Magnetic Resonance in Medicine 57:1119 –1130 (2007)

Longitudinal Gradient Coil Optimization in the Presence

of Transient Eddy Currents
A. Trakic, F. Liu, H. Sanchez Lopez, H. Wang, and S. Crozier*

The switching of magnetic field gradient coils in magnetic res- and experimental compensation methods are also required
onance imaging (MRI) inevitably induces transient eddy cur- for optimal results (9,10).
rents in conducting system components, such as the cryostat During the design of conventional gradient coils (cylin-
vessel. These secondary currents degrade the spatial and tem- drical, planar, etc.), the current distributions in one or
poral performance of the gradient coils, and compensation
more gradient layers are commonly optimized to obtain a
methods are commonly employed to correct for these distor-
tions. This theoretical study shows that by incorporating the
target gradient uniformity in the imaging volume while
eddy currents into the coil optimization process, it is possible to satisfying other design constraints such as minimum in-
modify a gradient coil design so that the fields created by the ductance, resistance, leakage fields, force, torque, and
coil and the eddy currents combine together to generate a maximum gradient efficiency (1). Traditionally, these de-
spatially homogeneous gradient that follows the input pulse. sign approaches do not take eddy currents into direct
Shielded and unshielded longitudinal gradient coils are used to consideration when optimizing the gradient coil. Conse-
exemplify this novel approach. To assist in the evaluation of quently, during the pulsing of the gradient current, eddy
transient eddy currents induced within a realistic cryostat ves- currents are induced in the cryostat vessel and other con-
sel, a low-frequency finite-difference time-domain (FDTD)
ducting materials that lead to degradation of the target
method using the total-field scattered-field (TFSF) scheme was
performed. The simulations demonstrate the effectiveness of
field uniformity and the field stability.
the proposed method for optimizing longitudinal gradient fields In 1986, Turner and Bowley (13) were among the first to
while taking into account the spatial and temporal behavior of introduce an analytical technique for passive magnetic
the eddy currents. Magn Reson Med 57:1119 –1130, 2007. screening. Their study considered spatial eddy-current
© 2007 Wiley-Liss, Inc. variations in a thick, highly conductive, infinitely long
Key words: eddy currents; MRI; gradient; FDTD; cryostat aluminum shield as the secondary source contributing to
target gradient fields. By varying the single-layer gradient
During the pulsing of magnetic field gradients in MRI, coil positions to accommodate for the spatial presence of
multiexponentially decaying eddy currents are always in- eddy currents in the metal sheet, they were able to show
duced within the conducting materials of the MR imager. that good gradient uniformities can be obtained. However,
Eddy currents in cold, highly conductive radiation shields the authors assumed that the skin depth in the aluminum
of the superconducting magnet produce particularly long- shield was much smaller than the radial thickness of the
acting effects relative to the image acquisition period (1– screen, which resulted in a purely spatial problem descrip-
2). These secondary magnetic fields are known to cause tion. Obviously, in a realistic cryostat vessel that often
spatial and temporal degradation of the gradient unifor- consists of more than one metal shield, the aforemen-
mity within the imaging volume, which often results in tioned assumptions are not always valid. Therefore, we
undesired misregistration and intensity-phase variations believe that it is necessary to compute the exact spatial
in both images and spectra. (radial, axial, and azimuthal) and temporal eddy-current
With the recent push of MRI towards high signal-to- variations in a model cryostat/gradient pair.
noise ratios (SNRs) and improved image resolution, tre- In this work we explore the possibility of including
mendous efforts have been made to prevent and minimize predicted eddy currents in the design process of the gra-
the eddy-current fields. For instance, active screening is dient coil. Transient eddy currents induced within a real-
often engaged to minimize leakage fields and hence spa- istic cryostat vessel during the pulsing of gradient coils
tially and temporally complex residual eddy currents in- can be applied constructively together with the gradient
duced in the cryostat vessel (3– 8). Unfortunately, the use currents that generate them to obtain a homogeneous gra-
of active shielding layer(s) occupies vital space inside the dient in the imaging volume with a desired temporal vari-
bore of the magnet, increases system cost, and reduces ation. This can be achieved by simultaneously tailoring
gradient efficiency. Furthermore, residual eddy currents the spatial distribution and temporal preemphasis of the
are never completely removed through active shielding, gradient coil current to control the effects of eddy currents.
Although this approach could be applied to any known
gradient coil and cryostat geometry, for the purposes this
School of Information Technology and Electric Engineering, University of study it is constrained to actively shielded and un-
Queensland, Brisbane, Australia.
shielded, symmetric, z-axis gradient coils. The simplicity
Grant sponsor: Australian Research Council.
*Correspondence to: Professor Stuart Crozier, School of Information Tech-
of the 2D problem means that it requires less computa-
nology and Electrical Engineering, University of Queensland, St. Lucia, Bris- tional effort than the optimization of a transverse gradient
bane, Qld 4072, Australia. E-mail [email protected] coil with eddy currents.
Received 28 June 2006; revised 30 January 2007; accepted 21 February
2007.
In order to assist in the computation of the transient
DOI 10.1002/mrm.21243 eddy currents within a realistic cryostat geometry during
Published online in Wiley InterScience (www.interscience.wiley.com). the optimization of the longitudinal gradient coils, a low-
© 2007 Wiley-Liss, Inc. 1119
1120 Trakic et al.

