You are on page 1of 11

Magnetic Resonance in Medicine 43:410 – 420 (2000)

Design of Improved Spectral-Spatial Pulses for Routine


Clinical Use
Yuval Zur*

Spectral-spatial pulses (spsp pulses) selectively excite spins subpulse to the other along the train is used to select the
at spatial location z and spectral frequency (due to chemical spectral frequency ␷.
shift and/or field inhomogeneity) ␷. In this work we discuss In this work, we present a method to generate optimal
the design of improved spsp pulses for fat signal suppres- and robust spsp pulses which we use in conjunction with
sion. Optimal pulses are designed as optimal constant rip- all standard imaging sequences. These pulses replace con-
ple FIR filters using the inverse SLR transform. Spsp pulses ventional FATSAT pulses since they provide improved
with thin slices are obtained by modifying the phases be- quality fat-free images. We distinguish between an spsp
tween subpulses, thereby eliminating unwanted magnetiza-
selective saturation pulse, which selectively saturates the
tion lobes. Robust spsp pulses at off-center slices are ob-
fat resonance, and an spsp selective excitation pulse,
tained with a prescan calibration. These pulses are used
which selectively excites the water resonance. The advan-
either for selective fat saturation or for selective water exci-
tages and disadvantages of each type are discussed. We
tation. It is shown that spsp pulses suppress fat signal better
than conventional fat saturation pulses. Using the techniques
show that spsp pulses are superior over FATSAT pulses
presented in this article, we replaced all the fat saturation because they are less sensitive to magnetic field variation
pulses on our systems with spsp pulses and obtained a due to inhomogeneity or magnetic susceptibility.
significant improvement in image quality. Magn Reson Med
43:410 – 420, 2000. © 2000 Wiley-Liss, Inc. PROPERTIES OF SPECTRAL-SPATIAL PULSES
Key words: spectral spatial pulses; fat signal suppression;
spectral selective pulses; radio frequency pulse design A train of N slice selective RF subpulses with tip angles
{␣i} (where i ⫽ 1 to N) applied in conjunction with an
Fat signal suppression is widely used in MRI to improve oscillating slice select gradient are shown in Fig. 1a,c. kz,
lesion conspicuity, reduce motion artifacts and chemical the k-space trajectory along the slice select direction z,
shift artifacts, and minimize shifting and blurring in spiral oscillates around kz ⫽ 0, thereby sampling k␷, the spectral
scans (1) and EPI (2). Conventional fat suppression tech- k-space axis. Each RF subpulse in the train is an inherently
niques use a spectrally selective RF pulse (FATSAT pulse) refocused pulse (11) with zero gradient area between sub-
to saturate the fat signal (3). This pulse is applied nonse- pulses. As shown in (11), the train of inherently refocused
pulses in Fig. 1a,c is equivalent to Fig. 1b,d, respectively,
lectively to the whole sample. Careful shimming of the
where a delta pulse replaces each selective RF subpulse.
magnet is required to avoid unwanted water signal sup-
The ith tip angle ␣i(z) at position z is proportional to the
pression. However, even with optimal shimming magnetic
Fourier transform of B1 of the ith selective subpulse as
susceptibility variations in the patient may impair the
given by Eq. [31] in Reference 11. An spsp pulse with RF
quality of fat suppression. Spectral spatial RF pulses (spsp subpulses applied only during the positive gradient lobes
pulses) are used increasingly to suppress fat signal (4 – 8). as in Fig. 1a,b is referred to as type I spectral-spatial pulse.
These pulses provide improved fat suppression over FAT- An spsp pulse with RF subpulses applied during positive
SAT pulses (4 – 6) because they are less sensitive to mag- and negative gradient lobes, as in Fig. 1c,d, is referred to as
netic field inhomogeneity. type II spectral-spatial pulse. The subpulses during the
An spsp pulse was originally described by Meyer et al. negative gradient lobes are inverted in time relative to the
(9) using the k-space formalism of Pauly et al. (10). This subpulses during the positive lobes because the kz trajec-
pulse selectively excites spins at spatial location z and tory during the negative lobes is time-inverted.
spectral frequency (due to chemical shift and/or field in- The k-space trajectory in the kz ⫺ k␷ plane is only ap-
homogeneity) ␷. For example, it can selectively excite wa- proximately inherently refocused because of the phase
ter protons or saturate fat protons at any location z and accumulation along the k␷ axis during and between the RF
generate a fat-free image at that location. In contrast, a subpulses (11). Therefore, the approximation shown in
conventional FATSAT pulse is applied nonselectively to Fig. 1b,d is true if
the whole sample. The spsp pulse consists of a train of
selective RF subpulses applied in conjunction with an ␥G⌬z Ⰷ ⌬v [1]
oscillating slice select gradient. Each subpulse in the train
selects a slice at z, while the phase accumulation from one where ⌬z is the excited slice thickness, G is the slice select
gradient amplitude, ␥G⌬z is the frequency bandwidth
along z and ⌬␷ is the frequency bandwidth along ␷.
GE Medical Systems Israel, Tirat Hacarmel, Israel.
*Correspondence to: Yuval Zur, GEMS Israel, Keren Hayesod St., P.O.B. Pulse Design Considerations
2071, Tirat Hacarmel, 39120, Israel. E-mail: yuval.zur@med.ge.com
Received 19 January 1999; revised 30 August 1999; accepted 25 October If Eq. [1] holds, the spsp pulse is equivalent to a train of N
1999. delta subpulses with tip angles {␣i (z)}, and ␶ seconds
© 2000 Wiley-Liss, Inc. 410
Design of Improved Spectral-Spatial Pulses 411

The limit set by Eq. [2] on ␶ restricts the duration of each


RF subpulse. On the other hand, the duration and band-
width of the subpulse must meet a specified slice profile
and minimum slice thickness. The minimum slice thick-
ness ⌬z and the time bandwidth product TB which deter-
mine the slice profile are given by:

␥S ⫽
TB
⌬z
where S⫽ 冕0
TRF
G共u兲du [3]

