Magnetic Resonance Imaging of Short T.sub.2 Species

This invention relates generally to magnetic resonance imaging (MRI) and spectroscopy, and more particularly the invention relates to imaging of species having short spin-spin relaxation times, T.sub.2.

Nuclear magnetic resonance (NMR) imaging, also called magnetic resonance imaging (MRI), is a non-destructive method for the analysis of materials and represents a new approach to medical imaging. It is completely non-invasive and does not involve ionizing radiation. In very general terms, nuclear magnetic moments are excited at specific spin precession frequencies which are proportional to the local magnetic field. The radio-frequency signals resulting from the precession of these spins are received using pickup coils. By manipulating the magnetic fields, an array of signals is provided representing different regions of the volume. These are combined to produce a volumetric image of the nuclear spin density of the body.

A descriptive series of papers on NMR appeared in the June 1980 issue of the IEEE Transactions on Nuclear Science, Vol. NS-27, pp. 1220-1255. The basic concepts are described in the lead article, "Introduction to the Principles of NMR," by W. V. House, pp.1220-1226, which employ computed tomography reconstruction concepts for reconstructing cross-sectional images. A number of two- and three-dimensional imaging methods are described. Medical applications of NMR are discussed by Pykett in "NMR Imaging in Medicine," Scientific American, May 1982, pp.78-88, and by Mansfield and Morris, NMR Imaging in Biomedicine, Academic Press, 1982.

Briefly, a strong static magnetic field is employed to line up atoms whose nuclei have an odd number of protons and/or neutrons, that is, have spin angular momentum and a magnetic dipole moment. A second RF magnetic field, applied as a single pulse transverse to the first, is then used to pump energy into these nuclei, flipping them over, for example to 90.degree. or 180.degree.. After excitation the nuclei gradually return to alignment with the static field and give up the energy in the form of weak but detectable free induction decay (FID). These FID signals are used by a computer to produce images.

The excitation frequency, and the FID frequency, is defined by the Larmor relationship which states that the angular frequency, .omega..sub.o, of the precession of the nuclei is the product of the magnetic field, B.sub.o, and the so-called magnetogyric ratio, .gamma., a fundamental physical constant for each nuclear species:

Accordingly, by superimposing a linear gradient field, B.sub.z =z.multidot.G.sub.z, on the static uniform field, B.sub.o, which defines the Z axis, for example, nuclei in a selected X-Y plane can be excited by proper choice of the frequency spectrum of the transverse excitation field applied along the X or Y axis. Similarly, a gradient field can be applied in the X-Y plane during detection of the FID signals to spatially localize the FID signals in the plane. The angle of nuclei spin flip in response to an RF pulse excitation is proportional to the integral of the pulse over time.

A spin echo technique has been employed in obtaining magnetic resonance signals from a body in a nonhomogeneous magnetic field. After nuclear spins are tilted and have been processed for a period of time, T, a 180.degree. refocusing RF field is applied to flip the nuclear spins 180.degree.. After a time period of T, the nuclear spins will refocus, at which time the magnetic resonance signals are detected.

However, there is an inherent delay in conventional slice-selective excitation between the peak of the RF waveform when the bulk of the transverse magnetization is created, and the end of the refocusing lobe when data acquisition can begin. Some species to be imaged have short spin-spin relaxation times (T.sub.2) after which the free induction signal significantly decays. The minimum T.sub.2 species that can be imaged is on the order of this delay time. This delay can be reduced somewhat by using offset RF waveforms and reducing the refocusing time. However the slice profile suffers.

The present invention is directed to providing a method for slice-selective excitation that allows the imaging of very sort T.sub.2 species.

An object of the invention is a method and apparatus for slice-selective excitation that allows the imaging of very short T.sub.2 species.

A feature of the invention is the use of at least two separate excitations and acquisitions of data with the data summed to provide a signal for a desired slice.

Another feature of the invention is the use of two separate excitations where each excitation corresponds to half of a conventional slice-selective excitation.

Briefly, in one embodiment of the invention two excitations are applied in sequence with data acquisition occurring after the application of each excitation. The two excitations correspond to half of a conventional slice-selective excitation, a sinc pulse for example, and the same acquisition gradients are used for each of the half excitations. The second half excitation is applied after data acquisition is completed for the first half excitation. The data acquisition following each of the excitations are then added to provide a signal for the desired slice.