frequency 2D axial-symmetric finite-difference time-do- currents. Pulsing of the gradient coil current generates a
main (FDTD) method based on the total-field scattered- time-varying primary magnetic field while at the same
field (TFSF) (14) technique was performed and verified time eddy currents are induced in nearby conductors,
with the commercially available software package FEM- which leads to formation of the time-varying secondary
LAB. The application of the TFSF-FDTD approach in cal- magnetic field. The primary magnetic field can be com-
culating the gradient-induced eddy currents is a novel puted using the Biot-Savart method. However, the compu-
feature of the design method. tation of the secondary magnetic field with analytical so-
The results of this study demonstrate that with a suitable lutions is not always mathematically straightforward, es-
temporal preemphasis and the optimized spatial current pecially when the eddy currents are produced in
distribution applied, the multiexponentially decaying geometrically complex conducting structures, such as a
eddy currents can be accurately predicted and construc- cryostat. Numerical schemes such as the FDTD and finite
tively used to assist in the design of longitudinal gradient element (FE) techniques are more suitable in these circum-
coils that generate very uniform and temporally stable stances, provided that repetitive computations of the gen-
gradient fields. erated eddy currents are sufficiently rapid during the op-
timization of the gradient coil.
MATERIALS AND METHODS Consequently, we extend our recent work on the low-
frequency FDTD method in cylindrical coordinates (11,12)
Longitudinal Gradient Coil Optimization in the Presence of
and improve the scheme by including the TFSF method-
Transient Eddy Currents
ology (14). With the TFSF-FDTD scheme in place, the
The approach outlined in this work considers the transient region of interest (ROI, i.e., the cryostat vessel) is enclosed
eddy currents induced in an example cryostat and the by four boundaries (Eq. [8]) to which the electromagnetic
corresponding effects in the diameter spherical volume ៮ r, H
(EM) fields (i.e., the H ៮ z , and E៮ ␸ field components in Eqs.
(DSV) region during the optimization of actively shielded/ [4]–[7]) from the gradient coil are propagated analytically
unshielded z-gradient coils. To achieve the desired gradi- using elliptic integrals for radial and axial magnetic and
ent homogeneity with desired temporal variation, both the azimuthal electric field components (16). The TFSF
spatial distribution and temporal preemphasis of the gra- boundaries mimic the distant gradient coil and act as
dient coil current are simultaneously tailored in the pres- artificial near-field sources. During FDTD field updating,
ence of the induced temporal eddy currents. It is essential the EM fields emerge from the TFSF boundaries, propagate
that the spatial distribution and temporal overshoot of the into free space, and are incident on the cryostat vessel,
gradient coil current be coupled during the optimization from which they are then partially reflected and partially
process in order to control the eddy-current effects in both transmitted. The region enclosed by the TFSF boundaries
space and time.
is termed the “total field” region in which both incident
and scattered fields exist, whereas the region surrounding
Eddy-Current nonoptimized z-Gradient Coil and Switching the TFSF boundaries contains only scattered fields and is
Sequence therefore called the “scattered field” region. The TFSF-
The shielded/unshielded z-gradient coil can be first pre- FDTD computational domain is terminated in a cylindrical
optimized without the consideration of eddy-current ef- split-version perfectly matched layer (PML), absorbing
fects using one of the many design schemes available (15). boundary condition (ABC), and perfect electrical conduc-
This z-gradient coil can then be used as a reference to the tor (PEC), which artificially truncate/absorb the scattered
eddy-current optimized coil. Alternatively, it is possible to fields (11,12). The application of the TFSF-FDTD method
start with an arbitrary current distribution as the initial to the longitudinal gradient coil/cryostat vessel model is
condition for the optimization of the spatial distribution shown in Fig. 1.
and temporal preemphasis of the gradient coil current. With the TFSF formulation, only the conductor and a
During this initial stage, it is possible to include any of the small region of air neighboring the conductor need to be
conventional optimization constraints and objectives. discretized. In this way the computational burden associ-
The stand-alone z-gradient coil is then theoretically ated with the redundant evaluation of the EM fields in free
placed within a realistic cryostat vessel and the desired space encircled by the cylindrical gradient coils can be
gradient switching sequence is applied. In this study the appreciably minimized. By employing the TFSF tech-
switching sequence is assumed to be trapezoidal and one- nique, we were able to accelerate the FDTD method by a
half period in duration. However, any continuous switch- factor of around 6. This computational speed-up is essen-
ing sequence type can be used, and more excitation cycles tial during the optimization and prototyping of the gradi-
can be included if required. ent coil in the presence of transient eddy currents. Briefly,
the central-difference method is applied to the Maxwell’s
Computation of the Primary and Secondary Magnetic curl equations in 2D axial-symmetric cylindrical coordi-
Fields nates to obtain a set of general FDTD update formulations
This section describes a new and efficient numerical for the fields inside the conductor (here the cryostat vessel)
method for the computation of eddy currents in MRI con- and surrounding medium (14):
ductors (first within the cryostat). The main purpose of
this proposed tool is to assist in the rapid optimization of
z-gradient coils in the presence of induced transient eddy H r兩 i,k⫹1/2
n⫹1/2
⫽ H r兩 i,k⫹1/2
n⫺1/2
⫹
⌬t
␮ 冉 E ␸兩 i,k⫹1
n

⌬z 冊
⫺ E ␸兩 i,k
n

[1]
Longitudinal Gradient Coil Optimization 1121

FIG. 1. MRI system setup showing the z-

gradient coil and the TFSF boundary en-
closing the cryostat vessel. The PEC en-
closes the TFSF-FDTD computational do-
main.

H z兩 1⫹1/2,k
n⫹1/2
⫽ H z兩 i⫹1/2,k
n⫺1/2
⫺
⌬t
␮r i⫹1/2 冉
r i⫹1E ␸兩 i⫹l,k
n

⌬r
⫺ r iE ␸兩 i,k
n

冊 [2]
where Hr and Hz denote the radial and axial components of
the magnetic field in [A/m] and E␸ is the azimuthal electric
field in [V/m]. The superscript n relates to the time in units

冉 冊 冉 冊
of ⌬t, and subscripts i and k relate to the spatial position in
2ε ⫺ ␴⌬t 2⌬t
E ␸兩 i,k
n⫹1
⫽ E ␸兩 i,k
n
⫹ units of ⌬r and ⌬z, respectively. The permeability ␮共r, z兲,
2ε ⫹ ␴⌬t 2ε ⫹ ␴⌬t permittivity ε共r, z兲, and material conductivity ␴共r, z兲 are

⫻ 冉 H r兩 i,k⫹1/2
n⫹1/2
⫺ H r兩 i,k⫺1/2
⌬z
n⫹1/2

⫺
H z兩 i⫹1/2,k
n⫹1/2
⫺ H z兩 i⫺1/2,k
⌬r
n⫹1/2

冊 [3]
defined at the center of the Yee cell. Variables ri , ri⫹1/ 2 , and
ri⫹1 are the radial coordinates in [m]. The TFSF boundary
is assumed to be a rectangular box with the following
Where the transient TFSF boundary conditions are discrete limits:
given by:
i ⫽ 兵i 僆 N兩i 0 ⱕ i ⬍ i 1其 ⵩ k ⫽ 兵k 僆 N兩k 0 ⱕ k ⬍ k 1其. [8]
BOTTOM INTERFACE:
The first terms in curly brackets on the right-hand side
H兩 n⫹1/2
⫽ 兵H 兩 n⫹1/2
其 (RHS) of Eqs. [4]–[7] are the generic time-stepping opera-
r i,k0⫺1/2 r i,k0⫺1/2 关1兴
tions as expressed in Eqs. [1]–[3]. The second terms on the

