Rapid MRI and Applications

From Taveras' et al. "Radiology"

by Mark Cohen

Introduction

The strength of the NMR1 signal reflects the rate of quantum state changes of protons - particles of infinitesimal size. Remarkably, however, these transitions occur very slowly, on a time scale of tenths of seconds to seconds for most biological materials. In this chapter, we shall see how the magnetic relaxation rate constants T1 and T2, which describe the rates of these quantum state changes, impose limitations on the practical speed of the MR imaging experiment. These limitations cause MRI as currently practised to be among the slowest of the diagnostic imaging modalities. Our primary focuses, however, will be on techniques developed in recent years to overcome some of these speed restrictions, on consequent changes in MR image contrast, and on applications the techniques engender. In particular, the reduced flip angle methods, known generally by the acronyms FLASH, GRASS, CE-FAST and MP-RAGE, will be discussed. Next, methods which improve the efficiency of the MR experiment by increasing the portion of the MR image that is collected after each excitation will be considered. Finally, the imaging results of the various methods will be compared.

The period from 1985 to the present has seen an explosive growth in the range and power of high speed MR imaging methods. Consequently, it is possible to discuss only a few of the diverse techniques designed to increase MR acquisition rates in a single, short chapter. Threrefore, the goal here is to provide a basis for understanding developments that shall continue to emerge in pace with the importance of MRI in radiology.

Determinants of Imaging Time

Spatial Encoding, Bandwidth and T2: why does MRI take so long?

In the basic MR experiment - in which a radio frequency (RF) excitation pulse elicits an RF signal from a magnetized sample - the MR signal decays rapidly. Its decay rate, T2, is only several tens of milliseconds for most biological samples. To collect the data for a complete image, as demonstrated in this chapter’s appendix, the limitations imposed by practically achievable gradient amplitudes and by analog-to-digital convertors require that the MR signal be sampled for a total of about one or two seconds. Since, following a single excitation, the signal is not present for this length of time, the data sampling is typically divided across one hundred or more separate excitation and data sampling cycles, or "data lines." Imaging time, in conventional imaging sequences, is therefore calculated as:

Reducing the Number of Phase Encodings

One simple strategy by which to shorten imaging time might be to reduce the number of phase encoding lines. If appropriate for the desired spatial resolution, and for the shape of the object, this is a simple and extremely effective way of reducing scan time. Typically, when isotropic resolution (that is, square pixels, or equal resolution) is desired, this will result in an oblong field-of-view. Care must be taken in this case to ensure that significant MR signal is not present from the patient outside of the field of view; otherwise aliasing (wrap around) artifacts may result. Alternatively, when isotropic resolution is not required, simply choosing a slightly lower resolution matrix is often a practical alternative. At this writing, many clinical users favor imaging matrices with 192 phase encode steps, as opposed to the more traditional 256, for the resulting 25% reduction in scan time. Two major consequences result from this strategy: on the one hand, the SNR is actually somewhat higher due to the larger voxels; but on the other hand the spatial resolution is compromised. Figure 1 shows the changes in resolution and signal to noise ratio as the imaging matrix is increased from 128 to 256 phase encode lines, producing rectangular pixels. Figure 2 shows the rectangular matrix (strip scan) technique, in which isotropic resolution is achieved for an oblong field-of-view.

Spin Echoes and Gradient Echoes

For this chapter, we will consider the spin echo technique introduced in chapter 1.32 to be "conventional" MR imaging. As outlined in that chapter, the MR signal is first formed using a 90° radio frequency pulse and, as it decays, the signal is recovered by a 180° pulse that forms a "spin echo." Many of the methods outlined below do not utilize a 180° pulse at all, using instead carefully timed activity in the imaging gradients in order to recover the MR signal. Though it may be something of a misnomer, such techniques are generally called "gradient echo" methods. Figure 3 below represents a family tree of the methods discussed in this chapter. Details of the differences in gradient echo and spin echo scanning will be discussed below.

Shallow Flip Angle Methods

Effects of Short TR on Spin Echo Scans

Decreasing the TR is an obvious means of reducing the overall scan time; and doing so has a variety of consequences, both positive and negative. Longitudinal magnetization (that is, crudely speaking, the extent to which the patient is "magnetized") recovers at the characteristic rate known as T1 , and is destroyed completely following each 90° RF excitation pulse. These T1 times are relatively long - on the order of seconds - so that complete re–magnetization may require several seconds after an excitation pulse. If the excitation pulse is repeated before complete longitudinal re–magnetization (i.e. with a TR short compared to T1), the strength of the MR signal following the next pulse will be reduced. Equation 2 (see chapter 1.32, equation 11) is the formula for the MR signal intensity on spin echo scans, as a function of TR, T1, TE, and T2.