TRF in Eq. [3] is the duration of the RF subpulse and S is


the area of the gradient G during this subpulse. In order to
save time and minimize ␶ we apply the RF subpulses
during the gradient ramps using the VERSE algorithm (16).
For type I spsp pulses the RF subpulse duration is equal
to ␶/2. For type II spsp pulses the subpulse duration is ␶
(see Fig. 1). The ratio R between the area S for type II and
type I pulses depends on the maximum gradient ampli-
tude and slew rate and ␶. If the slew rate is infinite, R 3 2.
In practice 2.5 ⬍ R ⬍ 4. For the examples presented in this
FIG. 1. a: A spectral spatial pulse with N ⫽ 3 subpulses that are article (see below) R ⫽ 3.1. According to Eq. [3], the min-
applied during the positive lobes of an oscillating gradient with imum slice thickness ⌬z for type II pulses is R times
amplitude G. The time between adjacent subpulses is ␶ seconds. b: smaller than for type I pulses.
Each subpulse in a is equivalent to a delta pulse with the same tip
In this work we consider spsp excitation pulses that
angle. c: A spectral spatial pulse with N ⫽ 5 pulses that are applied
excite a desired spectral component, and spsp saturation
during positive and negative gradient lobes. The time between
subpulses is half the time shown in a. d: Each subpulse in c is pulses that saturate an undesired spectral component. For
equivalent to a delta pulse with the same tip angle. the former pulses, we are interested in the transverse mag-
netization and for the latter pulses in the longitudinal
magnetization. For spsp saturation pulses, where the
between adjacent subpulses. We can calculate the N tip phase of Mxy is of no concern, we use a minimum phase
angles {␣i} to selectively excite along the spectral fre- FIR filter which enables us to reduce the pass-band to
quency ␷ with a Finite Impulse Response (FIR) filter of N stop-band transition width ⌬␷.
points and the inverse SLR transform (7,12,13). In order to saturate fat effectively, it is essential that the
The choice of the FIR filter depends on the application. saturation pulse uniformly excite the slice of the sequence
The constant ripple FIR filter is considered optimal be- that precedes the saturation. To this end, the RF subpulses
cause it has the narrowest pass-band to stop-band transi- in a saturation spsp pulse are designed from a constant
tion width ⌬␷ for a given pulse duration T ⫽ N␶ and ripple minimum phase FIR filter which has the sharpest
ripples (14). This filter is optimal for imaging applications possible slice transition. In addition, the nominal slice
because narrow transition provides the best immunity thickness of the subpulses in a saturation pulse is slightly
against field variations due to inhomogeneity and/or sus- wider than the slice thickness of the preceding sequence.
ceptibility. Another popular FIR filter uses the binomial
coefficients (15). A very low (⬍0.001) stop-band ripple is Excitation of Off-Center Slices
achieved at the expense of a wider ⌬␷. Its main use is in To select a slice at position z ⫽ zi we use the frequency
spectroscopy where very high water suppression must be synthesizer to modify our rotating frame frequency at any
achieved. In our opinion the binomial FIR filter is less time t according to the applied gradient G(t). The synthe-
optimal for imaging because of its wider ⌬␷. Therefore, sizer frequency f(t) is given by:
we will use the constant ripple FIR filter throughout this
article. f共t兲 ⫽ ␥G共t兲z i ⫹ ␯ 0 [4]
The spectrum of the FIR filter is periodic, with a period ␷0 is referred to as the synthesizer center frequency.
of 1/␶ Hz (14). The magnetization is also periodic and We can excite a spin with Larmor frequency ␷L by setting
consists of many lobes separated by 1/␶ Hz. We always try ␷0 ⫽ ␷L or by setting ␷0 ⫽ ␷L and add a phase of ␾ ⫽ 2␲␶(␷0
to decrease ␶ as much as possible in order to prevent ⫺ ␷L) radians between any two adjacent RF subpulses. For
excitation by a neighboring spectral lobe. If we want to example, we can set ␷0 to the Larmor frequency of fat and
excite only water protons without touching the fat, we excite water by adding a phase of ␾ ⫽ ⫺2␲␶ ⌬␷wf between
must ensure that: RF subpulses, where ⌬␷wf is the water-fat frequency offset.

1/␶ ⱖ 2⌬␯ wf [2] ANALYSIS OF THE MAGNETIZATION EXCITED BY A


SPECTRAL-SPATIAL PULSE
where ⌬␷wf is the water-fat frequency offset. If Eq. [2] is In this section we present an analysis of the magnetization
fulfilled, the fat resonance is far from the next spectral lobe excited by an spsp pulse. The results of this analysis are
by at least ⌬␷wf. used in a later section to improve the spsp pulse.
412 Zur

the precession frequency fn during even subpulses where


the gradient is negative:

f p共z, ␯兲 ⫽ ␥Gz ⫹ ␯ [6a]

f n共z, ␯兲 ⫽ ⫺␥Gz ⫹ ␯ [6b]