⫹ 冉 ⌬tE៮ ␸兩 i,k
␮⌬z
n
0
冊 E ␸兩 i,k
n⫹1
0
⫽ 兵E ␾兩 i,k
n⫹1
其 ⫺
0 关3兴 冉 2⌬tH៮ r兩 i,k
n⫹1/2

2ε⌬z ⫹ ␴⌬t⌬z
0 ⫺1/2
冊 RHS of Eqs. [4]–[7] are the added incident-wave terms that
arise from the artificial near-field sources present on the
TFSF boundaries. In general, the first terms on the RHS of
[4] Eqs. [4]–[7] are implemented first, after which the inci-
dent-wave terms are added. The temporal variation in the
TOP INTERFACE: fields due to the gradient coil (i.e., preemphasized trape-

冉 冊
zoidal excitation) is inherently introduced by the second
⌬tE៮ ␸兩 i,k
n
1 terms on the RHS of Eqs. [4]–[7]. A comprehensive treat-
H r兩 i,k
n⫹1/2
1⫹1/2
⫽ 兵H r兩 i,k
n⫹1/2
其 ⫺
1⫹1/2 关1兴
E ␾兩 i,k
n⫹1
⫽ 兵E ␾兩 i,k
n⫹1
其
1 关3兴
␮⌬z 1
ment of the TFSF theory, from which Eqs. [4]–[7] are

冉 冊
derived, is given in Ref. 14.
2⌬tH៮ r兩 r,k
n⫹1/2
1⫹1/2
⫹ [5] FDTD is a popular scheme for the solution of high-
2ε⌬z ⫹ ␴⌬t⌬z frequency problems, due to its simplicity and efficiency in
wave models. At low frequencies, however, Maxwell’s
LEFT INTERFACE: equations lose their symmetry, as the conducting current
is many orders of magnitude larger than the displacement

H z兩 in⫹1/2
0⫺1/2,k
⫽ 兵H z兩 in⫹1/2 其 ⫺
0⫺1/2,k 关2兴 冉 r i0⌬tE៮ ␸兩 in0,k
r i0⫺1/2␮⌬r 冊 E ␾兩 in⫹1
0,k
⫽ 兵E ␾兩 in⫹1 其
0,k 关3兴
current. Similarly to our recent work (12), the standard
FDTD scheme can be adapted to low frequencies by ap-
propriate linear scaling of certain physical constants. One
⫹ 冉 2⌬tH៮ z兩 in⫹1/2
2ε⌬r ⫹ ␴⌬t⌬r
0⫺1/2,k
冊 [6]
way to achieve the adaptation is to scale up the permittiv-
ity of free space ε0 and to leave the permeability ␮0 un-
changed. The downscaled speed of light constant c␣ is then
RIGHT INTERFACE: given by:

冉 冊
1 1
r i1⌬tE៮ ␸兩 in1,k c␣ ⫽ ⫽ [9]
H z兩 in⫹1/2
1⫹1/2,k
⫽ 兵H z兩 in⫹1/2 其 ⫹
1⫹1/2,k 关2兴
r i1⫹1/2␮⌬r
E ␾兩 in⫹1
1,k
⫽ 兵E ␾兩 in⫹1 其
1,k 关3兴
冑␮ 0共␣ε 0兲 冑␮ 0ε ␣

⫺ 冉 2⌬tH៮ z兩 in⫹1/2
2ε⌬r ⫹ ␴⌬t⌬r
1⫹1/2,k
冊 [7]
where ␣ is a dimensionless scaling factor, c␣ is the scaled
speed of the light constant [in ms⫺1 ], ␮0 is the permeability
1122 Trakic et al.

of free space 关4␲ 䡠 10⫺7 WbA⫺1 m⫺1 兴, ε0 is the permittivity Table 2

of free space 关8.8542 䡠 10⫺12 C2 N⫺1 m⫺2 兴 and ε␣ is the scaled Cryostat Vessel Properties
free space permittivity. The modification directly implies Inner
Conductivity Thickness
an alternation of the free space impedance, and if the # Material radius
(MS/m) (mm)
source is specified in terms of the electric field amplitude, (mm)
the magnetic field will be scaled up by a factor of 冑␣. 1 Stainless steel ⬃ 1.05 @ 300 K 5 450
However, if a magnetic field source is specified, the elec- 2 Aluminum ⬃ 100 @ 80 K 3 465
tric field will be scaled down by the same factor. In this 3 Aluminum ⬃ 500 @ 4.3 K 6 478
study the current source is in terms of the electric field,
which necessitates downscaling of the magnetic field re-
sults by the aforementioned factor at the end of simulation.
The TFSF-FDTD formulation allows direct and efficient domain. The dielectric properties of the cryostat vessel were
numerical computation of eddy currents induced within then input according to Table 2, and the gradient coil current
the conducting region (the cryostat vessel) during pulsing density of 1 A/m2 was amplitude-modulated by the pre-
of a magnetic field gradient. Using the superposition the- defined trapezoidal function with the parameters identical to
orem, the secondary transient magnetic field in the imag- the TFSF-FDTD model setup. The computational space was
ing volume of the gradient coil can be obtained by numer- then unevenly meshed with a total of 31560 FE nodes. The
ical integration, provided the eddy-current electric field in transient simulation was carried out for one excitation period
the conducting region is known. First the eddy current with temporal resolution of 10 ␮s. The relative and absolute
within every Yee cell of the conducting region I␸,edd y temporal tolerances were set to 1 ps. The direct UMFPACK
⫽ ␴E␸,edd y ⌬r⌬z is computed, followed by numerical inte- solver was then used to solve the transient problem, after
gration of the induced eddy current to obtain the corre- which the results were postprocessed.
sponding secondary axial magnetic field in the imaging
volume. z-Gradient Discrete Optimization Approach
To compensate for multiexponentially decaying secondary
Verification of the Computational Method
magnetic fields, current preemphasis is achieved using a
The low-frequency, cylindrical TFSF-FDTD formulation series of three exponential functions with characteristic
was coded in C and verified on a numerical model using RC-time constants and amplitudes that are readily adjust-
the FE software package FEMLAB©. An unshielded 20- able on a preemphasis network of the MR imager. The use
turn symmetric z-gradient was designed to have a diameter of only three exponential terms is sufficient in this case.
of 710 mm with axial coil coordinates detailed in Table 1. However, more exponential terms can be used if conver-
The gradient coil generates a DSV (here ⬍ 5% peak-to- gence of the optimization is inadequate.
peak (pp) gradient uniformity error) with radial and axial The time-varying primary Bzp (due to gradient coil) and
diameters of 0.45 m. secondary Bzs (due to eddy currents) axial magnetic fields add
A trapezoidal current density with the maximum ampli- vectorially to produce the resulting magnetic field gradient,
tude of 1 A/m2 at a frequency and rise time of 1 kHz and which must be homogeneous to 5%pp and have a pure
100 ␮s, respectively, was used to excite the coil. The trapezoidal temporal variation in the imaging volume. Com-
gradient coil is placed in a realistic cylindrical cryostat monly, the secondary magnetic field due to eddy currents
vessel consisting of stainless steel and two aluminum ra- distorts the spatial and temporal qualities of the gradient
diation shield layers. Table 2 lists the physical properties field. To compensate for these disturbances, the gradient-coil
of the cryostat vessel walls. The total length of the cryostat current distribution and temporal variation are optimized so
was assumed to be 1.4 m. that gradient-induced eddy currents will generate secondary
With radial and axial Yee cell sizes of 0.25 mm and magnetic fields that will add to the primary magnetic fields
7 mm, respectively, the cryostat bore was suitably dis- and produce a total spatially homogeneous and temporally
cretized to capture the spatial exponential decay (skin stable magnetic field gradient in the imaging volume. To
effect) of propagating waves as they penetrated the con- achieve this goal, a number N of temporal sample points
ducting walls. The 2D computational problem contained equally spaced in time are used to discretize the switching
61600 FDTD cells. sequence (i.e., the preemphasized trapezoidal waveform).
An identical model was implemented in FEMLAB as the The time spacing between adjacent temporal sample points
2D axial symmetric problem with the quasistatic solver for is related to the time increment ⌬␶ [s]. The nth-temporal
transient azimuthal currents. The gradient coil and the cry- sample point can be found at time tn关s兴:
ostat vessel were enclosed in a large rectangular air domain,
where axial symmetric and magnetic insulation boundary t n ⫽ n⌬␶ where n 僆 兵1…N其 [10]
conditions were appropriately applied to the borders of that
At every temporal sample point, the primary Bzp and
secondary Bzs axial magnetic fields are computed at M
Table 1 spatial sample points evenly distributed on the upper
Gradient Coil Parameters quarter of the DSV periphery, as illustrated in Fig. 1. The
distribution of spatial sample points was:
Axial coil coordinates (⫾ mm)
139 322 382 449 460 522 527 539 581 590
z m ⫽ R z ⫺ ⌬Z共m ⫺ 1兲
Longitudinal Gradient Coil Optimization 1123