This method has a number of advantages. First, for long T.sub.2 species the slice provide is the same as that for conventional slice selection. No sacrifice in slice profile is required for long T.sub.2 species. Second, the slice profile for short T.sub.2 species is limited only by T.sub.2 and the gradient strength. Third, no refocusing lobe is required. Data acquisition can begin immediately, limited only by the time it takes to shut down the RF transmitter.

The invention and objects and features thereof will be more fully understood from the following detailed description and appended claims when taken with the drawings.

FIGS. 1A-1D illustrate the arrangement of conventional MRI apparatus and magnetic fields generated therein.

FIG. 2 is a functional block diagram of MRI imaging apparatus.

FIG. 3 illustrates a conventional basic pulse sequence for imaging.

FIGS. 4A and 4B are a conventional slice-selective excitation and its k-space interpretation, respectively.

FIG. 5 is an excitation pulse in accordance with the invention.

FIGS. 6A-D illustrate the effect of T.sub.2 decay and off-resonance frequency on the slice profile of the short T.sub.2 slice-selective pulse.

FIG. 7 illustrates an RF waveform for a numerically optimized short T.sub.2 pulse with the tip angle designed to be 90.degree..

FIGS. 8A-C are slice profiles for the pulse for FIG. 7.

FIGS. 9A and 9B illustrate the ideal gradient and RF waveforms and the corresponding variable-rate waveforms which produce the same sinc weighting of k-space.

FIG. 10 illustrates pulse sequence for measuring the slice-profile of the short T.sub.2 slice-selective pulse.

FIGS. 11A and 11B illustrate acquired data from the two half k-space excitations with the tip angle being 30.degree..

FIG. 12 illustrates a plot of data resulting from adding the two acquisitions in FIGS. 11A and 11B (solid and dashed plots).

FIG. 13 illustrates a slice profile obtained by transforming the data of FIG. 12 (solid and dashed plots).

FIG. 14 illustrates a pulse sequence used to image short T.sub.2 species in which the proposed selective excitation pulse is followed immediately by projection-reconstruction acquisition gradients.

FIGS. 15A-D are cross-sectional images of six test tubes containing MnCl solutions with T.sub.2 's of 20, 20, 5,3,2, and 1 ms.

FIG. 16 is a pulse sequence for conventional sliceselective spin-warp imaging.

Referring now to the drawings, FIG. 1A is a perspective view partially in section illustrating coil apparatus in an NMR imaging system, and

FIGS. 1B-1D illustrate field gradients which can be produced in the apparatus of FIG. 1A. This apparatus is discussed by Hinshaw and Lent, "An Introduction to NMR Imaging: From the Bloch Equation to the Imaging Equation," Proceedings of the IEEE, Vol. 71, No. 3, March 1983, pp. 338-350. Briefly, the uniform static field B.sub.o is generated by the magnet comprising the coil pair 10. A gradient field G.sub.x is generated by a complex gradient coil set which can be wound on the cylinder 12. An RF field B.sub.1 is generated by a saddle coil 14. A patient undergoing imaging would be positioned along the Z axis within the saddle coil 14.

In FIG. 1B an X gradient field is shown which is parallel to the static field B.sub.o and varies linearly with distance along the X axis but does not vary with distance along the Y or Z axes. FIGS. 1C and 1D are similar representations of the Y gradient and Z gradient fields, respectively.

FIG. 2 is a functional block diagram of the imaging apparatus as disclosed in NMR - A Perspective on Imaging, General Electric Company, 1982. A computer 20 is programmed to control the operation of the NMR apparatus and process FID signals detected therefrom. The gradient field is energized by a gradient amplifier 22, and the RF coils for impressing an RF magnetic moment at the Larmor frequency is controlled by the transmitter 24 and the RF coils 26. After the selected nuclei have been flipped, the RF coils 26 are employed to detect the FID signal which is passed to the receiver 28 and thence through digitizer 30 for processing by computer 20.

FIG. 3 is a graph illustrating applied RF pulses for a 180.degree. flip angle and a 90.degree. flip angle followed by a "time reversal" 180.degree. pulse and the associated Z axis magnetic field gradient applied during application of the RF pulses. The illustrated pulses are conventional sinc pulses which have front and back lobes with a main RF pulse therebetween.