When fp ⫽ ⫺fn the odd tip angles are different from the
even tip angles because the former are the Fourier trans-
form of the subpulse at fp and the latter at ⫺fn. fn comes
with a negative sign because the kz trajectory during neg-
ative gradient lobes is time-inverted. The magnetization
Mxy(z, ␷) resulting from excitation by a type II spsp pulse
is analyzed in Appendix A. An analytical expression to
Mxy (z, ␷) was derived in reference (9). According to the
analysis in Appendix A, Mxy(z, ␷) is a sum of magnetiza-
tion lobes along the ␷ axis as shown in Fig. A2 (see Ap-
pendix A). We denote the even lobes M2n and the odd
FIG. 2. 兩Mxy(z, ␷)兩 for a type I spsp water-selective excitation pulse. lobes M2n⫹1 with n ⫽ 0, ⫾1, ⫾2…. The frequency offset
Pulse parameters are given in the text. Setting the synthesizer between even and odd lobes is 1/(2␶), where ␶ is the time
frequency on the water resonance excites the water protons. Note between subpulses. The central lobe, with n ⫽ 0, is M0. By
that the minimum slice thickness is about 9 mm. definition, M0 is centered at the rotating frame frequency
where the phase ␾ between subpulses is zero.
We show in Appendix A that each even lobe M2n is
Type I Pulse
symmetric about ␷ ⫽ 0 for any given z (Eq. [7a]) and
The precession frequency fp of a spin at (z, ␷) during the symmetric about z ⫽ 0 for any given ␷ (Eq. [7b]).
transmission of the RF subpulses is given by:
M 2n共z, ␯兲 ⫽ M ⫺2n共z, ⫺␯兲 [7a]
f p共z, ␯兲 ⫽ ␥Gz ⫹ ␯ [5]
M 2n共z, ␯兲 ⫽ M 2n共⫺z, ␯兲 [7b]
fp is the same for all subpulses because pulses are applied
only during the positive gradient lobes. If Eq. [1] holds, Each odd lobe M2n⫹1 is antisymmetric with respect to ␷ ⫽
then Mxy (z, ␷) as a function of ␷ for a constant z is 0 for any given z (Eq. [8a]) and antisymmetric with respect
proportional to the FIR filter spectrum, and Mxy (z, ␷) as a to z ⫽ 0 for any given ␷ (Eq. [8b]).
function of z for a constant ␷ is proportional to the slice
profile of each RF subpulse. M 2n⫹1共z, ␯兲 ⫽ ⫺M ⫺共2n⫹1兲共z, ⫺␯兲 [8a]
Example 1: We demonstrate a type I spsp pulse by
designing a 90° excitation pulse for our 1.9T magnet. The
M 2n⫹1共z, ␯兲 ⫽ ⫺M 2n⫹1共⫺z, ␯兲 [8b]
water is 280 Hz offset from fat. According to Eq. [2], the
maximum time between subpulses is ␶ ⫽ 1.8 msec and the
duration of each subpulse is ␶/2 ⫽ 0.9 msec. The maxi- The ␷ ⫽ 0 frequency in Eqs. [6], [7], and [8] is the synthe-
mum gradient amplitude and slew rate in our system is 19 sizer center frequency ␷0 from Eq. [4]. The antisymmetry in
mT/m and 40 mT/m/msec, respectively. Each subpulse is Eq. [8] implies that the odd lobes must be zero for all z at
sinc-shaped with time-bandwidth product of 3.0. From Eq. ␷ ⫽ 0 and must be zero for all ␷ at z ⫽ 0.
[3] the minimum slice thickness ⌬z is 9 mm. We use the Equations [7] and [8] hold only for the transverse mag-
following parameters: N ⫽ 8 subpulses, slice thickness netization Mxy. The longitudinal magnetization Mz ⫽
⌬z ⫽ 9 mm, effective pass band/stop band ripples of 0.005 冑1 ⫺ 兩Mxy兩2 consists of sidelobes located at the same spec-
and 0.015, respectively, and FIR filter bandwidth of 150 tral locations. However, all the lobes are symmetric with
Hz. The tip angles of the subpulses are 2.2°, 7.4°, 14.1°, respect to z and ␷ because Mz is independent of the sign
19.5°, 19.9°, 15.3°, 8.4°, and 3.1°. The magnetization 兩Mxy of Mxy.
(z, ␷)兩 in Fig. 2 was calculated by solving Bloch equations Example 2: A type II pulse is demonstrated by designing
with this spsp pulse. The frequency separation between a 90° excitation pulse with similar parameters as in Exam-
lobes on the ␷ axis is 1/␶ ⫽ 560 Hz and the shape of each ple 1. These parameters are N ⫽ 8 subpulses with time-
lobe along z is the slice profile of each subpulse. We excite bandwidth product of 3, ␶ ⫽ 1.8 msec, linear phase FIR
the water resonance by setting the synthesizer center fre- with effective pass band / stop band ripples of 0.005/0.015,
quency ␷0 (Eq. [4]) on the Larmor frequency of water. respectively, and FIR filter bandwidth of 150 Hz. The
duration of each subpulse is now 1.8 msec. The tip angles
are the same as in Example 1. Using the maximum avail-
Type II Pulse
able gradient amplitude and slew rate (19 mT/m and 40
For type II pulses the precession frequency fp during odd mT/m/msec, respectively), the minimum slice thickness is
subpulses, where the gradient is positive, is different from 2.9 mm. In order to excite the water protons, the synthe-
Design of Improved Spectral-Spatial Pulses 413

FIG. 3. 兩Mxy(z, ␷)兩 for a type II spsp water-selective excitation pulse.


Setting the synthesizer frequency on the water resonance excites
the water protons. Pulse parameters are given in the text. The
unwanted odd lobe at ⫺280 Hz excites the fat protons.

sizer center frequency ␷0 is set on the water resonance.


兩Mxy(z, ␷)兩 for this pulse is shown in Fig. 3. Setting ␷0 on the
water Larmor frequency excites the water resonance. The
even lobes M⫺2, M0 and M2 are shown at ␷ ⫽ ⫺560, 0 and
560 Hz, respectively, and the odd lobes M⫺1 and M1 are
shown at ␷ ⫽ ⫺280 and 280 Hz, respectively. This pulse
may be useful for selective water excitation provided the
fat signal located at –280 Hz is uniform throughout the
slice width. In this case, the signal will cancel out because
M⫺1 is antisymmetric in z. On the other hand, this pulse is
useless for selective saturation because Mz is always sym-
metric in z and ␷. Since spsp saturation pulses are very
useful clinically, it is necessary to eliminate the odd lobes.