冉 冉 冊冊
r m ⫽ R r sin a cos
zm
R2
where m 僆 兵1…M其 [11]
G z,max1 ⫽ max共Gz,max
tot
共tn 兲兲 ⵩ Gz,max2 ⫽ max共Gz,min
tot
共tn 兲兲 [15]

The average of Gz,max1 and Gz,max2 is computed and the

where rm and zm denote the radial and axial 2D axial- resulting magnitude is multiplied by the unit trapezoidal
symmetric cylindrical coordinates of the spatial sample response (discretized by N temporal sample points):
points in [m], respectively; ⌬Z is the fixed axial distance

冉 冊
between adjacent sample points in [m]; Rr and Rz are the
radial and axial DSV semi-axes in [m]; and m is the index G z,max1 ⫹ Gz,max2
G z,CD共t n兲 ⫽ 䡠 ⌶共tn 兲 [16]
of the spatial sample point. 2
The primary magnetic field due to the gradient coil is
computed with the Biot-Savart method and varies in time where ⌶共tn 兲 is the discretized unit trapezoidal response
in accordance with the preemphasized gradient coil cur- and Gz,CD 共tn 兲 is the closest desired (CD) magnetic field
rent excitation. The secondary magnetic field due to the gradient over time. Figure 2 visualizes the algorithm ap-
eddy currents induced in the cryostat is calculated using proach in generating the closest desired gradient.
the TFSF-FDTD method described previously, and its The next important step is to minimize the difference
time-space behavior is a function of the time-dependent between Gz,CD 共tn 兲 and the field extremes Gz,max tot
共tn 兲 and
primary magnetic field. Then the total time-varying axial
tot
Gz,min 共tn 兲. This is achieved by simultaneous optimization of
magnetic field Bz tot is given by: gradient coil current distribution and temporal current
preemphasis excitation used in driving the coil. Due to the
simplicity of the gradient coil, a discrete current distribu-
B ztot共r m,z m,t n兲 ⫽ B zp共r m,z m,t n兲 ⫹ B zs共r m,z m,t n兲 [12] tion was assumed and axial coil positions were varied in
the presence of eddy currents. However, analytical expres-
Here the term Bztot 共rm , zm , tn 兲 indicates that the total axial sions of the continuous current distribution can be also
magnetic field is calculated at a spatial sample point with optimized. In the preemphasis excitation, three amplitude
the radial and axial coordinates rm and zm , at time tn . Based and three RC-time constant coefficients were varied. In
on the total magnetic field, m total axial magnetic field this work we successfully employed the nonlinear Leven-
gradient values Gztot at every temporal sample point can be berg-Marquardt (LM) least-square optimization algorithm
derived based on the following difference equation: to minimize the following objective function:

冘 冋冉 冊冉 冊册
B ztot共r m⫹1,z m⫹1,t n兲 ⫺ B ztot共r m,z m,t n兲 N
G ztot共r m,z m,t n兲 ⫽ [13]
tot
Gz,max 共tn 兲 ⫺ Gz,CD 共tn 兲 2 Gz,min
tot
共tn 兲 ⫺ Gz,CD 共tn 兲 2
⌬Z ␺⫽ max ,
Gz,CD 共tn 兲 Gz,CD 共tn 兲
n⫽1

with m 僆 兵1…M ⫺ 1其. [17]

The eddy-current-distorted total gradient field on the
periphery of the imaging volume will always contain spa-
tial positions where gradient magnitudes are larger and/or The objective function ␺ is satisfied when its value at
smaller than the desired gradient field over time. Depend- every temporal sample point is less than 5%pp (the indus-
ing on the spatial distribution of eddy currents in the try standard).
cryostat vessel, there will be one spatial sample point
where a maximum gradient field waveform is traced in
time and usually another spatial different point with a
minimum waveform trace. The prime purpose of this al-
gorithm is to minimize the divergence between these two
extremes and the closest desired gradient field over time.
Therefore, for every temporal sample point, the maxi-
mum and minimum total magnetic field gradient values
are computed based on the dataset of total gradient values
on M spatial sample points:

tot
G z,max 共tn 兲 ⫽ max共Gztot 共rm ,zm ,tn 兲兲 ⵩ Gz,min
tot
共tn 兲
⫽ min共Gztot 共rm ,zm ,zm , tn 兲兲 [14]

tot
Here Gz,max 共tn 兲 and Gz,min
tot
共tn 兲 signify the maximum and
minimum total gradient field waveform traces over time,
respectively. These waveforms contain the eddy-current-
induced distortions.
To obtain the closest desired gradient of pure trapezoi-
dal waveform and with the largest possible magnitude
(i.e., for maximum gradient efficiency), we calculate the FIG. 2. Illustration of the algorithm approach for deriving the closest
maximum values of Gz,max tot
共tn 兲 and Gz,min
tot
共tn 兲: desired gradient over time, Gz,CD 共tn 兲.
1124 Trakic et al.