where SI is the signal intensity, TR the repetition time, TE the echo time (the time between the RF excitation pulse and data collection), T1 and T2 are tissue relaxation time parameters, and b is a proportionality constant that includes all instrumentation parameters, such as magnet field strength, RF coil efficiency, et cetera, as well as proton density dependence. Tissues differ greatly in their T1 recovery times and will therefore produce differing signal intensities: tissues with shorter T1 times will yield signals stronger than those with long T1’s. This principle, in fact, forms the basis of T1 contrast in MR imaging (see chapter 1.32). The effect of T1 on the MR signal intensity (the T1 contrast) is minimized when TR is long compared to T1, under which conditions the term 1-exp(-tr/T1) is approximately equal to 1. The T1 contrast is not always desirable; in particular, it compromises the usable contrast of T2-weighted images, which require that the dominant contrast mechanism be T2-, rather than T1-based. Figure 4 compares the contrast and signal-to-noise ratio as TR is decreased in both short and long TE scans.

Beyond the shorter imaging time, and the increase in T1 contrast, decreasing the TR also results in an overall loss in signal, and thus in SNR, especially for long T1 tissues (note that when TR is not large compared to T1, the term, 1-exp(-tr/T1), will be less than one). There will also be a decrease in volume coverage: multi-slice MR uses the time available between repetitions on a single slice to collect data from other slices. As the TR is reduced, the time available for multi-slice imaging is also decreased, and fewer slices may be collected. Often, in conventional imaging, the choice of TR for T1-weighted scans is a tradeoff between desired contrast and volume coverage.

Shallow Flip Angle Excitation

The RF flip angle is defined by the extent to which longitudinal magnetization becomes rotated into the transverse plane (and vice versa). FLASH and GRASS are representative of the gradient echo family of scanning methods, and use RF flip angles of less than 90° as excitation pulses. To introduce these reduced flip angle techniques, we will first present FLASH (Fast Low Angle SHot) scanning, the first such strategy to gain wide acceptance (Frahm, Haase et al. 1986). We will then explore some of the important variants of the method, such as GRASS (sometimes called FISP), that result in substantially different contrast behavior.

Recall (cf. chapter 1.32, figure 6) that the strength of the MR signal is reflected in the extent of the total magnetization in the transverse (or "imaging") plane. As shown in figure 5, the total magnetization is represented by the heavy arrow and includes a longitudinal (vertical) and transverse (horizontal) component. Prior to any RF excitation, all of the magnetization is longitudinal; the arrow would be vertical. If, for example, a 90° pulse were used, a would equal 90° and the magnetization would be rotated into the transverse (horizontal) plane.

Recall (cf. chapter 1.32, figure 6) that the strength of the MR signal is reflected in the extent of the total magnetization in the transverse (or "imaging") plane. As shown in figure 5, the total magnetization is represented by the heavy arrow and includes a longitudinal (vertical) and transverse (horizontal) component. Prior to any RF excitation, all of the magnetization is longitudinal; the arrow would be vertical. If, for example, a 90° pulse were used, a would equal 90° and the magnetization would be rotated into the transverse (horizontal) plane.

Note in the figure, that during the RF excitation pulse, the arrow is rotated towards the transverse plane. As the flip angle is increased, the projection of the magnetization onto the transverse plane, and therefore the signal intensity, is increased. Given that shallow flip angles yield less signal, of what use is a shallow flip technique? The answer lies in the realization that, with small flip angles, less time is required for the protons to re-magnetize after the excitation pulse. This allows for adequate longitudinal recovery, and therefore adequate signal, at shorter TR.

Figure 6 shows the relatively small loss of longitudinal magnetization that occurs when shallow pulse angles are used. It is only this small amount of longitudinal magnetization that must be recovered during the TR period.

Differences in longitudinal recovery between tissues represent T1 contrast. As a consequence, in FLASH imaging, shorter TR’s may be used to achieve images without T1 contrast. Figure 7 provides a key to understanding FLASH contrast behavior:

Here we can see that for long repetition times (plotted as the ratio of TR to the tissue T1, since a "long" repetition time depends upon the T1 of the tissue), larger flip angles produce greater signal. As the TR is decreased however, the shallow flip angles yield more signal. This is because, at shorter TR’s, the magnetization does not have the time to recover fully between repeated RF excitation pulses. In the very short TR regime, with TR less than 1/10th of T1, a 20° flip angle yields several times the signal intensity of a 90° flip angle. Note, on the other hand, that the maximum signal strengths can be achieved only with a 90° flip angle, and at long TR.

Analytically, the signal intensity in the FLASH technique is described by the formula:

where a is the flip angle and b is as defined above (equation 2). As we will see below, in shallow flip angle imaging, TE is defined as the time between the RF excitation pulse and the collection of the image data (note that this definition may differ slightly from that used for conventional spin echo imaging). In this equation, note that the effect of TE is to control the "T2*" contrast, as opposed to the "T2" contrast seen in conventional imaging. The differences in the two are discussed in more detail below in the context of the Spin Echo (see also Chapter 1.32, figure 13).