As noted above, here is an inherent delay in conventional slice-selective spin-warp imaging between the peak of the RF waveform when the bulk of the transverse magnetization is created, and the peak of the echo when most of the energy in the received signal is acquired. This delay is called the echo time T.sub.E, and is illustrated in FIG. 16. The minimum T.sub.2 species that can be imaged is on the order of this delay time.

This delay can be reduced within the slice-selective spin-warp framework by using an echo off-center in the data acquisition window, and by using an offset RF waveform with a reduced refocussing lobe. The offset echo requires a partial k-space reconstruction, which is not a fundamental problem. The offset RF pulse with the reduced refocussing lobe results in a degraded slice profile. This is fundamental.

We propose here a method for slice-selective imaging that allows the T.sub.E to be greatly reduced. The basic idea is that we will simply eliminate that part of the pulse sequence that contributes to T.sub.E. These are the second half of the slice select gradient, the refocussing lobe, the phase-encode lobe, the dephaser lobe, and the first half of the readout lobe. In FIG. 16 this corresponds to eliminating everything between the dotted lines. The T.sub.E is then potentially zero, although practical issues limit T.sub.E to a few hundred microseconds on our 1.5T GE Signa System.

There are two problems with this pulse sequence: how to perform data acquisition, and how to perform slice selection. The data acquisition problem is easily solved by using projection-reconstruction readout gradients. Then the low frequency data is always collected immediately. The slice selection problem is the subject of this paper.

The present invention is a method which defines a slice by a sequence of at least two separate excitations. Each of these corresponds to half of a conventional slice-selective excitation. The same acquisition gradients are used for each of these two half excitations, and the signals added. The resulting signal is that of the desired well defined slice.

This slice-selection method has a number of advantages. First, for long T.sub.2 species the slice profile is the same as that for conventional slice selection. No sacrifice in slice profile is required for long T.sub.2 species. Second, the slice profile for short T.sub.2 species is limited only by T.sub.2 and the gradient strength. Third, no refocusing lobe is required. Data acquisition can begin immediately, limited only by the time it takes to shut down the RF transmitter.

One disadvantage is some slice profile degradation due to off-resonance effects. This is not usually a significant problem in practice, as we will show in a simulation.

The proposed pulse is most easily understood in terms of a k-space analysis as described by Pauly, Nishimura and Macovski in J. Magn. Reson., to appear, (1988). Based on a small-tip-angle approximation, selective excitation is interpreted as a weighted trajectory through k-space. The slice profile is the Fourier transform of this weighted trajectory. The new excitation pulse may also be viewed as the consequence of the symmetries of the Bloch equation. This approach will be used in a subsequent section when the extension to large-tip-angle pulses is presented.

In conventional slice-selective excitation the k-space trajectory starts at some k.sub.min, passes through the k-space origin at the peak of the RF, goes to a k.sub.max, and returns to the origin with the refocusing lobe. See FIG. 4. The pulse sequence we propose scans k-space in two halves, as is illustrated in FIG. 5. The first half excitation starts at k.sub.min and ends at the k-space origin, right at the peak of the RF. At this point the read gradients can be applied and data collected. Next the second half excitation is applied, starting at k.sub.max and ending at the origin. This is the same as the first excitation, but with a negative gradient. The same read gradients are applied and data collected. The two acquired signals are added. In the absence of T.sub.2 decay and chemical shift the resulting signal is the same as that from a conventional slice-selective excitation.

Several advantages have been obtained, however. No refocusing is required. The readout occurs immediately after half of the main RF lobe has been applied. This is when the bulk of the transverse magnetization is created. The "echo time" or T.sub.E can be on the order of the RF transmitter shutdown time--about 200 .mu.s on our GE Signa system. Now the only factors limiting the visibility of short T.sub.2 species are the decay during excitation, which limits the sharpness of the slice profile for that species, and decay during readout, which for a given readout gradient limits the resolution with which that species can be imaged.