Table 3
Nonoptimized vs. Optimized Unshielded Z-Gradients
Axial coil coordinates (⫾ mm)
Coil A
Nonoptimized 140 340 345 355 425 455 463 480 485 500
Optimized 140 335 340 353 425 494 463 480 485 500
Change ⌬z 0 ⫺5 ⫺5 ⫺2 0 ⫹39 0 0 0 0
Coil B
Nonoptimized 139 322 382 449 460 522 527 539 581 590
Optimized 127 310 313 327 471 477 494 522 531 533
Change ⌬z ⫺12 ⫺12 ⫺69 ⫺122 ⫺11 ⫺45 ⫺33 ⫺17 ⫺50 ⫹57

Single-Layer Longitudinal Gradient Optimization Example mum performance with eddy currents, it was necessary to
A shorter version of the z-gradient coil from Table 1, with simultaneously reoptimize the z-gradient coil wire posi-
an overall length of 1 m, was first optimized without the tions and the preemphasis excitation. The optimization
presence of eddy currents using the LM-optimization was terminated when the maximum deviation of less than
method. It was assumed that the coil is made of circular 5%pp from the desired gradient over time was achieved.
copper wires with 3-mm diameter. Rectangular copper
wires could have been also used. Furthermore, it was Study of Different Gradient Rise Times
assumed that the target field uniformity, gradient effi-
ciency, adequate gradient, and DSV geometry are the only It is well known that eddy currents depend not only on the
applicable optimization constraints. The gradient coil was gradient coil/cryostat geometry and intrinsic properties,
then positioned inside the three-layer cryostat vessel de- but also on the gradient current rise time and switching
tailed in Table 2, and a maximum current of 1A was sequence employed (17). It is therefore essential to study
pulsed trapezoidally through the gradient coil at a fre- different excitations and the resulting implications for eddy-
quency of 1 kHz with 100 ␮s rise time. For illustration current spatial and temporal behavior to ensure that the
purposes, only one-half of the trapezoidal period was dis- eddy-current optimized gradient design is stable. In this
cretized with N ⫽ 50 temporal sample points and used in study the axial coil positions of the eddy-current opti-
all optimization calculations. The DSV size constraint was mized coil A (Table 3) were assumed to be constant, and
assumed to be identical to the stand-alone single-layer thus only the current overshoot was tailored to account for
z-gradient coil (i.e., 0.45 ⫻ 0.45 m). Due to the model the rise-time-specific eddy currents induced in the cryo-
symmetry, only one-quarter of the DSV periphery was stat bore. Four simulations were conducted with rise times
discretized by M ⫽ 15 spatial sample points. For maxi- of 150, 200, 250, and 300 ␮s, respectively.

FIG. 3. Transient validation results: (a) first

wall (stainless steel), (b) second wall (alumi-
num), and (c) third wall (aluminum), where
waveforms 1 and 2 are spatial points taken
at the inner surface 0.21 m and 0.07 m
axially from the center of axis of symmetry,
respectively. Subplot d represents the sec-
ondary axial magnetic flux density due to
the whole cryostat vessel at R ⫽ 0; Z ⫽
0.225 m.
Longitudinal Gradient Coil Optimization 1125

Actively Shielded Longitudinal Gradient Optimization The eddy-current nonoptimized, actively-shielded z-

Example gradient coil was then situated inside the cryostat vessel
In this part of the study an actively shielded stand-alone as detailed in Table 1, and a maximum current of 1A was
symmetric z-gradient coil was optimized without the con- pulsed trapezoidally through the gradient coil at a fre-
sideration of eddy currents. The gradient-coil optimization quency of 1 kHz with 100␮s rise time. A spatial and
aims were to minimize the leakage fields at the first eddy- temporal discretization (N ⫽ 50, M ⫽ 15) identical to the
current source (inner cryostat wall), maximize the gradient single-layer z-gradient coil optimization was used. The
efficiency, and obtain at most 5%pp gradient uniformity in gradient current overshoot and axial coil positions in
a 500 ⫻ 560 mm working volume. The shielding effective- both the primary and secondary gradient layers were
ness of the actively shielded gradient coil is computed at then optimized to account for the presence of secondary
100 spatial points equally spaced along the z-axis (z ⫽ magnetic fields in the imaging volume. The leakage
⫺ 1. . .1m) at radial distances of 0.45 m (first eddy-current fields were also minimized during the same optimiza-
source) and 0.5 m (inner surface of the first superconduct- tion process.
ing magnets). The inner and outer coil diameters were
fixed at 700 mm and 800 mm, respectively. It was assumed
RESULTS
that this coil is also made of circular copper wires with
3-mm diameter. During the optimization procedure, the Verification of the Computational Method
axial half lengths of the primary and secondary layers were On a dual XEON 3.6GHz/4GB RAM PC platform and with
constrained to 480 mm and 530 mm, respectively. a scaling factor of ␣ ⫽ 9 䡠 1010 , the TFSF-FDTD method

FIG. 4. Gradient field vs. time due to the

single-layer longitudinal gradient coil (here
coil A) at 500A transport current and eddy
currents induced in the cryostat vessel (left),
and gradient uniformity snapshot at 0.11 ms
after the start of the gradient waveform
(right). The gradient waveforms (left) show
the (A) closest desired gradient Gz,CD 共tn 兲, (B)
minimum total gradient Gz,min tot
共tn 兲, (C) maxi-
tot
mum total gradient Gz,max 共tn 兲, (D) sketch of
scaled preemphasized current excitation.
and (E) total gradient in the center of the
tot
DSV over time Gz,center 共tn 兲. The subplots il-
lustrate (a) a spatially and temporally non-
optimized coil, (b) a spatially nonoptimized
coil driven with optimal preemphasis, and
(c) a spatially optimized gradient coil driven
with optimal preemphasis.
1126 Trakic et al.