In MRI, SNR is always a limited quantity. It is a function primarily of the strength of the received signal; the noise level is relatively constant. To make matters worse, contrast is generally achieved only by reducing the signal (and thus the SNR) of one tissue compared to another. The maximum possible SNR is therefore achieved only with a 90° flip angle and a long TR - a condition of minimal contrast. As presented above, this means long imaging times. The virtue of FLASH lies in the fact that, when shorter TR times are used for whatever reason, a shallower flip angle can yield an SNR that is higher overall than that with a 90° flip angle for comparable TR.

T1 contrast and RF flip angle

The reason that reducing the RF flip angle results in more signal is that it reduces the dependence of the signal on tissue T1; for flip angles that are sufficiently small, the signal intensity becomes effectively T1-independent. One obvious consequence is that T1 contrast is reduced in shallow flip angle methods. Whereas in conventional methods the T1 contrast depends primarily on the TR, it depends on flip angle as well in the case of FLASH scanning. In practice it becomes possible to trade TR against flip angle to achieve any desired degree of T1 contrast, with the caveat that shorter scan times will result, generally, in reduced SNR. For purposes of contrast, the following are roughly equivalent:

Figure 8 compares FLASH contrast in clinical MR images as TR and flip angle are adjusted. Note the apparent ability to trade TR against flip angle for purposes of contrast and the variation in SNR as the scan time (TR) is reduced.

The Spin Echo

In the traditional MR image acquisition, a 90° and 180° RF pulse pair are used to form a spin echo, and a set of three, mutually orthogonal, magnetic field gradients is used to encode the signal spatially (chapter 1.32). Use of the spin echo serves a variety of purposes: by altering the direction of Larmor precession, but not its velocity, the spin echo technique tends to minimize so-called T2* losses in signal which result from variations in local magnetic field (and thus spin frequency). Signal decay is thus determined primarily by intrinsic tissue properties (T2). The spin echo technique also allows the sampling of the MR signal to be delayed somewhat from the excitation pulse, leaving adequate time for spatial encoding before the signal is collected.

180° pulses are not used in FLASH scanning

The advantages of shallow flip angle excitation cannot (though there are some clever exceptions) be realized when 180° RF pulses are used to form spin echoes. Recall that the FLASH method works, in part, because the longitudinal magnetization is relatively undisturbed by a shallow excitation pulse. By contrast, one effect of a 180° pulse is to invert the longitudinal magnetization. Rather than reducing the time required for longitudinal recovery, the use of a shallow excitation flip angle in combination with 180° pulses actually prolongs the time required for signal recovery, because the longitudinal magnetization must now recover, not from zero, but from a negative initial value. When the 180° pulse is omitted, as it must be for shallow flip angle imaging, the MR signal decays at the rate T2* rather than at the T2 rate (see Chapter 1.32, Figure 13).

Contributors to T2*

Susceptibility

Even when the field supplied by the external magnet is highly uniform, the magnetic field in the human body varies substantially. Susceptibility (usually denoted as c) is a measure of the extent to which a tissue magnetizes (specifically, it is the ratio of the magnetization of the tissue to the applied magnetic field). Because tissues differ in their susceptibilities, the field strength experienced by the protons, and thus their spin frequency, will differ from one tissue to the next. If the spin frequency differs by as little as one part per million (ppm), the protons of these tissues will become 180° out of phase with each other in only 23 msec in a 1 Tesla magnet and only 15 msec in a 1.5 Tesla magnet. This implies that, at the interface between these tissues where such variations in susceptibility exist, there will be no signal left after only 15 to 20 msec. A familiar sight in FLASH images is a dark border surrounding organs; this is due primarily to the dephasing that occurs at that interface, resulting from the locally differing magnetic fields.

Dramatic variations in susceptibility are also found within tissues having a complex macroscopic structure. The fine trabeculations in bone, for example, result in large non-uniformities in the magnetic field and in very short T2* decay times: the internal cavities of normal bone will appear dark in most FLASH images. Of perhaps greater clinical significance, the air-filled regions in the temporal bone (e.g. the internal auditory canal) and the suprasellar region, including the pituitary fossa, may show reduced signal intensity as compared to spin echo images. A FLASH sequence with an ultra short TE (e.g. 5 msec or less) will show better performance than a longer TE FLASH sequence in these regions. The auditory nerves in particular are very difficult to image with simple gradient echo techniques. The susceptibility effects are a function of the field strength of the magnet: the same loss of signal due to susceptibility differences will occur in 9 msec at 1.5 Tesla, 14 msec at 1.0 Tesla and 27 msec at 0.5 T. Figure 9 shows typical susceptibility-related signal losses in the sella with gradient echo imaging as opposed to RF spin echo methods.

