Magnetic Resonance Im

Robust prescan calibration for multiple spin-echo sequences:

application to FSE and b-SSFP

Chris V. Bowena,4, Joseph S. Gatib, Ravi S. Menonb,c,d

aInstitute for Biodiagnostics (Atlantic), National Research Council, Halifax, Nova Scotia, Canada B3H 3A7bThe Laboratory for Functional Magnetic Resonance Research, Robarts Research Institute, London, Ontario, Canada N6A 5K8

cDepartment of Medical Biophysics, University of Western Ontario, London, Ontario, Canada N6A 5K8dDepartment of Diagnostic Radiology and Nuclear Medicine, University of Western Ontario, London, Ontario, Canada N6A 5K8

Received 16 January 2006; revised 29 March 2006

Abstract

The collection of fast imaging techniques that use multiple spin-echo (MSE) sequences relies on a precise phase relationship between spin

echoes and stimulated echoes that form along the radiofrequency refocusing pulse train. Failure to achieve this condition produces dark

banding artifacts that result from destructive interference between signal coherence pathways. Satisfying this condition on the microsecond

timescale required is technically challenging for conditions involving strong diffusion-weighted gradients, for arbitrary orientation

acquisitions and at large field strengths with high-resolution acquisitions. Two clinically significant MSE sequences, fast spin echo (FSE) and

balanced steady-state free precession (b-SSFP), are investigated in this work using a 4-T whole-body scanner. We developed a readout-

projection-based prescan technique that ensures coherent signal formation by utilizing banding artifacts to automatically adjust gradient

balance. Subsequent image acquisition using the results of this prescan permits the formation of coherent-echo images, which are robust

under challenging imaging conditions. The robustness of this approach is demonstrated for FSE and b-SSFP images obtained from the knees

of human volunteers. We believe that the use of this prescan calibration technique for the alignment of signal pools in MSE sequences is

critical at high fields and will facilitate the implementation of high-quality clinically significant sequences such as FSE and b-SSFP.

D 2006 Elsevier Inc. All rights reserved.

Keywords: Fast spin echo; TrueFISP; FIESTA; SSFP; Artifact; Prescan; Knee; MRI

1. Introduction

In recent years, the use of high-field magnetic resonance

scanners for human body imaging has grown rapidly owing

to improved signal-to-noise ratio (SNR) and achievable

susceptibility contrast [1,2]. These advantages may be

employed to acquire images with greater resolution (hence,

greater anatomic detail) or images with greater sensitivity to

susceptibility-related signal intensity changes, as with

functional magnetic resonance imaging (fMRI) [3,4].

However, coupled with these benefits are additional

technical challenges resulting from an increased difficulty

in producing adequate shim quality (B0) and radiofrequency

(RF) homogeneity (B1) [5]. These problems are most

apparent within the family of multiple spin-echo (MSE)

sequences, where signals from multiple refocusing path-

0730-725X/$ – see front matter D 2006 Elsevier Inc. All rights reserved.

doi:10.1016/j.mri.2006.04.008

4 Corresponding author. Tel.: +1 902 473 1858; fax: +1 902 473 1851.

E-mail address: chris.bowen@nrc-cnrc.gc.ca (C.V. Bowen).

ways must coherently combine to prevent destructive signal

interference [6,7]. Any microsecond-scale misalignment of

primary spin echoes and stimulated echoes within the

readout window produces a large spatial variation of image

signal intensity for MSE sequences, such as fast spin echo

(FSE) [8,9] and balanced steady-state free precession (b-

SSFP) [10,11].

Although high-quality FSE and b-SSFP images have

become commonplace in the clinical setting over the past

decade,MSE acquisitions can present signal incoherence arti-

facts at any field strength. In addition to specific difficulties

presented at high fields (greater than 3 T) [12], the use of

transverse magnetization preparation schemes, such as diffu-

sion, T2 or T2* weighting, can produce phase errors that

cause destructive interference between coherence pathways

[13,14]. A great number of system calibration imperfections

can be problematic for MSE sequences. Difficulties associ-

ated with gradient fidelity, such as unbalanced amplifier

gain along each gradient polarity, amplifier nonlinearity,

aging 24 (2006) 857–867

C.V. Bowen et al. / Magnetic Resonance Imaging 24 (2006) 857–867858

unaccounted gradient delays or imperfect eddy current

compensation, are not uncommon [14–17]. Although gradi-

ent fidelity problems are an ongoing consideration even with

low-field-strength MSE acquisitions, they are of particular

concern with increased field strength where high-resolution

acquisitions are often used to exploit SNR advantages. For a

given gradient performance, high-resolution acquisitions

require both increased gradient strength and RF pulse

refocusing times. Each condition further exposes technical

imperfections in gradient fidelity and field homogeneity and

increases difficulty in aligning multiple signal coherence

pathways relative to that normally observed clinically.