FIG. 4. Elimination of odd lobes for type II water-selective spsp


ELIMINATION OF ODD LOBES FOR excitation pulses. a: The synthesizer center frequency ␷0 is set on
1/(2␶) ≥ ␷ ≥ ⴚ1/(2␶) the water resonance. The odd lobe M⫺1 at ⫺1/(2␶) Hz excites the fat
If Eq. [2] is fulfilled, it is sufficient to eliminate the odd resonance. b: ␷0 is set on the fat resonance. A phase ␾ ⫽ ␲ radians
lobe in the range 1/(2␶) ⱖ ␷ ⱖ ⫺1/(2␶) in order to prevent is applied between subpulses to shift M0 and M2 to ⫺1/(2␶) and
1/(2␶) Hz, respectively. M1 at ␷ ⫽ 0 is eliminated and the fat is not
unwanted excitation as in Example 2. The elimination of
excited. c: The fat-water frequency offset is smaller than 1/(2␶). ␷0 is
odd lobes for 1/(2␶) ⱖ 兩␷兩 is demonstrated in Fig. 4. set on the fat resonance. M0 is shifted by ⫺␷1 Hz from the fat so that
Suppose we want to excite water protons and the fat only the water is excited by M2. The frequency offset between M0
resonance is 1/(2␶) Hz below the water resonance. If ␷0 is and M2 is 1/␶ Hz.
set on the water resonance as in Example 2, the fat reso-
nance is excited by M⫺1 as shown in Fig. 4a. Instead, ␷0 is
set on the fat resonance. By definition, this is the ␷ ⫽ 0
frequency. A phase shift of ␾ ⫽ ␲ radians is added between to excite the water (fat) resonance. ␷0 is set on the fat
adjacent subpulses, which shifts the frequency of the even (water) resonance frequency. The even lobe M0 (M2) is
lobes M0 and M2 to ␷ ⫽ ⫺1/(2␶) and ␷ ⫽ 1/(2␶) Hz, respec- shifted by ⫺␷1 Hz from ␷0 by adding a phase of ␾ ⫽ 2␲␷1␶
tively, as shown in Fig. 4b. M1 is eliminated because it is radians between subpulses. ␷1 is determined in such a way
located at ␷ ⫽ 0 and the water resonance is selectively that only the water (fat) resonance is excited by M2 (M0).
excited by M2. In case we want to excite fat protons, ␷0 is In the next examples we demonstrate the elimination of
set on the water resonance and a phase of ␾ ⫽ ␲ radians is the odd lobe M1 and design a water-selective excitation
added between adjacent subpulses. The fat is selectively pulse and a fat-selective saturation pulse. These pulses are
excited by M0. designed for 1.9T where the water-fat frequency offset is
The general case, where 1/(2␶) is larger than the water- 280 Hz. The maximum gradient amplitude and slew rate
fat offset frequency is shown in Fig. 4c. Suppose we want are 19 mT/m and 40 mT/m/msec, respectively.
414 Zur

Example 3: We design a 90° water-selective excitation


pulse with a linear phase FIR. The spsp pulse parameters
are the same as in Example 2. The effective pass band/stop
band ripples are 0.005/0.015, respectively, and the band-
width is 150 Hz, N ⫽ 8 subpulses, and ␶ ⫽ 1.8 msec. Each
subpulse is a sinc pulse with a time bandwidth product of
3.0. The minimum slice thickness is 2.9 mm. To eliminate
the odd lobes we set ␷0 on the fat resonance and add a
phase of ␲ radians between subpulses in order to excite the
water. The RF and gradient waveforms are shown in Fig.
5a and the magnetization 兩Mxy兩 in Fig. 5b. The amplitudes
of the even subpulses in Fig. 5a are inverted because of the
␲ radians phase shift between subpulses. The water is
excited by M2. The fat resonance is not excited because M1
at ␷ ⫽ 0 is eliminated.
Example 4: In this example we design a fat selective
saturation spsp type II pulse. We use a minimum phase
FIR filter. The pass band/stop band effective ripples are
0.06/0.003 and the bandwidth is 170 Hz. There are N ⫽ 8
subpulses and ␶ ⫽ 1.8 msec. The tip angle is ␣ ⫽ 110°. For
each subpulse we use a minimum phase FIR filter with in
slice/out of slice ripples of 0.05/0.005, respectively, to
ensure sharp slice profile. The time bandwidth product is
3.6. The subpulses tip angles are ⫺6.4°, ⫺10.6°, ⫺1.7°,
13.1°, 33.0°, 38.6°, 29.9°, and 14.1°. To eliminate M1 we set
␷0 on the water resonance and add a phase of ␲ radians
between subpulses. The RF and gradient waveforms are
shown in Fig. 6a. The minimum phase RF waveform of
each subpulse is interpolated according to the gradient
waveform as required by the VERSE algorithm. The even
subpulses are inverted in time relative to the odd sub-
pulses, and the amplitudes of the even subpulses are in-
verted as in Example 3. Mz(z, ␷) is shown in Fig. 6b. The
negative of Mz is displayed to improve the visibility of the
graph. The fat resonance at ␷ ⫽ ⫺280 Hz is excited by M0.
The water resonance at ␷ ⫽ 0 is not excited because M1 is
eliminated. For this pulse the elimination of odd lobes is
critical, because otherwise the water resonance would be
saturated by M1.
Other techniques were suggested to eliminate odd lobes.
With the first technique (4) the scan runs twice: once with
a gradient G and then with a gradient ⫺G. The raw data
from the two scans are added. The odd echoes are canceled
because of their antisymmetry with respect to z (Eq. [8b]).
With the second technique (6) an alternated line scanning
of k-space with two excitations is used to obtain simulta-
neous water and fat images. In practice, however, these FIG. 5. a: RF and gradient waveforms for a type II spsp water-
techniques are not very useful because: 1) saturation selective excitation pulse. Pulse parameters and tip angles are the
pulses cannot be used because Mz is the same for both same as in Fig. 3. The synthesizer frequency is set on the fat
scans and cannot be cancelled; 2) scan time increases resonance and a phase of ␲ radians is applied between subpulses
because at least two excitations are required; and 3) patient by inverting the amplitudes of even subpulses. b: 兩Mxy(z, ␷)兩 for this
movement during the scan may cause the raw data from pulse. The fat is not excited and the water at ␷ ⫽ 280 Hz is
selectively excited by M2.
the two excitations to be inconsistent.