The dependence of the slice profile on T.sub.2 can be shown analytically by introducing a T.sub.2 term into the analysis of Ref. Pauly et al., supra. It is more easily seen conceptually. The k-space weighting of high spatial frequencies occurs early in the pulse, and decays away with time constant T.sub.2 until data acquisition occurs. The effective RF is then the applied RF weighted by the exponential T.sub.2 decay

where the excitation starts at -T and ends at zero. In the small-tip-angle regime the slice profile is the Fourier transform of the weighted k-space trajectory. This will be the infinite T.sub.2 slice profile convolved with the Lorenzian line-shape of the short T.sub.2 species. For very short T.sub.2 species the slice profile will be dominated by T.sub.2 decay. Since .gamma.G is the velocity through k-space, the product .gamma.GT.sub.2 is the spatial frequency at which the k-space weighting has fallen to 1/e. Its reciprocal (.gamma.GT.sub.2).sup.-1 is the ultimate 3 dB slice-width achievable for a given T.sub.2 and gradient strength.

To illustrate the effect of T.sub.2 on slice profile the pulse illustrated in FIG. 2 was simulated with no T.sub.2 decay and with T.sub.2 's equal to one and two times the component pulse duration. The slice profiles are plotted in FIG. 6. The infinite T.sub.2 slice profile is plotted in (a). If the component pulses are 2 ms in length then plots (b) and (c) correspond to T.sub.2 's of 2 and 1 ms. The effect of short T.sub.2 values is a reduction in the sharpness of the slice profile and some loss in signal.

The short T.sub.2 slice-selection pulse has some sensitivity to off-resonance frequency. This is minimized by keeping the component pulses short relative to the period of the off-resonance frequency. For example, at 1.5 T the period resulting from 1 ppm inhomogeneity is about 15 ms. Most component pulses will be a factor of 10 shorter than this, causing no problems. The period resulting from the 3.6 ppm lipid shift is about 4.5 ms. Most component pulses will be half to a quarter of this in length. The result is that the slice profile at the lipid frequency will be somewhat degraded. Plot (d) in FIG. 6 is the slice profile resulting from the pulse illustrated in FIG. 2 at one-half cycle off-resonance over the duration of the component pulse. This corresponds to an off-resonance frequency of 250 Hz or 3.9 ppm at 1.5 T for 2 ms duration component pulses. There is some moderate degradation in the slice profile. In addition there is a constant phase shift that is not shown in the plot.

The basic idea of adding the data acquired from two separate excitations to define a slice can be extended to large-tip-angle pulses. In this case the analysis is based on the spatial symmetry of the Bloch equation when excited by a real pulse. The spatial response of any real-valued pulse possesses skew-Hermitian symmetry (2), ##EQU1## If we ignore off-resonance effects, then the sum of the original slice and the slice obtained with the gradient negated is ##EQU2## Because the M.sub.z component is not observed, one need optimize only the My component of the response when designing an RF waveform.

From this viewpoint off-resonance effects can be considered to be misregistration of the two component slices. An off-resonance frequency .delta. corresponds to a spatial shift of .DELTA.z=.delta./.gamma.G. This provides another measure of whether off-resonance effects will be significant for a particular pulse. If .DELTA.z is small relative to the slice width the slice profile degradation should be minimal. The principle contribution to the slice profile degradation will be imperfect cancellation of the M.sub.x components of the slices produced by the two pulses. This is given approximately by ##EQU3## The new aspect of the pulse design in the large-tip angle regime is the nonlinearity of the Bloch equation. There are several computational techniques available for the nonlinear inversion of the Bloch equation. The most general technique is based on optimal control theory. Suppose the RF pulse is a piecewise-constant waveform with N independent values, and that the user has defined a scalar measure of the fitness of a pulse--usually the integrated squared distance between the desired and the achieved slice profile. The optimal control algorithm allows for fast (order N) computation of the gradient of some scalar measure of the fitness of a pulse to the desired task. Once the gradients of the measure are computed, the pulse can be iteratively optimized with any number of nonlinear programming tools. We have used the software package NPSOL, which computes an approximation to the Hessian of the "goodness" measure with respect to each of the points of the RF. The Hessian is used to approximate a second-order optimization algorithm. Convergence is achieved in about 20 iterations.