Compensation for these problems through the use of

postprocessing algorithms is not possible with MSE acquis-

itions, as with echo-planar imaging [18]; thus, a prospective

correction is required.

A number of pulse sequence modifications have been

proposed to suppress artifact formation in the presence of

technical imperfections. In the case of FSE, suggested

approaches include ensuring that refocusing pulses are

precisely 1808 [19], suppressing all coherence pathways

other than the primary echo [13,20] or using a non-Carr–

Purcell–Meiboom–Gill (CPMG) phase-cycling scheme for

the refocusing pulses to limit sensitivity to phase errors [21].

These approaches are limited by practical considerations or

by reduction in SNR efficiency due to the suppression of

useful signal coherence pathways [22]. For b-SSFP acquis-

itions, the reduction of artifact through maximum intensity

projections of multiple b-SSFP images, where the artifact

has been deliberately shifted to alternate spatial locations, is

feasible but suffers from reduction in SNR efficiency [23].

Two clinically significant MSE sequences, FSE and

b-SSFP (also known as TrueFISP and FIESTA), are

investigated in this work using a 4-T whole-body scanner.

We developed a readout-projection-based prescan technique

that automatically adjusts the read/slice gradient balance to

ensure coherent signal formation to first order in these

two dimensions. This technique utilizes banding artifacts

that form through a deliberate misadjustment of readout

gradient balance from a nominal value in order to calculate

calibration parameters. The theoretical basis for artifact

formation with FSE and b-SSFP acquisitions is first reviewed

prior to the description of a prescan strategy suggested from

this. The robustness of the prescan approach is demonstra-

ted through multiple prescan acquisitions from the heads

of volunteers. Subsequent image acquisition using the results

of this prescan calibration is demonstrated for FSE and

b-SSFP images obtained from the knees of human volunteers.

The significant signal heterogeneity produced by the anato-

my of the knee presents a challenging condition for the

prescan approach. We believe that the use of this prescan

calibration technique for the alignment of signal pools

in MSE sequences is critical at high fields for high-

resolution acquisitions and will facilitate the implementation

of high-quality clinically significant sequences such as FSE

and b-SSFP.

2. Theory

The generation of artifact-free images for the family of

MSE sequences relies upon a precise phase relationship

between transverse magnetization and refocusing pulse B1

vectors [6,13,22]. Failure to achieve this condition results in

destructive interference between echoes that are formed

through a broad range of signal coherence pathways; a

significant problem with as many as 750 distinct signal

pathways present after only eight spin echoes [24].

Deviation from proper phase coherence can result from

nonidealities associated with either the phase of transverse

magnetization or the phase of the refocusing pulse B1 field.

The phase of different spin packets in the transverse plane

(which sum to transverse magnetization) is altered by

factors such as imperfectly compensated eddy currents,

gradient amplifier nonlinearity, unaccounted gradient delays

and off-resonance, while errors in the phase of the

refocusing pulse B1 field may be introduced through

imperfect referencing of synthesizer phase following off-

center slice excitation. Regardless of the source of deviation,

appropriate coherence conditions are described below for

the two MSE sequences investigated in this work (FSE and

b-SSFP), and the consequences for an appropriate calibra-

tion prescan are described.