SPECTRAL-SPATIAL PULSE CALIBRATION


accurately measure these phase shifts, allowing their full
In this section we show that unwanted phase shifts arise cancellation during the scan.
between subpulses of type II spsp pulses due to unavoid- For off-center slices the synthesizer frequency f(t) must
able small gradient distortions. For pulses designed for follow the slice select gradient G(t), as in Eq. [4]. f(t) is
selective water excitation, these phase shifts cause fat ex- controlled digitally and it accurately follows the desired
citation and vice versa. A prescan sequence is used to pattern. On the other hand, G(t) deviates from the desired
Design of Improved Spectral-Spatial Pulses 415

FIG. 7. Demonstration of the generation of an unwanted phase due


to a deviation of the gradient pattern G(t) from the synthesizer
pattern f(t). The thick line is f(t). f(t) is delayed by ⌬t seconds from
G(t). The phase shift ␺0 between the first and the second subpulses
is proportional to the dotted area. An opposite phase shift is gen-
erated between the second and third subpulses which is propor-
tional to the gray area.

␺共t兲 ⫽ ␥z i 冕
0
t
⌬G共t⬘兲dt⬘ ⫹ ␥ 冕
0
t
⌬B 0共t⬘兲dt⬘ [9]

where zi is the slice position. Equation [9] shows that for


off-center slices where zi is large, ␺(t) may be significant
even with small gradient deviation ⌬G(t).
This effect is demonstrated in Fig. 7, where there is a
small gradient delay ⌬t between f(t) and G(t). Because of
this delay a phase shift of ␺0 ⫽ 2␥ziG⌬t radians, which is
equal to the dotted area in Fig. 7, is generated during the
gradient fall between odd and an adjacent even subpulse.
G is the peak gradient amplitude. Similarly, a phase shift
of ⫺␺0 radians is generated during the gradient rise be-
tween even and an adjacent odd subpulse. This phase shift
is equal to the gray area in Fig. 7. Consequently, the phase
between even subpulses (or odd subpulses) is ␺0 ⫺ ␺0 ⫽ 0
radians. A similar effect occurs if the duration/time con-
stant of the distortions/eddy currents is short compared to
FIG. 6. a: RF and gradient waveforms for a type II spsp fat-selective the time ␶ between subpulses. In this case, the gradient fall
saturation pulse. A minimum phase FIR filter is used to calculate the
induces a phase shift of ␺0 radians and the gradient rise
tip angles {␣i} of the subpulses and a minimum phase FIR filter is
induces an opposite phase shift of ⫺␺0 radians.
also used for each subpulse to ensure optimal slice profile. The
pulse parameters are given in the text. The even subpulses are For eddy currents with time constant comparable to ␶,
inverted in time relative to the odd subpulses. A phase shift of ␲ the phase ␺0 varies from one subpulse to the other until a
radians is added between subpulses by inverting the amplitudes of steady state is achieved, where ␺0 between odd and even
even subpulses. b: Mz(z,␷) for this pulse. The odd lobe M1 at ␷ ⫽ 0 subpulses equals to ⫺␺0 between even and odd subpulses.
is suppressed. The fat resonance at ␷ ⫽ ⫺280 Hz is selectively In order to ensure that steady state is achieved prior to the
excited by M0 and the water resonance is not excited. first subpulse, the spsp pulse is preceded with dummy
gradient pulses. Eddy currents with time constants longer
than the total spsp pulse duration have negligible effect
pattern because of gradient delay, eddy currents, and dis- on ␺0.
tortions generated by the current amplifier (17). This de- The effect of the phase shift ␺0 is analyzed in Appendix
viation, ⌬G(t), as well as B0 eddy currents (4) ⌬B0(t) gen- B. When ␺0 ⫽ 0 the unwanted magnetization lobe M1 in
erate an undesired phase shift of ␺(t) radians between the Fig. 4 reappears at 1/(2␶) Hz away from the excitation lobe
spins and the rotating frame defined by the synthesizer: M0, causing M0 itself to decrease. The ratio between the
416 Zur

FIG. 9. 兩Mxy兩 vs. z and ␷. The image shows the lobes generated by
FIG. 8. A prescan sequence to measure the phase ␺0 between the spectral spatial pulse of Fig. 5. The vertical axis is z and the
adjacent subpulses of a type II spsp pulse. The synthesizer pattern horizontal axis is ␷. Applying a constant gradient during excitation
f(t) follows the gradient pattern G(t) in segment B according to simulated the spectral axis ␷. The lobe M1 at the center of the ␷ axis
Eq. [4]. (␷ ⫽ 0) is eliminated, as predicted by Fig. 5.