Motion and T2*

Movements of protons within the magnet lead to a relatively rapid loss of phase coherence and a short T2*. As the TE times are increased in FLASH imaging, the signal from moving tissues decays rapidly, especially where there is no explicit flow compensation. As a result, although the CSF might be bright in such an approach it is not practical to use FLASH with a long TE for myelographic contrast behavior in the C-spine: the signal from CSF in-vivo decays rapidly as TE is increased, due to motion effects. These flow effects are even more important determinants of contrast in steady state methods (such as GRASS and SSFP), as we will see below.

Chemical Shift

Variations in resonance frequency are also present in proton NMR as a consequence of chemical effects. Specifically, as a consequence of the sharing of electrons in chemical bonds, protons in water molecules precess at a higher Larmor frequency than those in lipids; the fat/water chemical shift is about 3.5 ppm (or 220 Hz at 1.5 Tesla). Unlike the dephasing that occurs in magnetic field gradients, the chemical shifts result in separation of the NMR signal into distinct frequency bands. The signal therefore periodically dephases and rephases following the excitation. In a 1.5 Tesla magnetic field, the fat and water signal will rephase every 4.5 msec. With TE’s that are integer multiples of 4.5 msec, the fat and water signal in a voxel will add constructively. At intermediate TE’s, the signal from the two moieties will tend to be cancel one another. Note that if a voxel contains only fat or only water, there will be no signal cancellation, even though chemical shift artifacts may still be present. This behavior, of the MR signal from voxels containing a mixture of fat and water, is shown in figure 10.

Transverse Steady State

The discussion of shallow flip angle excitation presented above deals explicitly only with longitudinal excitation and relaxation. Following the data collection, however, some transverse magnetization usually remains. Because most of the popular MR imaging strategies require repeated excitations, it is important to understand the way in which residual transverse magnetization is incorporated into the MR signal by successive RF excitation pulses.

The FLASH steady state

Perhaps the easiest pulse sequence to understand fully is FLASH. In this technique, steps are taken to destroy any residual transverse magnetization prior to each excitatory RF pulse. In this way, only longitudinal magnetization is incorporated into the steady state. Figure 11 illustrates the behavior of the tissue magnetization cycle in FLASH imaging.

In FLASH, after the signal is collected (2) in the form of a gradient echo, a gradient pulse called a "spoiler" is used to destroy any remaining transverse magnetization (3) (see below). From there, the magnetization recovers longitudinally at the tissue T1 rate. The brevity of the TR in FLASH necessitates the use of a spoiler; without it, some transverse magnetization would remain at the beginning of the next RF pulse. With a short TR, the longitudinal magnetization does not recover fully during period 1; with a constant TR, however, the degree of recovery is the same with each cycle, and thus reaches a steady state. The extent of the signal-producing transverse magnetization at the time of the data collection depends on the amount of longitudinal recovery, as described above, and the amount of transverse signal decay that takes place during TE: the time between excitation (1) and data collection (2).

Spoiling

The simplest way to destroy transverse magnetization and the mechanism used for the original FLASH sequence is to apply a magnetic field gradient. Because spins in different locations in the magnet thereby experience a variety of magnetic field strengths, they will precess at differing frequencies; as a consequence they will quickly become dephased. Unfortunately, magnetic field gradients are not very efficient at spoiling the transverse steady state. To be effective, the spins must be forced to precess far enough to become phased randomly with respect to the RF excitation pulse. In clinical MR imagers, the field gradients are set up in such a way that they increase and decrease relative to the center of the magnet; the magnetic field at the magnet "isocenter" does not change. As a result, a common artifact in gradient-spoiled FLASH imaging is a contrast non-uniformity in the central lines of the image near the magnet isocenter due to incomplete spoiling.

Given that effective spoiling requires randomization of the phase with respect to the RF excitation, an alternative technique, known as RF spoiling, was developed (Darrasse, Mao et al. 1988). In this method, the phase of the RF excitation and receiver channel are varied pseudo-randomly with each excitation cycle. The method is highly effective in yielding consistent FLASH contrast. In practice, however, RF spoiling requires a fully digital RF system that, as of this writing, is only now becoming widely available on clinical MR instruments (figure 12).

GRASS or FISP Imaging

An alternative to spoiling is to incorporate residual transverse magnetization directly into the longitudinal steady state. This is the behavior achieved in FISP (Fast Imaging with Steady-state Precession) or GRASS (Gradient Recalled Acquisition of the Steady-State) pulse sequences (GRASS and FISP are manufacturer’s designations for essentially equivalent methods).

GRASS is very similar to FLASH, except that the spoiler pulse is eliminated. As a result, any transverse magnetization still present at the time of the next RF pulse is incorporated into the steady state. Figure 13 illustrates the GRASS magnetization cycle.