To exploit the flexibility in M.sub.x and M.sub.z for the interleaved excitation problem discussed here, we employ a measure of goodness that penalizes only the deviation of the M.sub.y from a desired slice. Consider a simple example where in-slice we desire that M.sub.y be close to one, whereas M.sub.z and M.sub.x should be close to zero. Suppose that a certain pulse achieves a magnetization profile at a position with total error in M.sub.x and M.sub.z of magnitude 0.2. Because magnetization length is preserved, the My component must be .sqroot.1-0.2.sup.2 .apprxeq.0.98. Hence, in this case, a 20% error in two components implies only a 2% error in the other component. This geometrical asymmetry should be exploited in the design of interleaved pulses because we only care about the M.sub.y component in the slice.

The final design specifications to consider are bandwidth, pulse duration, and slice-edge sharpness. For a fixed-duration pulse, the energy and slice thickness increase with the bandwidth, but the slice profile improves. We settled on a compromise solution with intrinsic bandwidth equivalent to a sinc pulse with 3 zeroes on the left side of the pulse. The initial pulse was chosen by the k-space analysis. The transition regions were unpenalized in the measure of goodness, and the width of the transition was varied. The best pulse is shown in FIG. 7. Its simulated response is plotted in FIG. 8. Only the My component contributes to the resultant slice.

The excitation pulse as shown in FIG. 5 is difficult to implement. The gradient must go to zero instantly at the end of the RF waveform. In practice the maximum gradient slew rate is limited, on our GE Signa system to 2 G/cm/ms. A practical gradient would be constant during most of the pulse, and then ramp down to zero at the slew rate limit. There are then two choices. One is that the RF can be confined to the constant part of the gradient waveform. The gradient down-ramp then defocuses the slice, and must be compensated with a small refocusing lobe. This wastes time while the signals of interest are decaying. A better solution is to play the RF during the down-ramp, but to compensate the RF amplitude to insure that the proper k-space weighting is still achieved. This is simply an application of the variable-rate, or VERSE principle. This is the solution we use here. Both the RF waveform and the slice select gradient end at the same time, and data acquisition can start as soon as the RF transmitter can be shut down.

The VERSE calculations for variable-rate-gradient RF pulses are generally done numerically. In this case the gradients are sufficiently simple that the calculation of the required RF waveform may easily be done analytically. Ideally an RF waveform B.sub.1 (t) is played in the presence of a gradient with amplitude G.sub.0. The gradient switches off instantly at time zero, the end of the pulse. Time and spatial frequency are related by the expression ##EQU4## For this ideal gradient, time and spatial frequency are linearly related

so the RF weighting corresponding to a specific spatial frequency is B.sub.1 (k/.gamma.G.sub.0).

The gradient we wish to use is initially at G.sub.0 and ramps down to zero at the slew-rate limit, reaching zero at time zero. If the slew-rate limit is S the gradient is constant until time ##EQU5## and then ramps down linearly. The gradient may be written The spatial frequency as a function of time is ##EQU6## The new RF may now be computed as a change of variables, ##EQU7## The variable rate RF waveform is then ##EQU8## The variable-rate gradient and the RF waveforms t produce a sinc k-space weighting are illustrated in FIG. 9, along with the corresponding waveforms for the ideal unrealizable waveforms.

Another practical note is that demodulators used in magnetic resonance imaging often have a DC bias error. This can be suppressed by alternating the sign of the RF on every other excitation, and subtracting each pair. This can also be done at no additional cost for the short T.sub.2 slice-selection pulse. Rather than just inverting the gradient on the second component excitation, both the RF and the gradient are inverted. The data corresponding to the two component excitations are then subtracted rather than added.

In this section we present experimental verification of the short T.sub.2 selective excitation pulse. This demonstrates several points. First, that the two component excitations do in fact each cover only half of k-space, and that the two halves are complementary. Second, that the variable-rate RF and gradient derived in the previous section do produce the desired k-space weighting. Finally, that the slice profile produced is the one expected analytically.

The pulse sequence used to measure the slice profile is simply the excitation pulse applied along the z-axis, followed by dephase and readout gradient lobes along the same axis. This is illustrated in FIG. 10. The raw data acquired is then the spatial frequency weighting produced by each component excitation pulse. The slice profile is obtained by adding the data from the two component excitations, and computing the Fourier transform of the result.