amplitudes of M1 and M0 as a function of ␺0 is given by Eq. gradients G(t) there are synthesizer patterns f(t), which
[A3]. From Eq. [A3], if ␺0 ⫽ ␲/2 radians 兩M1兩 ⫽ 兩M0兩 and if follow the gradient patterns in segment B according to
␺0 ⫽ ␲ radians all the magnetization is shifted from M0 to Eq. [4]. The gradient lobe C in Fig. 8 is identical to the
M1. This unwanted magnetization excites fat when we gradient lobe under each subpulse in the spsp pulse that
want to excite water and vice versa. Therefore ␺0 must be we use in the scan itself. The prescan sequence runs
minimized so that M1 becomes negligible. twice: 1) the gradients and synthesizer patterns G(t) and
According to the analysis in Appendix B, if we require f(t) in segment B run with positive polarity and the
that M1/M0 ⬍ 5% then ␺0 ⬍ 0.1 radian. Using the example phase of the echo signal is calculated; 2) the gradients
in Fig. 7, ␺0 ⫽ 2␥ziG⌬t radians where G is the peak gradi- and synthesizer patterns in segment B run with opposite
ent, ⌬t is the gradient delay and zi is the slice position. We polarity, i.e., ⫺G(t) and ⫺f(t) and the signal phase is
may minimize ⌬t by shifting the gradient demand forward calculated again. For each slice the signal phase of the
in time so that G(t) and f(t) almost coincide and ␺0 ⬍ 0.1 second run is added to the signal phase of the first run.
radians. This means that for G ⫽ 19 mT/m and zi ⫽ 10 cm, The result is the phase ␺0 between the odd and even
兩⌬t兩 ⬍ 0.1 ␮sec. In practice we cannot determine ⌬t with subpulses in the scan itself. This measured ␺0 is then
such accuracy. We face the same problem with gradient used to compensate the phase errors in the spsp pulse
distortions and eddy currents. In practice the gradient during the scan. A phase of ⫺␺0 radians is added to the
delay, distortions, and eddy currents cannot be accurately synthesizer between odd and even subpulses and a
controlled or measured and they vary significantly with phase of ␺0 radians is added to the synthesizer between
gradient axes, gradient amplifiers, and even gradient cur- even and odd subpulses. This process accurately elimi-
rent amplitude (17). To solve this problem we developed a nates the phase errors in the spsp pulse.
special prescan sequence. We run this prescan immedi- Segments A and B are both flow-compensated to ensure
ately before the scan starts and measure ␺0 using the slice accurate phase measurement in areas of motion and blood
positions and gradient demands that are actually used in flow. Another reason is that the two gradient lobes in
the scan. ␺0 is corrected by applying an opposite phase segment B prior to gradient lobe C ensure correct phase
shift, ⫺␺0, during the scan. Note that ␺0 does not affect measurement in case of eddy currents with longer time
type I spsp pulses because RF subpulses are applied only constant because they are equivalent to the dummy gradi-
during the positive gradient lobes and the phase between ent pulses that precede the spsp pulse.
subpulses remains zero. Therefore, no prescan or calibra-
tion is required for type I pulses.
EXPERIMENTAL RESULTS
The prescan sequence is shown in Fig. 8. This is a spin
echo sequence with only a slice-select gradient that runs The results shown in this section were implemented on an
with the same slice positions and slice thickness as the Elscint Prestige 1.9T whole body imaging system (Elscint
scan itself. The sequence has two segments: A and B. Ltd. Haifa, Israel).
Segment A is a flow-compensated spin echo sequence First we verify experimentally the elimination of odd
with standard slice selective 90° and 180° pulses. Seg- magnetization lobes as simulated in Fig. 5. Two gradients
ment B is also flow-compensated. In addition to the were applied during the spsp pulse excitation: the oscil-
Design of Improved Spectral-Spatial Pulses 417

FIG. 10. Demonstration of the accu-


racy of the calibration prescan. The im-
ages shown were acquired with a spin-
echo sequence where the 90° pulse
was the water-selective spsp pulse
from Fig. 5. Scan parameters: TR ⫽
500 msec, TE ⫽ 18 msec, FOV ⫽ 23
cm, acquisition matrix ⫽ 256 x 256. a:
Image acquired at z ⫽ 4 cm from the
magnet center. The phase shift ␺0,
measured by the prescan, was com-
pensated by applying phase shifts of
opposite sign between the subpulses.
The fat signal is fully suppressed. b:
Image acquired at z ⫽ 4 cm from the
center. The scan parameters are iden-
tical to the scan in a, but the compen-
sating phase shifts were set to zero. c:
Another slice from the scan in a. z ⫽ 5
cm from the center. ␺0 was compen-
sated and the fat signal is fully sup-
pressed. d: Another slice from the scan
in b. z ⫽ 5 cm from the center. No
phase compensation. Note that the fat
in d is brighter than in b because ␺0 at
z ⫽ 5 cm is larger than at z ⫽ 4 cm.

lating slice-select gradient G(t) along one direction and a DISCUSSION


constant gradient along an orthogonal direction. The con-
stant gradient simulated the spectral axis ␷. A 180° slice- Consistent and reliable fat signal suppression in a clinical
selective pulse was applied along the third direction. A environment is a difficult task due to magnetic field inho-
spin warp gradient echo sequence imaged the resulting mogeneity and susceptibility induced by tissue. Next, we
magnetization. The imaged object was a cylinder with 15 explain why fat signal suppression obtained with spsp
cm diameter and height filled with doped water. The re- pulses is more robust than fat suppression obtained with
sulting image is shown in Fig. 9. The horizontal axis is the conventional FATSAT pulses.
spectral axis ␷ and the vertical axis is the slice axis z. The For scans where many slices are acquired in each TR, an
odd magnetization lobe M1 at the center of the ␷ axis (the spsp saturation or excitation pulse excites each slice only
␷ ⫽ 0 frequency) is zero as predicted by the simulation. once per TR. On the other hand, a conventional fat saturation
Next we demonstrate the importance and accuracy of pulse excites each slice every TR/N where N is the number of
the calibration prescan sequence for type II spsp pulses. slices. Suppose the water resonance moves slightly towards
The head images in Fig. 10 were acquired with a spin-echo the frequency of the saturation pulse due to field variation. A
sequence where the 90° pulse was a water-selective spsp conventional fat saturation pulse will saturate it, whereas an
type II pulse similar to the pulse presented in Fig. 5. In spsp pulse will very slightly affect it. For example, suppose
order to ensure coherent water excitation, we ran a prescan we use a spin-echo sequence with a TR of 1 sec and 30 slices
sequence and compensate the unwanted phase shifts ␺0 as with fat saturation. An spsp pulse excites each slice only
explained before. This compensation is very critical at once per second, whereas a FATSAT pulse excites each slice
off-center slices. The images in Fig. 10a,b were acquired at every 1000/30 ⫽ 33 msec. Therefore, for a given field varia-
4 cm from the magnet center. The images in Fig. 10c,d tion the water resonance will be much more saturated by the
were acquired at 5 cm from the center. In Fig. 10a,c we conventional FATSAT pulse.
compensated the unwanted phase shifts ␺0 by adding Water-selective spsp pulses for low tip angle short TR
phases of opposite signs between subpulses. The fat signal gradient echo sequences are a better alternative for fat
in these images is perfectly eliminated. In Fig. 10b,d this suppression than conventional fat saturation pulses which
compensation was removed, and the fat signal suppression use large flip angles of 100°–120°. The fat saturation pulse
failed. Note that the fat signal is brighter in Fig. 10d be- may saturate the water signal in case of field variation,
cause this image is further from the center and ␺0 is larger whereas the low-tip-angle spsp pulse affects the water
than in Fig. 10b. signal much less.
418 Zur