GRASS uses an RF pulse that alternates in sign. During the period between RF pulses (TR), some T1 recovery and T2* dephasing take place. Note that because there is still some remaining transverse magnetization at the time of the RF pulse, an RF pulse of a degrees (Figure 13) flips the spins less than a degrees from the longitudinal axis. There are a number of immediate consequences in GRASS imaging:

1. If TR is made much longer than T2* there will be no transverse magnetization at the time of the next RF pulse: GRASS and FLASH image contrast will become identical. GRASS must therefore be used with short TR in order to achieve its characteristic contrast behavior. As a result, except in 3D imaging (discussed below), GRASS is essentially a single slice technique.
2. When GRASS is used with small flip angles , as in FLASH, very little longitudinal magnetization is lost and the image contrast becomes almost independent of T1. Using a very short TE eliminates T2* effects, so that the images become proton density weighted.
3. As the flip angle is increased, the contrast becomes increasingly dependent on T1 and T2*. In general, the greater the ratio between T2* and T1, the brighter the tissue will appear (see equation 4), as in SE imaging, where long T2 relaxation times and short T1's yield high signal intensities. Unlike SE imaging, the T1 and T2 effects cannot be separated.
4. The special contrast behavior of GRASS imaging depends upon the phase relationship of the RF pulses to the protons. Factors that tend to alter the spin phase in the body tissues, in particular motion and magnetic susceptibility, will destroy the steady state behavior. For this reason, the signal from even slowly moving tissues such as CSF may not be as strong as expected.

Figure 14 shows the contrast behavior of GRASS scanning as a function of TR and flip angle. Comparing this with figure 8, note that for small flip angles and long TR’s, the contrast behavior of FLASH and FISP (GRASS) are very similar. It is only in the domain of large flip angle and short TR that FISP (GRASS) exhibits vastly different contrast.

Areas of Clinical Advantage with FLASH

For FLASH, as shown in figures 7 and 8, when flip angles approach 90°, the contrast becomes similar to that of conventional spin echo. However, several key differences remain. T2* contrast is substituted for T2 contrast as the TE is increased. Less obviously, the presence of a 180° pulse in SE pulse sequences results in a loss of signal as the TR becomes small compared to the TE. Extremely short TR times are therefore not possible with the conventional SE technique. Having no 180° pulse, FLASH and GRASS do not suffer from this limitation. As a result they provide a mechanism for gaining extremely high T1 contrast by imaging with TR times as brief as 20 to 30 msec while retaining reasonable signal levels.

Turbo-FLASH - Magnetic Preparation MP-RAGE

Advances in magnet technology and computer control have recently enabled the minimum TR in gradient echo scans to be reduced to well under 5 milliseconds (Frahm, Merboldt et al. 1990). The acquisition of a complete 128 line MR image, using "conjugate synthesis" (a method for reducing the number of phase encoding lines that must be acquired explicitly - see below), can therefore be completed in about 300 msec. In order to achieve reasonable signal levels, RF flip angles of only 5° or so are used. However, images obtained in this way have very little intrinsic contrast: maintaining adequate signal requires that losses (and therefore contrast) due to tissue T1 be minimized. Gradient echo techniques of this kind, which form complete images in times short compared to T1, are often known as Turbo-FLASH methods.

Because the acquisition times in these ultra-fast FLASH images are comparable to, or less than, tissue T1’s, however, one plausible strategy for improving contrast is to precede the acquisition with one or more "preparation pulses" (leading to the acronym MP-RAGE for Magnetization-Prepared Rapid Acquisition with Gradient Echoes) so that the longitudinal steady state, prior to the FLASH acquisition, is altered. Typical preparatory sequences might be an inversion (180°) pulse to add T1 contrast (Figure 15), or a 90° - 180° - 90° series to add T2 contrast. When the acquisition time is comparable to, or longer than, the tissue T1, this can lead to some contrast anomalies as the signal changes during sampling. To compensate for this effect, some investigators have suggested partitioning the data collection into segments, each having a relatively short duration, separated by a recovery period and an additional preparation pulse. Such techniques are in their early phases, and a detailed comparison of temporal efficiency and achievable contrast is premature at this writing.

Other Steady-State Techniques

Though the amplitude of the spin echo, as demonstrated by Hahn, is largest when a 90° excitation pulse is followed by a 180° pulse (Hahn 1950), similar spin echoes occur following any pair of RF pulses. As in the 90°-180° method, the echo follows the second RF pulse at a time equal to the delay between the pulses. In a continued train of a°pulses, the echoes will occur simultaneously with the excitation pulses.

In the technique known as CE-FAST (Contrast Enhanced Fast Acquisition with Shallow Tip), the MR signal is sampled immediately prior to each RF pulse. Because the signal is formed by a true spin echo, its contrast is predominantly T2-, rather than T2*-based and is less sensitive to artifacts and signal losses related to field non-uniformity and susceptibility variation. While the CE-FAST method offers the advantage of good contrast, it does so at the cost of somewhat limited SNR (Figure 16).