FIG. 5. Gradient field vs. time due to single-

layer longitudinal gradient coil (here coil A)
at 500A transport current and eddy currents
induced in the cryostat vessel, where the
axial coil positions have been fixed and only
the rise-time-specific preemphasis has
been tailored: (a) sketch of scaled preem-
phasized current excitation ⌶共tn 兲, (b) closest
desired gradient uniformity over time
Gz,CD 共tn 兲, (c) optimized gradient uniformity
over time where eddy-current effects have
been taken into account, and (d) gradient
uniformity over time degraded due to eddy
currents (with optimal preemphasis not ap-
plied).

required around 8 s to obtain the transient field result for fields. The terms “coil A” and “coil B” in Table 3 signify
one period of the trapezoidal sequence. The model prob- the short and long single-layer z-gradient coils.
lem was that described in the Materials and Methods sec- Figure 4a (left) illustrates the deviation of the gradient
tion under “Verification.” In contrast, the commercial soft- uniformity from the target gradient over time for an eddy-
ware package FEMLAB took around 34 s on the same current nonoptimized coil (Table 3, coil A—nonopti-
platform to solve the identical problem. Although the mized) in the presence of eddy currents. The results illus-
number of nodes in the FE setup was about twice as large trated in Fig. 4b (left) indicate that if only the optimal
as the number of cells used in the FDTD method, the preemphasis is applied to the eddy-current nonoptimized
TFSF-FDTD formulation was around four times faster than coil A, the target field uniformity cannot be guaranteed.
the FE method. Figure 4c (left) shows that with simultaneous application
Figure 3 compares the transient, azimuthal eddy-current of optimal current preemphasis and appropriate variation
electric fields at different locations within the cryostat in axial coil positions (Table 3, coil A, optimized), the
vessel between the TFSF-FDTD and FEMLAB simulations. target gradient uniformity (⬍5% pp gradient uniformity
The results are in good agreement. error) over time can be obtained. The scaled version of the
required preemphasized current excitation is also illus-
Single-Layer Gradient Optimization Results trated in the same plot. Figure 4 (right) depicts the associ-
On the dual XEON PC platform, the single-layer z-gradient ated worst-case gradient uniformity snapshots within the
coil eddy-current-based optimization consumed around working volume at 500A transport current for all three
28 min of CPU time and 6.3 MB of memory. Table 3 lists aforementioned cases at 0.11 ms after the start of the gra-
the axial coordinates of the eddy-current nonoptimized dient waveform. The same optimization procedure was
and optimized gradient coils, indicating the changes in the applied to the longer z-gradient coil from Table 1 with
axial positions required to accommodate the eddy-current similar CPU time and memory requirements, and analo-

FIG. 6. Spatial gradient deviation from the

target gradient along the axial coordinates
of the upper-quarter DSV periphery at
0.11 ms: (a) spatially nonoptimized axial coil
positions of coil A and optimal current pre-
emphasis (Fig. 5), (b) spatially optimized ax-
ial coil positions of coil A and optimal cur-
rent preemphasis (Fig. 5), (A) target gradient
uniformity, and gradient uniformity at (B)
150 ␮s, (C) 200 ␮s, (D) 250 ␮s, and (E) 300
␮s rise time, respectively.
Longitudinal Gradient Coil Optimization 1127

Table 4
Nonoptimized vs. Optimized Actively Shielded z-Gradients
Axial coil coordinates (⫾ mm)
First layer
Nonoptimized 98 196 221 282 362 395 411 419 452 455 458 461 464 467 480
Optimized 95 193 223 286 362 392 410 415 445 449 455 459 464 472 479
Change ⌬z ⫺3 ⫺3 ⫹2 ⫹4 0 ⫺3 ⫺1 ⫺4 ⫺7 ⫺6 ⫺3 ⫺2 0 ⫹5 ⫺1
Second layer
Nonoptimized 115 198 269 341 380 416 434 463 482 526
Optimized 112 200 278 339 380 413 433 467 483 528
Change ⌬z ⫺3 ⫹2 ⫹9 ⫺2 0 ⫺3 ⫺1 ⫹4 ⫹1 ⫹2

gous results to Fig. 4 were obtained (not shown). By incor- first superconducting magnets). It can be observed that high-
porating the eddy currents in the design process, it may be er-frequency spatial oscillations in the eddy-current source
possible to shape larger DSV regions than by using the are present when the actively shielded gradient coil is placed
single-layer gradient coil alone. within the vicinity of the cryostat bore.
Figure 8 shows the gradient uniformity within the working
Gradient Rise Time Results volume at 500A transport current for nonoptimized (a) and
In Fig. 5 the axial coil positions of the eddy-current opti- optimized (b) case of the actively shielded gradient coil at
mized coil A (Table 3) were assumed to be constant and 0.11 ms after the start of the gradient waveform. Figure 8a
only the current overshoot was tailored to account for the illustrates that active shielding does not guarantee the target
rise-time-specific eddy-current temporal effects generated field uniformity. By applying the optimization method, the
in the imaging volume. As expected, the slower the gradi- target field uniformity can be obtained (Fig. 8b).
ent current rise time, the smaller the temporal eddy-cur- Figure 9 shows the deviation of the gradient uniformity
rent distortions and hence less preemphasis is required. from the desired gradient over time and the optimized solu-
Increasing the current rise time from 150 to 300 ␮s de- tion that follows the prescribed trapezoidal pattern with less
creased the required overshoot in the gradient-coil current than 5%pp deviation from desired gradient field. Also illus-
by a factor of more than 2.2. Figure 6 illustrates snapshots trated is the scaled version of the current preemphasis.
of the spatial gradient deviation from the target field uni- Figure 10 illustrates the eddy-current density at the inner
formity along the axial coordinates of one-quarter DSV surface of three cryostat walls along the z-axis vs. time for the
periphery (0.45 ⫻ 0.45 m) for the different rise-time con- nonoptimized and optimized unshielded and actively
ditions at 0.11 ms. shielded z-gradient coils. The plot shows the effects of pre-
Figure 6a shows the spatial gradient deviation assuming emphasis and perturbed current distribution on the spatial
the fixed, eddy-current nonoptimized coil A (Table 3) and and temporal characteristics of the induced eddy currents.
the rise-time-dependent overshoots from Fig. 6, while Fig.
6b) illustrates results for the eddy-current optimized coil
A. The preemphasized, spatially nonoptimized gradient
coil caused more than 10%pp gradient deviation from the
target field uniformity under all rise-time conditions that
were studied (Fig. 6a). Other gradient coils could poten-
tially generate more deviation under these circumstances.
However, in the case of the preemphasized, spatially op-
timized gradient coil, the gradient deviation was below
5%pp for all rise times (Fig. 6b). These results illustrate
that a priori optimization of spatial current distribution is
advantageous for achieving the target field uniformity and
ensuring stability. In addition, Fig. 6 illustrates that in
both cases (subplots a and b), the spatial gradient deviation
tends to decrease slightly as the rise time increases.

