a pixel size of 470 X 470 pm. Image distortion due to B, inhomogeneity is not observable on this image. Eight scans were accumulated, with a TR of 1 s. The other imaging parameters were AT = 122 ms, TE = 140 ms. The maximum oscillating gradient was 20 mT m -’.

406 COMMUNICATIONS

FIG. 4. (a) The 64 X 64 EPI image of a section in the x-y plane through the brain of another cat, using the 150-mm prototype gradient set and a birdcage head rf coil. The field of view was 80 X 80 mm. Four scans were averaged, with a TR of 5 s. AT = 4 1 ms, TE = 62 ms. The oscillating gradient had an amplitude of 30 mT m-’. Compared with the 260-mm gradient set, the smaller gradient set switched substantially faster to the gradient values used, thus preventing the “uniform sampling” artifact mentioned earlier. (b) The 64 X 64 EPI image of a selected region of the slice shown in Fig. 4a, with dimensions 20 X 20 X 2 mm, zoomed to a FOV of 40 X 40 mm. Sixteen scans were averaged, with a TR of 5 s. AT = 45 ms, TE = 65 ms. The oscillatinggradient had an amplitude of 55 mT m-’.

RESULTS

The cat brain images shown in Figs. 2-4 were obtained at 200 MHz. After shim- ming, the measured linewidths at half height were about 20 Hz on the whole head without volume selection, and about 9 Hz on the selected volume of 20 X 20 X 4 mm within the brain. Using the STEAM sequence for slice selection and eventually for volume selection, images were acquired with both the standard spin-warp and

COMMUNICATIONS 407

FIG. 4-Continued

the MBEST imaging schemes. The MBEST images were zero-filled before Fourier transformation to yield 128 X 128 pixel images. The STEAM images were zero-filled in the phase-encode direction to give 128 X 128 pixel images. Figures 2 and 3 show images in the x-z plane obtained using the standard Acustar shielded gradient coils, and Figs. 4a and 4b show earlier images in the x-y plane obtained more rapidly using the larger gradients available from the prototype 15-cm-bore shielded gradient set.

DISCUSSION AND CONCLUSION

The efficiency of volume selection as shown by the images (Figs. 2-4) is consistent with that achieved previously for spectroscopic purposes ( 7). Essentially no back- ground signal can be seen in the unselected region, at least for the conventional spin- echo images. MBEST imaging of a zoomed VOI (Figs. 3b and 4b) clearly avoids the

408 COMMUNICATIONS

distortion caused by static field inhomogeneity seen in the unzoomed EPI image (Fig. 2b). The improvement comes partly from the increased readout gradient strength used to decrease the FOV, but mostly arises from the selection of a volume of interest in which the Bo field is already quite homogeneous, and from localized shimming. Fine image details are well visualized and the spatial resolution is certainly the origi- nal pixel size of 0.5 mm. The signal-to-noise ratio is quite acceptable, given the total acquisition time of only 16 s for the 16 scans.

A faint echo-reversal ghost is visible on the MBEST images shown in Figs. 2b, 3b, and 4. This is a common feature of MBEST images and arises from the asymmetry between even and odd echoes after the time reversal of alternate echoes. The time-consuming adjustment of acquisition parameters necessary to minimize the in- tensity of that ghost can be avoided by implementing new echo-planar sampling schemes ( 9 ) .

By obtaining high resolution images of cat brain, we have demonstrated that the field inhomogeneity problem often encountered when using the EPI technique can be overcome through inner volume imaging. The principal hardware factor limiting this technique is the gradient switching time of the instrument.

Instead of the STEAM localization technique, any single-shot volume selection method may be used in combination with EPI. In particular, the PRESS technique, which uses a double echo ( 10, 11 ) may be advantageous, since the full magnetization may be used for imaging. In addition, the latter localization technique is less motion- sensitive. STEAM localization, however, has the advantage that heavy diffusion weighting may be applied while the echo time is kept short ( 12).

ACKNOWLEDGMENTS

The authors acknowledge the invaluable assistance of Daryl Despres and Scott Chesnick. The work was performed at the NIH In Vivo NMR Center.

REFERENCES

1. P. MANSFIELD, J. Phys. C: Solid State Phys. 10, L55 ( 1977). 2. R. TURNER AND D. LE BIHAN, J. Mugn. Reson., in press. 3. P. R. LUYTEN AND J. A. DEN HOLLANDER, “Proceedings, 4th Meeting of the Society of Magnetic

Resonance, London, 1985,” p. 102 1. 4. J. GRANOT, J. Mugn. Reson. 70,488 ( 1986). 5 . J. FRAHM, K. D. MERBOLDT. AND W. HANICKE, J. Mugn. Reson. 72,502 ( 1987). 6. D. FEINBERG AND J. S. HALE, “Proceedings, 5th Meeting of the Society of Magnetic Resonance, Mon-

treal, 1986,” p. 950. 7. P. C. M. VAN ZIJL, C. T. W. MOONEN, J. R. ALGER, J. s. COHEN, AND A. S. CHESNICK, Mugn. Reson.

Med. 10,256(1988). 8. G. JOHNSON AND J. M. S. HUTCHINSON, J. Mugn. Reson. 63, 14 (1985): R. RZEDZIAN AND I. L.

PYKETT, Amer. J. Roentgenol. 149,245 ( 1987); A. M. HOWSEMAN, M. K. STEHLING et ul., Brit. J. Rudiol. 61,822 (1988).

Y. D. FEINBERG, R. TURNER, P. JAKAB, AND M. VON KIENLIN, Mugn. Reson. Med. 13,162 ( 1990). 10. P. A. BOTTOMLEY, U.S. Patent 4,480,228, 1984. I 1. R. J. ORDIDGE, M. R. BENDALL, R. E. GORDON, AND A. CONNELLY, in “Magnetic Resonance in

Biology and Medicine” (Govind, Khetrapal, and Saran, Eds.), p. 387, Tata McGraw-Hill, New Delhi, India, 1985.

12. C. T. W. MOONEN, P. C. M. VAN ZIJL, J. GILLEN, P. DALY, M. VON KIENLIN, J. WOLF, AND J. S. COHEN, “NMR in Biomedicine,” in press.