K-Space Strategies

MR imaging time may be reduced dramatically by modifying the strategies used for spatial encoding of the MR signal. The simplest such techniques, varying the size of the acquisition matrix, have already been discussed above. Through the process of frequency and phase encoding, the MRI raw data set is the 2D Fourier transform (2DFT) of the MR image. By convention, the two-dimensional space describing the MR data prior to its conversion to an image is called k-space. To understand the more sophisticated approaches to data collection, it is useful to present a few principles concerning the 2D Fourier transform NMR method.

Conjugate Synthesis

In the 2DFT approach, increases in spatial resolution (reductions in voxel size) require sampling of a larger area of the raw data space which, in turn, requires larger or longer gradient pulses. There is an important exception, however. It can be shown that under certain conditions, including uniform magnetic and RF field, no motion, and extremely long T2, the MR raw data set has the property of conjugate symmetry: only one-half of the data must be collected in order to form a complete image. In practice, it is possible to form images from real-world data sets by collecting just over half of the data. This technique, known as conjugate synthesis, is equally applicable to conventional and gradient echo images. Since its introduction in 1985, it has led to an immediate, two-fold reduction in scan times and is presently an important component of the majority of MR examinations under the monicker "Half-Fourier," "Partial NEX," "Partial K-Space," or "Conjugate Synthesis." In interpreting images collected with this technique, it is useful to note that it may be somewhat more motion sensitive than conventional acquisition and that it will generally result in about a 30% reduction in SNR, consistent with its reduced scan time.

Echo-Planar Methods

The methods known as "echo planar imaging," or EPI (Mansfield 1977), collect sufficient data to form a complete MR image following a single excitation pulse (often called a single "shot"). Since no repetitions, and thus no TR, are required, this can result in radical, imaging time reductions of 20,000-fold as compared to the spin echo counterparts. In the original EPI implementation, a small "phase-encoding" gradient was applied continuously during data collection, while large, rapidly alternating pulses were applied to the readout gradient. Most current implementations, including "Instascan" and "M-BEST" (Modulus Blipped Echo-Planar Scan Technique), use a pulse sequence that differs somewhat from Mansfield’s original proposal. Rather than remaining continuously on, the phase encoding gradient is pulsed briefly following each sign reversal of the readout gradient. The gradient encoding method is equally compatible with spin echo methods (where spatial encoding is performed during the Hahn spin echo) and shallow flip angle methods, which, by analogy to FLASH scanning, have no 180° echo-forming pulse. The applications and signal behavior of each method will be presented below. Figure 17 shows the MR pulse sequence used in spin echo Instascan. In this "pulse sequence diagram," the activity in each of the hardware components used to create an MR image is shown. The line labeled, "RF," shows the transmission of a 90° pulse, followed by a 180° pulse. Each of the other three lines indicate the activity in one of the three gradient channels: Slice selection, Phase encoding and Frequency encoding.

The rapidly oscillating "readout" gradient results in the formation of a series of gradient echoes, each of which is used to sample a separate line of k-space, as determined by the short pulses of the "phase-encoding" gradient. Sampling all of k-space during a single spin echo, requires not only extremely rapid sampling (and wide receiver bandwidth, see appendix), but large amplitude magnetic field gradients that must be switched rapidly.

As described in the appendix, conventional gradient amplifiers are limited in both speed and strength. Gradients that reach their peak amplitudes in less than about 500 µsec, or gradients stronger than about 1 Gauss/cm, require unconventional technology. Rzedzian developed a unique method that allows the generation of the gradient waveforms required for these "single-shot" images (Rzedzian 1987b). The technology takes advantage of the inductance of the gradient coil, coupled with a parallel or series capacitance, to produce a resonant system in which the large amount of energy required for the high amplitude gradients is stored alternately as charge across the capacitance and as current within the gradient coil.

Instascan Contrast Behavior

When used in the spin echo mode, the contrast behavior of the Instascan pulse sequence is virtually identical to that of spin echo imaging (Equation 2), a decided advantage of the technique. Figure 18 shows a series of cardiac images, obtained at various time points throughout the cardiac cycle, and demonstrates the freedom from motion artifacts and wide contrast range available with the spin echo Instascan method. In fact, because the echo-planar methods do not require repeated excitation pulses, the T1-dependence of the contrast can be easily controlled, or even eliminated, resulting in contrast behavior equivalent to infinite TR (Cohen and Weisskoff 1991). Under these circumstances, the contrast variations in the images are dominated by T2 differences and are controlled by changes in TE (Figure 19). Preceding the basic data collection with either a 90°, or a 180° pulse adds T1 contrast to the images. By convention, the time between 90° pulses is designated TR and the time between an inversion pulse and the 90° excitation pulse is denoted TI. Examples of contrast variation as a function of TI are shown in Figure 20.

If the 180° pulse is omitted, and the signal is sampled during the free induction decay (as in FLASH) following an RF excitation pulse of less than 90°, the contrast behavior already described above for FLASH methods is achieved. In echo planar imaging, however, each excitation results in a complete image; it is thus possible to obtain a series of images in rapid succession (short TR) with minimal signal loss. This has been used effectively in cardiac imaging, for example, where a series of images covering the cardiac cycle can be obtained in a single heartbeat (Figure 21).