2.1. FSE calibration

For MSE sequences where nonzero gradient encoding is

present during the application of refocusing pulses, spatially

invariant phase-cycling conditions cannot exist [20]. For

instance, alignment of transverse magnetization and the

refocusing pulse B1 field, in accordance with CPMG

condition, cannot be achieved for all spatial locations with

FSE. Considering the effects of imaging gradients alone, the

phase of transverse magnetization relative to the refocusing

pulse B1 field immediately prior to the first and second

refocusing pulses is given by /1 and /2, respectively,

as follows:

/1 x; y; zð Þ ¼ cZD=2

0

Gd r dt ¼ c Ax1xþ Ay1yþ Az1z� �

/2 x; y; zð Þ ¼ cZ3D=2

D=2

Gd r dt � /1

¼ c Ax2xþ Ay2yþ Az2z� �

� /1 ð1Þ

where c is the gyromagnetic ratio, D is the refocusing pulse

interval, and G and r are the imaging gradient and spatial

position vectors, respectively. The logical-read, phase-

encode and slice directions are defined by x, y and z,

respectively, and have the indicated gradient–time integrals

Ax, Ay and Az, respectively. In practice, phase encoding is

completely rewound prior to each refocusing pulse; hence,

C.V. Bowen et al. / Magnetic Resonance Imaging 24 (2006) 857–867 859

Ay1 and Ay2 are ideally zero, while the nonzero values of Ax

and Az are determined by the applied read and slice gradient

crusher area, respectively. Therefore, neglecting the phase-

encode axis and substituting from Eq. (1), we define D/ as

the difference in phase between /1 and /2, as follows:

D/u /2 � /1 ¼ c Ax2 � 2Ax1ð Þxþ Az2 � 2Az1ð Þz�½ ð2Þ

Fig. 1 illustrates the phase condition required for the

coherent addition of stimulated echoes and spin echoes for

FSE imaging. The first refocusing pulse separates free

induction decay (FID) magnetization (with phase /1) into

two coherence pathways that contribute to image formation:

one comprising a spin echo and another stored on the

longitudinal axis to be excited by a subsequent refocusing

pulse as a stimulated echo [24]. The number of coherence

pathways that potentially contribute to image formation

expands with each subsequent refocusing pulse in this

Fig. 1. Illustration of the FSE signal phase condition in the transverse plane.

Magnetization vectors are shown immediately before (left side) and after

(right side) the application of the first (A), second (B) and third (C)

refocusing pulses along the yV axis, respectively. Subscripts indicate

whether the magnetization is FID (fid), spin echo (se), stimulated echo (ste)

or stored on the longitudinal axis (z), with the subscript number indicating

the number of refocusing pulse intervals traversed by the respective signal

pools. The angles associated with Mz components indicate the encoded

phase produced by subsequent stimulated echoes, with /1 and /2 defined

for some spatial position by Eq. (1). After three refocusing pulses (C),

assuming that the condition defined by Eq. (3a) is met, all transverse

magnetization components pool into those stimulated echoes that have had

an even number (Me) or an odd number (Mo) of refocusing interval

passages, with phase relative to the yV axis given by /1 and /2,

respectively. The spin-echo signal combines with the Me component.

fashion, with the phase of the spin-echo magnetization given

by /2 immediately prior to the second refocusing pulse.

Two conditions are required for the constructive summa-

tion of signal from all coherence pathways: the gradient–time

integral applied during each refocusing pulse interval must

be identical, and the difference in phase D/ between /1 and

/2 must be zero [6]. The first condition ensures that signals

combine coherently, while the second condition ensures that

signals combine constructively. These conditions are defined

by Eqs. (3a) and (3b), respectively, as follows:

/n ¼ /2 for all n N 2 ð3aÞ

D/ ¼ 0 ð3bÞ

where /n is the phase accrued during the refocusing pulse

interval prior to the nth refocusing pulse and D/ is defined

in Eq. (2). As is evident from Fig. 1C, the satisfaction of Eq.

(3a) causes magnetization from each of the rapidly expand-

ing number of coherence pathways to converge along one

of two orientations, with phase /1 or /2, depending on

whether the stimulated-echo signal has traversed an even

number or an odd number of refocusing pulse intervals,

respectively [25]. These two magnetization pools rapidly

approach with equivalent strength and with comparable

signal intensity after as few as four echoes [20,24,25].

Failure to satisfy Eq. (3a) causes incoherent mixing of

signals from various coherence pathways and apparent rapid

signal decay along the FSE train.