Sample data from the two component excitations are shown in FIG. 8. Each component excitation only covers half of k-space, ending at the k-space origin. Very little transverse magnetization is created at unexcited spatial frequencies. In this experiment the slice select gradient is only 0.15 G/cm for 2 ms, so that the slice profile can be resolved with a readout gradient of reasonable amplitude and duration. The readout gradient is 0.9 G/cm for 2.66 ms. Data is acquired at 48 kHz, giving 256 samples per acquisition. The RF amplitude was scaled to produce a 30.degree. tip angle on resonance for each of the component pulses. The experiments were performed on a 1.5 T General Electric Signa system, equipped with shielded gradient coils. The standard anti-aliasing filter has been replaced to allow sampling at higher data rates.

The combined data from the two component excitations are shown in FIG. 9, where we have zoomed-in on the excited region of k-space. Also shown is the theoretical sinc k-space weighting, scaled to match the width and amplitude of the acquired data. The agreement is quite good.

The experimental slice profile is obtained by performing a Fourier transform of the acquired data of FIG. 12. The result is shown in FIG. 13, along with the Fourier transform of the sinc weighting. Again the agreement is quite good. The slice width at half amplitude is 2.7 cm. The ripple in the passband of the pulse is due to the choice of a sinc RF pulse. Any other RF waveform could be substituted.

In the previous section we have demonstrated that the proposed slice selection method does produce the expected slice profile. Here we demonstrate its utility for imaging short T.sub.2 species.

The pulse sequence used in this section is simply the proposed slice selective excitation pulse followed by a projection-reconstruction acquisition gradient sequence. The reason for using the projection-reconstruction sequence is that data can be collected immediately, starting at low frequencies. Moderate T.sub.2 decay during acquisition will reduce resolution, but not the imaged signal level of a short T.sub.2 species.

As was the case for the excitation gradient, the read gradients cannot change levels instantly. Gradient ramps must be introduced to keep the slew rates within the capabilities of the gradient amplifiers. In this case the gradients start at zero and ramp up to the desired read levels. The slew rate is chosen so that the slew-rate limit is reached when the read gradient falls along one of the gradient axes.

The low frequency data is acquired on the up-ramp of the read gradients. This corresponds to non-uniform sampling of k-space. The reconstruction algorithm compensates both for this non-uniform sampling, as well as the polar-to-rectangular mapping inherent in projection-reconstruction data collection. For the images presented here the reconstruction algorithm consists simply of mapping each point in the desired rectangular grid into the acquired projection-reconstruction data, and interpolating to find a data value. The resulting rectangular data grid is then transformed to produce an image.

The pulse sequence is illustrated in FIG. 14. There is a delay between the end of the excitation pulse and the beginning of the read gradients. This is can be adjusted to introduce a specific T.sub.2 weighting.

The images in FIG. 15 are a cross-section through six test tubes, each containing MnCl solutions with different T.sub.2 's. The T.sub.2 's are 20, 10, 5, 3, 2, and 1 ms, arranged as illustrated in the lower right. The slice thickness is 7 mm, the field of view of 24 cm, and data collection is 1.58 ms in duration at a rate of 48 kHz. The images are reconstructed on a 256 by 256 grid and have a resolution of slightly better than 1 mm. Only the central 12 cm of the images are shown. The upper two images were acquired using two different delay times between the end of the RF pulse and the beginning of the read gradients and data acquisition: 250 .mu.s in the upper left image and 750 .mu.s in the upper right image. Both images clearly show all the different T.sub.2 solutions. Only the 1 ms solution shows a significant reduction in signal intensity.

An image that is selective for short T.sub.2 species may be formed by subtracting these two images. The result is shown in the lower left image. The 1 ms test tube is the brightest, while the 10 and 20 ms test tubes are barely visible. This method only works when the object has only a single spectral component. With more than one spectral component, such as water and lipid, the signals from the various components will interfere differently at the two different delay times. This will produce a contribution to the difference image even in the absence of T.sub.2 decay. A more sophisticated approach is required in this case.

The present invention presents a new slice selective excitation pulse which allows very short T.sub.2 species to be imaged without any penalty in slice profile. The experimental images show both that the slice profile is the same as that for conventional slice-selective excitation, and that T.sub.2 's on the order of 1 ms can easily be imaged on a commercial imaging system.

While the invention has been described with reference to specific two-pulse embodiments, the description is illustrative of the invention and is not to be construed as limiting the invention. Various modifications and applications may occur to those skilled in the art without departing from the true spirit and scope of the invention as defined by the appended claims.