K-Space Tiling

Spatial resolution in echo planar methods is limited ultimately by maximum achievable gradient amplitudes, high speed sampling requirements and the inevitable loss in SNR resulting from reduction in sampling time and resultant increases in bandwidth. With present technology, it is practical to obtain single-shot images with pixel sizes of 1.5 x 1.5 x 7 mm (figure 19 or 20). Further improvements in spatial resolution in general require a hybrid technique, such as Mosaic or MESH (Rzedzian 1987a), in which a different portion (typically one-fourth to one-half) of the raw data is acquired during each of one to four excitations. With these methods, usable in-plane resolution of 0.8 mm has been demonstrated. The multiple-shot methods effectively increase the total sampling time and therefore confer an advantage in SNR. Analogous techniques have direct applications in very short TR ("Turbo" or "snapshot") FLASH imaging: by separating the acquisition into two or more segments, the contrast aberrations that would otherwise occur during acquisitions with duration comparable to tissue T1 are mitigated somewhat. In fact, a broad range of k-space tiling schemes is possible. These span the range between conventional acquisition, in which a single line of raw data is collected following each excitation, to the echo-planar method, in which all of the data collection follows a single excitation (shot). The optimal combination would be a balanced tradeoff of tissue T2, receiver bandwidth, achievable gradient strengths, and requisite SNR and contrast.

RARE

Repeated 180° pulses, following an excitation pulse, will form multiple spin echoes. In the RARE technique developed by Hennig (Hennig, Nauerth et al. 1986), each of a series of two or more echoes is separately phase-encoded to form a single line of MR raw data. The total scan time for an image is reduced in direct proportion to the number of separate echoes. For example, if sixteen echoes are used, the scan time is reduced by a factor of sixteen, when compared to a conventional acquisition. Each of the later echoes is progressively more T2-weighted, but it is known that the contrast in the final MR image is dominated by the signal collected in a particular part of the raw data space (near the origin of k-space (Twieg 1983)). Using suitable strategies of ordering the raw data collection, it is possible by this method to obtain a variety of image contrasts from proton density to heavy T2-weighting. Good image quality and contrast have been achieved in as little as six seconds using modified RARE methods (Figure 22). As in the shallow flip angle methods, a distinct advantage of RARE over echo-planar imaging is that it can be applied without substantial modifications to the standard MR acquisition hardware. The RARE method, and its analogs (such as FAISE, or so-called Fast Spin Echo) necessarily trade volume coverage (in the form of multiple slices) against acquisition time per slice. Where long TR’s are desired, as in T2-weighted imaging, and where very large numbers of slices are not required, the RARE method is likely to prove extremely useful in clinical practice.

Summary & Perspective

The more than ten years of clinical experience with MRI have witnessed a steady decrease in imaging times: from hours, to the minutes required for today’s typical clinical applications. State-of-the-art technology, including gradient echo imaging and echo-planar methods will see scan times reduced still further to fluoroscopic, real-time, speeds. With these improvements, not only will the time required for a patient exam be reduced, but increases in temporal resolution will further enable new applications in functional imaging. Already, significant results have been shown in the dynamic assessment of clinically relevant parameters, from cardiac wall motion to changes in brain perfusion with mental activity (Belliveau, Kennedy et al. 1991). While the apparent practical limit to present MR acquisition rates is loss of SNR when acquisition times become short compared to T2, technological advances in RF systems and digital signal processing are likely to allow still further reductions while maintaining adequate SNR. Potential applications for such enormous speed extend beyond what can now be imagined.

Appendix -- Spatial Encoding, Bandwidth and T2

The position of the MR signal in space is generally determined by its frequency, as established by applied magnetic field gradients. A central theorem of signal processing states that in order to represent a signal accurately, it is necessary to sample it at least twice as fast as the maximum frequency of interest. For example, to sample frequencies from 0 to 10 kHz, one must sample 20,000 times/second (20 kHz). Furthermore, to detect n different, lower, frequencies, it is necessary to collect at least n samples. For example, to detect 256 different frequencies between 0 and 16 kHz, one must collect 256 samples at the rate of 32,000 samples/second - a total sampling time of 8 msec (=1/32,000 seconds/sample * 256 samples).

As described elsewhere (cf. Chapter 1.32 and 1.33), imaging gradients are used to cause the frequency of the MR signal to vary with position. Practically speaking, relatively exotic technology is needed to achieve gradient field strengths greater than about 1 gauss/cm (G/cm). For protons, this is equivalent to a difference in frequency of about 4 kHz/cm. The requirements that these factors impose on imaging time can be seen from a practical example. In order to obtain a pixel size of, say, 0.3 mm, it is necessary to detect frequency differences of 120 Hz (=4 kHz/cm * 0.1 cm/mm * 0.3 mm) that, in turn, necessitates a sampling duration of 8 msec (&Mac197;1/120 Hz). Therefore using practical imaging gradients of 1 G/cm, each line of MR raw data will require about 8 msec to collect. In a typical clinical scan we desire 256 lines, so the total sampling time is 256 * 8 msec, or about 2 seconds. The MR signal, however, is detectable for only a short period, limited by T2, of about 1/10th of a second. It is for this reason that the conventional image is built up line by line, rather than as a single data collection.