Failure to satisfy Eq. (3b) causes resultant magnetization,

which is derived from a combination of even and odd echo

magnetization pools (Me and Mo in Fig. 1C) to be reduced,

as follows [6]:

M 2 ¼ M 2e þM 2

o þ 2MeMocosD/ ð4Þ

where D/ is defined in Eq. (2), and a spatial cosine mod-

ulation of signal strength is observed if Eq. (3b) is not met.

In the case of unbalanced readout gradients (i.e., Ax2p 2Ax1

within Eq. (2)), a strong cosine modulation in the readout

direction is observed for late effective-echo FSE images

[6] — a property that is central to the prescan calibration

approach involving the read profiles described below.

2.2. b-SSFP calibration

Unlike the case with FSE, b-SSFP is an MSE sequence

that requires a complete rewinding of all gradient encodings

prior to subsequent RF excitation [7,11,26]. One parameter

that determines b-SSFP signal response is the phase

evolution of magnetization on a TR interval /TR. Again,

considering the effects of imaging gradients alone, /TR is

given, as follows:

/TR x; y; zð Þ ¼ cZTR

0

Gd r dt ¼ c Axxþ Ayyþ Azz� �

ð5Þ

where Ax, Ay and Az are gradient–time integrals over a TR

interval in the read, phase-encode and slice directions,

C.V. Bowen et al. / Magnetic Resonance Imaging 24 (2006) 857–867860

respectively. As with FSE, the two conditions required for a

coherent and a constructive addition of multiple coherence

pathways are given by Eqs. (6a) and (6b), respectively, as

follows [7,11]:

/ nð ÞTR ¼ / nþ1ð Þ

TR for all n ð6aÞ

/TR ¼ 0 ð6bÞ

where n is the number of TR interval in the RF pulse train.

Failure to satisfy Eq. (6a) causes incoherent summation

of signals from multiple coherence pathways, with the

extreme limiting case being RF spoiling, where random

phase jumps are deliberately added between RF pulses to

suppress echo formation.

Failure to satisfy Eq. (6b) results in destructive interfer-

ence between multiple signal coherence pathways and

causes periodic variation in signal intensity for changes in

/TR. However, unlike the spatial cosine variation for FSE

imaging, as described in Eq. (4), b-SSFP signal intensity

varies with /TR as a complex function of T1, T2, TR and flip

angle [7,11]. Nonetheless, unbalanced readout gradients

(Axp 0 from Eq. (5)) result in periodic banding profile in the

readout direction; thus, both FSE and b-SSFP are amenable

to the same readout-profile-based prescan calibration tech-

nique described in this work.

3. Methods

All experiments described in this work were performed on

a 4-T Varian/Siemens whole-body scanner equipped with

Siemens Sonata imaging gradients (an amplitude of 40 mT/

m operating at a slew rate of 120 T/m/s). Phantom studies

involved an 11-cm-diameter Ni-doped water sphere and a

27-cm-diameter quadrature birdcage coil. Phantom acquis-

itions had certain imaging parameters for FSE [14 cm field

of view (FOV), 256�256 matrix, 5 mm slice, 15 ms echo

spacing, 8 echoes, 3000/60 ms TR/TE] and b-SSFP (14 cm

FOV, 256�256 matrix, 5 mm slice, 10/5 ms TR/TE). Normal

human volunteers were used for both head and knee imaging

studies. All measurements in the human head used either

a 27-cm-diameter quadrature birdcage coil or an 11-cm-

diameter quadrature surface coil pair. For head imaging,

certain imaging parameters were used for FSE (22 cm FOV,

256�256 matrix, 4 mm slice, 15 ms echo spacing, 8 echoes,

4500/60 ms TR/TE) and b-SSFP (22 cm FOV, 256�256 matrix, 4 mm slice, 8/4 ms TR/TE). An 18-cm quadrature

birdcage coil was used for all high-resolution knee imaging

with certain acquisition parameters for FSE (16 cm FOV,

512�512 matrix, 2 mm slice, 15 ms echo spacing, 8 echoes,

3000/60 ms TR/TE) and b-SSFP (16 cm FOV, 512�512 matrix, 2 mm slice, 13/6.5 ms TR/TE).

3.1. Pulse sequences

Both 2D FSE and 2D/3D b-SSFP sequences were

developed in-house following conventional approaches.