An important application of spsp pulses is fat-free time- pulses with thin slices were obtained by modifying the
of-flight angiography where a type I water-selective spsp phases between subpulses, thereby eliminating unwanted
pulse is used (18). In this case only type I pulses can be magnetization lobes. The elimination of unwanted lobes is
used because they are inherently flow-compensated (19). critical for saturation spsp pulses. Robust spsp pulses at
The advantage of the spsp pulse over a fat saturation pulse off-center slices were obtained with a prescan calibration.
is demonstrated with an example. Suppose we acquire a We have shown that spsp pulses suppress fat signal better
3D time-of-flight angiogram of the carotid bifurcation. We than conventional fat saturation pulses. Therefore, we
use a receive-only neck coil and transmit with the body have replaced all the fat saturation pulses on our scanner
coil. If we use a conventional fat saturation pulse to elim- with spsp pulses and obtained a significant improvement
inate the fat signal we face two problems: 1) the large tip in image quality.
angle of the fat saturation pulse (⬎90°) can saturate the
blood signal even with small magnetic field variations, as
APPENDIX A
explained above; and 2) field variations anywhere from the
heart to the carotid bifurcation will saturate the blood In this appendix, we analyze Mxy (z, ␷) for a type II spsp
signal. If the water resonance anywhere in the heart is pulse.
shifted and saturated, blood signal will be saturated every- An spsp pulse with N subpulses is equivalent to a se-
where, including the carotid bifurcation. By using a low- quence of N delta pulses with tip angles {␣i} as shown in
tip-angle 3D spsp water-selective excitation pulse we over- Fig. 1. The time between subpulses is ␶. We write the set of
come these problems. calculated tip angles {␣i} as a sum of the odd subpulses and
Another advantage of spsp pulses is that we can inde- the even subpulses with tip angles {␣io} and {␣ie}, respec-
pendently vary the spectral selection frequency at each tively. This is demonstrated in Fig. A1 for N ⫽ 6 sub-
slice according to the water (or fat) frequency at that slice pulses.
(8,20). A prescan sequence measures the water frequency The precession frequency of a spin at position z and
at every slice and the synthesizer frequency of the spsp spectral frequency ␷ during odd and even subpulses, given
pulse during the scan is modified accordingly. This greatly by Eq. [6], is fp and fn, respectively. When fp ⫽ ⫺fn the tip
improves the quality of fat signal suppression over large angle of each odd subpulse is different from the tip angle
volumes. of an adjacent even subpulse. Therefore, the odd tip angles
The use of an spsp selective excitation pulse or spsp of the pulse become A{␣io} and the even tip angles B{␣ie}
selective saturation pulse depends on the application. where A and B are scalars. The N-pulse excitation function
Both have advantages and drawbacks. An excitation spsp with tip angles A{␣io} ⫹ B{␣ie} is defined as b(t), and the
pulse is used mostly in short TR low-tip-angle scans in odd and even excitation functions A{␣io} and B{␣ie} are
order to avoid the large tip angles of a saturation pulse, as called bo(t) and be(t), respectively. The magnetization gen-
explained above. Its drawback is its long duration, which erated by bo(t) and be(t) is Mo and Me respectively and is
means long TE and increased sensitivity to patient motion shown in Fig. A2. The frequency offset between magneti-
and field inhomogeneity. A saturation spsp pulse is used zation lobes in Mo and Me is 1/(2␶) and its amplitude is
mostly in conjunction with spin-echo and fast-spin-echo proportional to A and B, respectively. Since be(t) is shifted
scans. A saturation pulse does not affect the TE. Therefore, in time by ␶ seconds from bo(t), the polarity of the lobes in
pulses with long duration and sharp transition width can Me alternates. The magnetization M generated by b(t) is the
be used. Another advantage is that a saturation pulse can
be used in conjunction with any scan without affecting its
performance. It does not affect the TE, echo space, echo
train length, or any other parameter of the scan. The main
drawback of a saturation pulse is its sensitivity to RF field
(B1) inhomogeneity. When B1 changes, the tip angle
changes and the saturation becomes incomplete.
Another consideration is when to use type I spsp pulses
and when type II. Type I pulses do not require a prescan
calibration and are flow-compensated. On the other hand,
the minimum slice thickness is often too large, especially
at a high field strength where water-fat frequency offset is
large. Type II spsp pulses require a prescan calibration, but
the minimum slice thickness is adequate even at high field
strength.

CONCLUSIONS
In this work we discussed the design of improved spectral-
spatial pulses for fat signal suppression. Pulses for selec- FIG. A1. A type II spsp pulse with N ⫽ 6 subpulses and tip angles
tive fat saturation and selective water excitation were con- {␣} ⫽ ␣1 to ␣6 is divided into odd subpulses with tip angles {␣o} ⫽
sidered. Optimal pulses were designed as optimal constant ␣1, ␣3 and ␣5, and even subpulses with tip angles {␣e} ⫽ ␣2, ␣4
ripple FIR filters using the inverse SLR transform. Spsp and ␣6.
Design of Improved Spectral-Spatial Pulses 419

cient B of the even tip angles {␣ie} at a precession fre-


quency ⫺f:

A共 f 兲 ⫽ B共⫺f 兲 [A2]

From Eqs. [A1] and [A2] we conclude that the amplitudes


of the even lobes and the odd lobes of a spin at (z1, ⫺␷1) are
B ⫹ A and B – A, respectively. That is, for any given z ⫽
z1 the amplitudes of the even lobes M2n are symmetric
with respect to ␷ ⫽ 0 and the amplitudes of the odd lobes
M2n⫹1 are antisymmetric. In a similar way we can show
that for any given ␷ ⫽ ␷1 the amplitudes of M2n are sym-
metric with respect to z and the amplitudes of M2n⫹1 are
antisymmetric. These conclusions are summarized by Eqs.
[7] and [8].
Note that ␷ ⫽ 0 is defined as the center synthesizer
frequency during transmission ␷0, because the amplitudes
A and B are determined by the difference between the
Larmor frequency and the synthesizer frequency during
transmission.