Actively Shielded Gradient Optimization Results

Using the same computing platform detailed in “Verification
of the Computational Method” above, it took around 52 min
to obtain the solution for the actively shielded case. Table 4
lists the axial coordinates for the nonoptimized and opti-
FIG. 7. Shielding characteristics at r ⫽ 0.45m (inner surface of
mized gradient coils, indicating the changes in axial posi- cryostat) and r ⫽ 0.50m (inner surface of first superconducting
tions required in the presence of eddy currents. Figure 7 main magnet) along the z-axis for the actively shielded and un-
illustrates the shielding performance of the stand-alone ac- shielded longitudinal gradient coils. The shielding effectiveness
tively shielded z-gradient assembly at radial distances of 兩Bshielded /Bunshielded 兩2 where B ⫽ 冑Bz 2 ⫹ Br 2 (total magnetic flux den-
0.45 m (first eddy-current source) and 0.5 m (inner surface of sity) is around 0.012745.
1128 Trakic et al.

FIG. 8. Gradient uniformity snapshots at

500A transport current at 0.11 ms of trape-
zoidal excitation due to the (a) nonoptimized
gradient coil and induced transient eddy
currents, and (b) spatially and temporally
optimized gradient coil and eddy currents.
The gradient uniformity in b is preserved
during the whole excitation.

DISCUSSION
The results obtained in this theoretical study clearly dem-
onstrate the possibility of including predicted transient
eddy currents into the design process for shielded and
unshielded longitudinal gradient coils. Although this
study focused on axial gradients, it certainly seems possi-
ble to apply the described method to more general cases
(i.e., transverse, planar, and other coil geometries). How-
ever, the optimization of 3D gradient systems (i.e., trans-
verse cylindrical coils) could be computationally inten-
sive. This computational burden can be effectively miti-
gated through parallel computing, which we recently
demonstrated for Cartesian and cylindrical systems (18).
A numerical description of the eddy current associated
with gradient switching is of significant utility in the MRI
field. A robust analysis scheme allows for a more complete
design and optimization of gradient coil/cryostat/magnet
combinations and gives indicative predictions of the com-
pensation waveforms required (11,12). To assist in the
efficient computation of EM fields during gradient switch-
ing and thus enable the optimization of z-gradient coils in
the presence of eddy currents, we extended our recently
proposed low-frequency FDTD method by including the
TFSF scheme, which effectively accelerates the original
computational method. Unlike most analytical formula-
tions, the low-frequency TFSF-FDTD numerical algorithm
can be used to compute spatially distributed temporal
eddy currents in realistic cryostat vessel geometries and
possibly other conducting materials induced by a z-gradi-
ent coil with almost any continuous switching sequence
applied. In a typical experiment, the TFSF boundaries,
PEC, and PML layers would be placed around a set of
conducting objects of interest, mainly close to the cryostat
vessel, and the TFSF-FDTD simulation would be carried
out. We note here that we have focused only on the eddy
FIG. 9. The gradient field vs. time for actively shielded coil at currents induced in the cryostat as they are the main
500A transport current: (a) sketch of scaled preemphasized cur- sources of the secondary magnetic fields, and RF-related
rent excitation, (b) closest desired gradient uniformity over time electrical components (RF shielding and coils) are ne-
G z,CD共t n兲, (c) optimized gradient uniformity where eddy currents
glected in this formulation. If all of the conducting MRI
are taken into account, and (d) maximum spatial gradient unifor-
components are considered, a domain-decomposition ap-
mity degradation over time due to eddy currents (nonoptimized
gradient case). Shown below is the focused plot at around 0.1 ms proach can then be utilized. We emphasize also that this
of trapezoidal excitation, illustrating that the optimized gradient efficient TFSF formulation will make the eddy-current
uniformity is following the closest desired gradient (e.g., trape- evaluations relatively easy to implement for 3D gradient
zoidal profile). coil designs.
Longitudinal Gradient Coil Optimization 1129

FIG. 10. Eddy-current density along the z-

axis vs. time at the inner surface of the
stainless steel, aluminum I, and aluminum II
cryostat walls for the nonoptimized and op-
timized unshielded (coil A) and actively
shielded symmetric z-gradient coils at 1A
transport current.

In the optimization approach we have assumed that the uniform gradient fields with pure trapezoidal temporal
target field uniformity, adequate gradient and DSV geom- characteristics, provided the single-layer z-gradient coil
etry, leakage fields, gradient efficiency and eddy-current current is both spatially and temporally optimized in the
spatial and temporal distortions are the only applicable presence of eddy currents. Therefore, depending on the
optimization constraints. Obviously, other parameters can maximum possible performance of the gradient amplifi-
be included in the optimization kernel. For instance, dur- ers/preemphasis unit, it may be possible to remove the
ing each optimization loop, the total inductance and resis- active gradient screen and use a suitably designed cryostat
tance can be calculated and their implications on gradient vessel as the controlled passive (“natural”) shield instead.
switching performance can be included in the model. The This could potentially improve the gradient-coil efficiency
approach presented here is somewhat different from tra- (since the cryostat is further away from the primary coil
ditional methods in that the B0-shifts due to eddy currents than the conventional active shield), minimize the gradi-
are not separately corrected for, since eddy-current fields ent construction costs, and, most importantly, free a sig-
are strictly taken into account during the system optimi- nificant amount of vital space inside the MRI bore, which
zation and B0-shifts are inherently minimized. could then be used for other purposes. The downsides of
The numerical results obtained in the eddy-current- this approach include greater field penetration and in-
based single-layer z-gradient optimization study promote creased heat load in the cryostat, and the fact that the
the possibility of employing the primary gradient layer design optimization is specific to a gradient/cryostat pair.
and the cryostat vessel as the passive shield to generate We note that designs in between a shielded and an un-
1130 Trakic et al.