PULSE SEQUENCES

Name (as described here)	Also known as:	Contrast Summary	Typical TR & TE
FLASH	SPGR	T1: Increases with Flip Angle T2*: Increases with TE	TR: 30-500 ms TE < 15 ms
GRASS	FISP, FFE, PFI	T1/T2*. T2* component decreases with long TR and short TE. T1 component controlled by flip angle (like FLASH)	TR < 50 ms TE < 15 ms
CE-FAST	PSIF, SSFP	T1: controlled by TR and flip (as in FLASH) T2: increases with TE TR: 20-50 msec	TE: &Mac197; 2 times TR

ACQUISITION SCHEMES

Name (as described here)	Also known as:	Contrast Summary	Typical TR & TE
Rectangular FOV	Strip Scan	No change in contrast. SNR is reduced as compared to a square field of view image with similar resolution	No restrictions
Conjugate Synthesis	Half-NEX	Half-Fourier Scans with less than 100% raw data collection may have increased motion artifacts and flow voids. SNR is reduced.	If used to reduce scan time, minimum TE may be increased
Instascan	Echo Planar Imaging (EPI)	Spin Echo or Gradient available. SNR and/or resolution lower than conventional scans. Increased susceptibility and chemical shift artifacts.	TR: 50 msec to &Mac176; TE: &Mac179; 10 msec
Turbo-FLASH	Snapshot, Turbo-FFE, Turbo-GRASS	MP-RAGE T1: Adjusted by TI (inversion time) T2: Presently unavailable	TR: 3-5 msec TE: 2 msec Flip angle: 5

Abbreviations and Definitions

SI	Signal Intensity
TR	Repetition Time
TE	Echo Time (time between excitation and sampling)
T1	Longitudinal Relaxation Rate
T2	Transverse Relaxation Rate
FLASH	Fast Low Angle SHot
GRASS	Gradient Recalled Acquisition of the Steady State
FISP	Fast Imaging with Steady-State Precession
RF	Radio Frequency
SNR	Signal to Noise Ratio
NEX	Number of Selective Excitations
MR	Magnetic Resonance
SE	Spin Echo
MP-RAGE	Magnetization-Prepared Rapid Acquisition with Gradient Echoes
CE-FAST	Contrast Enhanced Fast Acquisition with Shallow Tip

---

REFERENCES

• Belliveau, J. W., D. N. Kennedy, R. C. McKinstry, B. R. Buchbinder, R. M. Weisskoff, J. M. Vevea, K. Nadeau, C. M.S., T. J. Brady and B. R. Rosen. (1991). Functional mapping of the human visual cortex by susceptibility-contrast NMR. Ninth Annual Meeting of the Society for Magnetic Resonance Imaging.
• Cohen, M. and R. Weisskoff. (1991). "Ultra-Fast Imaging." Magnetic Resonance Imaging. 9(1): 1-37.
• Darrasse, l., L. Mao and H. Saint-Jalmes. (1988). Spoiling techniques in very fast TR imaging. Society for Magnetic Resonance in Medicine seventh annual meeting. 611.
• Frahm, J., A. Haase and D. Matthaei. (1986). "Rapid NMR imaging of dynamic processes using the FLASH technique." Magn Res Med. 3: 321-327.
• Frahm, J., K. Merboldt, H. Bruhn, M. Gyngell, W. Hänicke and D. Chien. (1990). "0.3-second FLASH MRI of the human heart." Magnetic Resonance in Medicine. 13(1): 150-157.
• Hahn, E. (1950). "Spin echoes." Physical Review. 80(4): 580-594.
• Hennig, J., A. Nauerth and H. Friedburg. (1986). "RARE imaging: a fast method for clinical MR." Magnetic Resonance in Medicine. 3: 823-833.
• Mansfield, P. (1977). "Multi-planar image formation using NMR spin echoes." Journal of Physics. C10: L55-L58.
• Rzedzian, R. (1987a). High speed, high resolution, spin echo imaging by Mosaic scan and MESH. Society for Magnetic Resonance in Medicine. 51.
• Rzedzian, R. (1987b). A method for instant whole-body MR imaging at 2.0 Tesla and system design considerations in its implementation. Society for Magnetic Resonance in Medicine. 229.
• Twieg, D. (1983). "The k-trajectory formulation of the NMR imaging process with applications in analysis and synthesis of imaging methods." Medical Physics. 10: 610-621.

®Mark Cohen 2^5^99