The 2D multislice FSE sequence employed CPMG phase

cycling along the refocusing train and combined slice select/

gradient crusher waveforms for refocusing pulses to achieve

minimum echo-spacing conditions [8,9]. Refocusing pulses

with a 1608 flip angle were employed for SAR consid-

erations at 4 T [12] and to facilitate a more rapid equi-

libration of stimulated and spin echoes along the refocusing

pulse train [25]. RF pulses were optimally designed using

the SLR algorithm [27]. In order to satisfy the phase-balance

condition described in Eq. (3b) for the logical frequency-

encode axis x, the area of the readout prewinder gradient

(Ax1) was adjusted by scaling gradient amplitude according

to a procedure described in the Results section. Eq. (3b) was

satisfied for the slice select axis z by adjusting the phase of

the transmitter for the first refocusing pulse (/1) according

to the calibration procedure developed in the Results

section. The phase is adjusted for Varian console through

specific off-resonance frequency hops for a finite amount of

time (coincident with the readout prewinder gradient lobe)

to achieve target phase offset.

The 2D and 3D b-SSFP sequences were developed in-

house according to an approach described in the literature

for cardiac imaging [26]. This version of the b-SSFP

sequence can be gated to the cardiac cycle or to the

respiratory cycle through the use of a catalyzing train of RF

pulses to rapidly achieve steady-state condition. A catalyz-

ing train of 20 RF pulses was applied prior to data

acquisition, where the flip angle was linearly increased to

the target flip angle a. This catalyzing pulse strategy is

known to be robust with respect to off-resonance for a rapid

approach to steady state [28] — an important consideration

at higher field strengths. The phase-balance condition

described in Eq. (6b) was satisfied for both the frequency-

encode axis (x) and the slice axis (z) in the same manner as

that described above for the FSE sequence.

Several approaches were taken to reduce the phase-

encode ghosting induced by signal oscillation along the

acquisition train. A �a/2 bflipbackQ pulse was applied with

a delay of TR/2 after the last RF pulse in the acquisition

train in order to return magnetization to equilibrium con-

dition along the longitudinal axis. This promoted a more

rapid approach to steady-state condition by subsequent

catalyzing trains [29]. A linear phase-encode pattern was

used in order to prevent signal oscillation from being

introduced by subtle differences in /TR for adjacent TR

intervals. The presence of residual eddy currents causes

differences in /TR that make adjacent TR intervals more

pronounced when the phase-encode increment is large [30].

Additionally, in order to minimize the phase-encode

increment for adjacent TR intervals, each uncollected echo

in the catalyzing train was phase-encoded using the first

phase-encode value in the acquisition train [31].

3.2. Readout profile prescan acquisition

Prior to every FSE or b-SSFP acquisition, an array of

10 readout profiles having a different readout prewinder

C.V. Bowen et al. / Magnetic Resonance Imaging 24 (2006) 857–867 861

gradient area was acquired for each slice (phase-encode

gradients turned off). The readout prewinder gradient area

(Ax1 from Eq. (1) for FSE, or Ax from Eq. (5) for b-SSFP)

was deliberately adjusted from a nominal value by a fraction

fx of the full readout gradient area (acquisition window).

The fx array was uniformly spaced by about 0% deviation

from nominal, where the maximum fx deviation was

selected to produce a number of signal intensity modulation

bands in the readout direction Nb of approximately 30. As

can be deduced from Eq. (2) and (5) for FSE and b-SSFP,

respectively, this corresponds to an fx value having a max-

imum offset fraction of Nb /Nro, where Nro is the number of

frequency-encode points in the acquisition.

The total acquisition time for this prescan was 10 TR

cycles for FSE, while the acquisition of b-SSFP profiles

required approximately 21 TR cycles per slice per fx value

(20 catalyzing pulses to approach steady state plus one

profile acquisition). This corresponds to a prescan acqui-

sition time of approximately 30 s in each case (10 fxprofiles with a TR of 3 s for FSE, or 10 fx profiles with a

TR of 10 ms and 15 slices for b-SSFP). For the most part,

the profiles were acquired with identical acquisition

parameters and slices, as prescribed for imaging. However,

for b-SSFP prescan acquisitions, the flip angle was

Fig. 2. Demonstration of banding in frequency-encode profiles obtained from a pha

(A) and from b-SSFP acquisitions (B) with deliberate imbalance in the read prew

readout gradient area. The 1D Fourier transform magnitude of the same profiles

profile is obtained from the peak location for the first harmonic (indicated by arrow

harmonic for FSE profiles but a peak location similar to that of b-SSFP.

adjusted to 708, independent of the prescribed flip angle,

since magnitude banding profiles in the readout direction

are more sinusoidal for flip angles that exceed the Ernst

angle [23].