shielded system are possible using this method; that is, a account the transient eddy currents induced in the cryo-
“lightly” shielded system with high efficiency and good stat vessel. The compensation can be achieved through
temporal stability and linearity can be obtained. coupled optimization of spatial distribution and preem-
Based on the results illustrated in Figs. 4 – 6, the re- phasis of the gradient-coil current that accounts for the
quired spatial homogeneity can not be always realized presence of the eddy-current field, and should be prefera-
with the current preemphasis alone, and spatial reoptimi- bly applied during the design process of the gradient coil.
zation of gradient current distribution to account for the Through a priori spatial correction of eddy-current fields,
secondary fields in the imaging volume is also required for the preemphasis can be tailored to achieve improved target
the optimal results. The corrections in wire positions de- field uniformities over time under different rise-time con-
tailed in Tables 3 (single-layer z-gradient) and 4 (actively- ditions. In future work, we intend to study the implication
shielded z-gradient) to accommodate for the transient of different imaging sequences, rise times, and gradient
eddy currents are feasible. Clearly, with active shielding coil/cryostat vessel geometries on the target gradient uni-
the magnitude of the eddy currents in the cryostat is re- formity using this new method. Furthermore, we plan to
duced. An approximately 4.0% preemphasis overshoot extend the method to three dimensions and apply paral-
was required, and yet variation in axial coil positions lelism to the algorithm in order to make eddy-current
remains advantageous for producing the best coil perfor- optimization of transverse gradient coils possible.
mance. The unshielded gradient coil (Fig. 4 and Table 3) is The proposed method provides distinct advantages for
around 1.46 times more efficient than the shielded gradi- systems engineering of gradient coil/magnet pairs because
ent coil (Fig. 9 and Table 4) under the same transport the cryostat and gradient set can be designed together in
current conditions. Obviously, displacing the active shield terms of eddy-current induction and field profiles. This
further away from the primary coil would improve the also allows the rapid design and prototyping of cryostats of
gradient efficiency. However, the unshielded coil requires various shapes and properties.
much more voltage overdrive (Fig. 4) than the actively
shielded gradient and hence imposes a greater burden on
REFERENCES
the amplifier system. The final result of these gradient-coil
1. Badea EA, Craiu O. Eddy current effects in MRI super-conducting
optimizations is not a highly stable gradient waveform, but magnets. IEEE Trans Magn 1997;33:1330 –1333.
rather a gradient waveform that is stable enough to be 2. Ries G. Eddy current transients and forces in cryostat walls of super-
corrected using typical magnitudes of amplifier preempha- conducting solenoids. IEEE Trans Magn 1988;24:516 –519.
sis. The final system specification for gradient waveform 3. Shvartsman S, Morich M, Demeester G, Zhai Z. Ultrashort shielded
gradient coil design with 3D geometry. Concepts Magn Reson Part B
fidelity on clinical scanners is typically on the order of
2005;26B:1–15.
0.02% (after eddy-current calibrations are completed). 4. Mansfield P, Chapman B. Active magnetic screening of static and
Figure 10 illustrates the effects of preemphasis and per- time-dependent magnetic field generation in NMR imaging. J Phys E Sci
turbed current distribution on the spatial and temporal Instrum 1988;19:540:545.
characteristics of the induced eddy currents in cases of 5. Crozier S, Doddrell DM. A design methodology for short, whole-body,
shielded gradient coils for MRI. Magn Reson Imaging 1995;13:615– 620.
shielded and unshielded z-gradients. Four cases are con- 6. Jehenson P, Westphal M, Schuff N. Analytical method for the compen-
sidered: 1) eddy-current nonoptimized single-layer, 2) eddy- sation of eddy-current effects induced by pulsed magnetic field gradi-
current optimized single-layer, 3) eddy-current nonopti- ents in NMR systems. J Magn Reson 1990;90:264 –278.
mized actively shielded z-gradient coil, and 4) eddy-cur- 7. Bowtell R, Mansfield P. Gradient coil design using active magnetic
screening. Magn Reson Med 1991;17:15–21.
rent optimized actively shielded z-gradient coil. As
8. Brown RW, Shvartsman S. Supershielding: confinement of magnetic
expected, the flow direction of induced eddy currents at fields. Physics Rev Lett 1999;83:1946 –1949.
100 ␮s is opposite to the nearest gradient currents, while 9. Gach HM, Lowe IJ, Madio DP, Caprihan A, Altobelli SA, Kuethe DO,
the eddy-current temporal variation resembles the switch- Fukushima E. A programmable preemphasis system. J Magn Reson Med
ing sequence employed (trapezoidal: 1 kHz frequency, 1998;40:427– 431.
10. Crozier S, Eccles CD, Beckey FA, Fields J, Doddrell DM. Correction of
100 ␮s rise time) in driving the gradient coils. At 0.5 ms, eddy-current-induced B0 shifts by receiver reference-phase modula-
the eddy currents are reversed in direction to counteract tion. J Magn Reson 1992;97:661– 665.
the new current direction in the gradient coils. In all cases 11. Liu F, Crozier S. An FDTD model for calculation of gradient-induced
the largest eddy currents are induced in the second cryo- eddy currents in MRI system. IEEE Trans Appl Superconduct 2004;14:
1983–1989.
stat wall, as this is the first EM shield with reasonably high
12. Trakic A, Wang H, Liu F, Sanchez Lopez H, Crozier S. Analysis of
material conductivity. The EM energy diffuses slowly transient eddy currents in MRI using a cylindrical FDTD method. IEEE
through this second layer and, as expected, the eddy cur- Trans Appl Superconduct 2006;16:1924 –1936.
rent excitation in the third wall is considerably delayed in 13. Turner R, Bowley RM. Passive screening of switched magnetic field
time. The current overshoot in the optimized unshielded gradients. J Phys E Sci Instrum 1986;19:876 – 879.
14. Taflove A. Computational electromagnetics—the finite-difference time-
(single-layer) z-gradient is evident. In addition, one can domain method. London: Artech; 1995.
observe different decay terms within the cryostat vessel 15. Turner R. Gradient coil design: a review of methods. Magn Reson
due to diverse material properties and unique wall geom- Imaging 1993;11:903–920.
etries within the cryostat vessel. 16. Abramowitz M, Stegun IA. Handbook of mathematical functions.
London: Dover; 1965.
17. Ahhn CB, Cho ZH. Analysis of eddy currents in nuclear magnetic
CONCLUSIONS resonance imaging. Magn Reson Med 1991;17:149 –163.
18. Wang H, Trakic A, Liu F, Bialkowski ME, Crozier S. A high perfor-
Longitudinal gradient coils can be implemented to pro- mance FDTD scheme for MRI. In: Proceedings of the 14th Annual
duce very uniform gradient fields over time by taking into Meeting of ISMRM, Seattle, WA, USA, 2006 (Abstract 1385).