APPENDIX B
In this appendix, we analyze the effect of the phase ␺0
which alternates between odd and even subpulses. The
train of the N subpulses with tip angles {␣i} is written as a
sum of the odd subpulses with tip angles {␣io} and the even
subpulses with tip angles {␣ie}, as shown in Fig. A1. When
␺0 ⫽ 0, the even subpulses {␣ie} are phase shifted by ␺0
radians with respect to {␣io}.This is similar to the analysis
FIG. A2. Generation of magnetization lobes for a type II spsp pulse.
in Appendix A, where the even tip angles were propor-
The amplitude of the magnetization lobes Mo generated by the odd
tional to B. Here the even tip angles are proportional to
subpulses bo(t) ⫽ A{␣o} is proportional to A. The amplitude of the
magnetization lobes Me generated by be(t) ⫽ B{␣e} is proportional to exp(i␺0). Therefore, in analogy with Appendix A, the ratio
B with alternating sign. Their sum, M, consists of alternating lobes between the even lobes M2n and the odd lobes M2n⫹1 is:
with amplitudes proportional to A ⫹ B and A ⫺ B.
M 2n⫹1 1 ⫺ exp共i␺ 0兲
⫽ [A3]
M 2n 1 ⫹ exp共i␺ 0兲
sum of Mo and Me, with lobes of alternating amplitude. M
consists of even lobes M2n, where n ⫽ 0, ⫾1, ⫾2, … with REFERENCES
amplitudes proportional to A ⫹ B and odd lobes M2n⫹1
1. Meyer CH, Hu BS, Nishimura DG, Macovski A. Fast spiral coronary
with amplitudes proportional to A ⫺ B, as shown in Fig. artery imaging. Magn Reson Med 1990;28:202–213.
A2. The frequency offset between an even lobe and an 2. Schick F, Forster J, Machann J, Kuntz R, Claussen CD. Improved clin-
adjacent odd lobe in M is 1/(2␶). Note that the frequency of ical echo⫺planar MRI using spectral-spatial excitation. J Magn Reson
the central even lobe M0 in Fig. A2 is the on resonance Imag 1998;8:960 –967.
3. Haase A, Frahm J, Hanicke W, Matthaei D. H1 NMR chemical shift
frequency, where the phase ␾ between adjacent subpulses
selective (CHESS) imaging. Phys Med Biol 1985;30:341–344.
is zero. 4. Block W, Pauly J, Kerr A, Nishimura D. Consistent fat suppression with
Consider a spin at ␷ ⫽ ␷1 and z ⫽ z1. The precession compensated spectral-spatial pulses. Magn Reson Med 1997;38:198 –
frequencies fp and fn are ␥Gz1 ⫹ ␷1 and –␥Gz1 ⫹ ␷1, respec- 206.
tively. For another spin at ␷ ⫽ ⫺␷1 and the same z ⫽ z1 the 5. Schick F, Forster J, Machann J, Huppert P, Claussen CD. Highly selec-
tive water and fat imaging applying multislice sequences without sen-
precession frequencies fp and fn are ␥Gz1 - ␷1 and ⫺␥Gz1 ⫺ sitivity to B1 field inhomogeneities. Magn Reson Med 1997;38:269 –
␷1 respectively. In summary: 274.
6. Schick F. Simultaneous highly selective MR water and fat imaging
using a simple new type of spectral-spatial excitation. Magn Reson Med
f p共␯ 1兲 ⫽ ⫺f n共⫺␯ 1兲 [A1]
1998;40:194 –202.
7. Pauly J, Spielman D, Macovski A. Echo planar spin-echo and inversion
A and B at any precession frequency f are proportional to pulses. Magn Reson Med 1993;29:776 –782.
the Fourier transform of the subpulse at f (11). Every RF 8. Purdy DE, Thomasson DM, Finn JP. Improved multi-slice fat suppres-
subpulse during negative gradient lobes is time-inverted sion in inhomogeneous fields using variable-frequency spectral-spatial
water excitation pulses. In: Proc ISMRM 3rd Annual Meeting, Nice,
with respect to every subpulse during positive gradient 1995. p 653.
lobes. Therefore, the coefficient A of the odd tip angles 9. Meyer CH, Pauly JM, Macovski A, Nishimura DG. Simultaneous spatial
{␣io} at any precession frequency f is equal to the coeffi- and spectral selective excitation. Magn Reson Med 1990;15:287–304.
420 Zur

10. Pauly J, Nishimura D, Macovski A. A k-space analysis of small tip-angle 16. Conolly S, Glover G, Nishimura D, Macovski A. A reduced power
excitation. J Magn Reson 1989;81:43–56. selective adiabatic spin-echo pulse sequence. Magn Reson Med 1991;
11. Pauly J, Nishimura D, Macovski A. A linear class of large tip-angle 18:28 –38.
selective excitation pulses. J Magn Reson 1989;82:571–587. 17. Ganin A, King K. Eddy current characterization using response surface
12. Pauly J, Le Roux P, Nishimura D, Macovski A. Parameter relations for methodology. In: Proc ISMRM Sixth Annual Meeting, Sydney, 1998. p
the Shinnar-Le Roux selective excitation pulse design algorithm. IEEE 2012.
Trans Med Imag 1991;10:53– 65. 18. Harvey PR, Rotem H, Stokar S. Fat free angiography using flow compen-
13. Shinnar M, Bolinger L, Leigh JS. The use of finite impulse response sated binomial spectral spatial slice variable tip angle RF slice selection.
filters in pulse design. Magn Reson Med 1989;12:75– 87. In: Proc ISMRM 5th Annual Meeting, Vancouver, 1997. p 1838.
14. Oppenheim AV, Schafer RW. Filter design techniques. In: Discrete-time 19. Fredrickson JO, Meyer C, Pelc NJ. Flow effects of spectral spatial excita-
signal processing, chapter 7. Englewood Cliffs, NJ: Prentice-Hall; 1989. tion. In: Proc ISMRM 5th Annual Meeting, Vancouver, 1997. p 113.
15. Hore PJ. Solvent suppression in Fourier transform nuclear magnetic 20. Spielman D. Dynamic shimming for multislice spectroscopy and im-
resonance. J Magn Reson 1983;55:283–300. aging. In: Proc ISMRM 3rd Annual Meeting, Nice, 1995. p 652.

You might also like