Processing of prescan data to extract calibration param-

eters (according to a strategy described in the Results

section) was performed online and automatically, using

compiled routines written in MATLAB (The Mathworks,

Natick MA). These routines are stand-alone (do not require

a MATLAB license or do not require a running MATLAB

interface) and require approximately 4 s of processing to

compute and to set calibration parameters (computation on a

SUN Ultra80 workstation).

4. Results

A calibration procedure for satisfying the required phase

conditions for FSE (Eq. (3b)) and b-SSFP (Eq. (6b)) is first

developed below. This prescan procedure requires the

acquisition of a number of readout profiles for each slice

prior to imaging. The efficacy of this approach for in vivo

head and knee imaging is then demonstrated for oblique

slices in the case of volume and surface coil acquisitions for

both b-SSFP and FSE at 4 T.

ntom. Read profiles are shown from FSE acquisitions using the fourth echo

inder gradient equal to 2.7% (solid line) and 5.3% (dashed line) of the full

is shown for FSE (C) and b-SSFP (D). The banding frequency for a given

s for profiles with fx =2.7%). Note the additional relative height of the first

Fig. 3. Determination of calibration parameters. (A) The banding frequency

determined from read profiles of FSE acquisitions (see Fig. 2) is shown as a

function of fractional read prewinder gradient error ( fx) for two slices

(� and o symbols). The optimum fx value for imaging ( fx(opt)) is

determined from a linear fit extrapolated to zero banding frequency.

(B) The phase of 1D Fourier transform profiles corresponding to the

location of the first harmonic magnitude peak (see Fig. 2C and D) is shown

as a function of fx for the same two slices. The arrow indicates the band

phase (/0) of one slice corresponding to fx(opt), as obtained from linear

extrapolation. (C) Slice calibration parameters are obtained from the slope

( Psl) and the intercept ( P0) of a variance-normalized weighted linear

regression of the band phase (/0) versus slice position (z). Error bars

indicate the variance of the residuals from curves shown in (B).

C.V. Bowen et al. / Magnetic Resonance Imaging 24 (2006) 857–867862

4.1. Readout-profile-based calibration procedure

Fig. 2 shows read profiles obtained from a uniform

spherical phantom for both b-SSFP (Fig. 2A) and FSE

(Fig. 2B) acquisitions, where the readout prewinder lobe has

been deliberately scaled from a nominal value. The prom-

inent feature for magnitude profiles obtained from each

sequence is a banding of the signal intensity in the readout

direction. For FSE acquisitions, profiles obtained from the

fourth echo demonstrate a sinusoidal banding profile in

accordance with predictions from Eq. (4). Additionally, null

signal regions in the profiles (stopband) have nearly 100%

signal loss, indicating nearly equivalent strength between

the even and the odd echo magnetization pools by the fourth

echo (Me and Mo in Fig. 1C). For b-SSFP acquisitions, the

profiles exhibit a more complex banding modulation, with a

flatter passband that occupies more than 50% of each cycle

and with less pronounced stopbands.

However, profiles from both FSE and b-SSFP acquis-

itions show increased banding frequency as the readout

prewinder gradient is deliberately adjusted from a nominal

value by a fraction fx of the full readout gradient area. These

profiles correspond to the satisfaction of Eqs. (3a) and (6a),

but not of Eqs. (3b) and (6b), for FSE and b-SSFP,

respectively. The spatial frequency of modulation is

extracted from the 1D Fourier transform of magnitude

profiles (Fig. 2C and D for FSE and b-SSFP, respectively).

Spatial frequency is determined from the peak location of

the first harmonic in Fourier transform magnitude profiles

(arrows in Fig. 2C and D). Fourier transform profiles have

been premultiplied by a Hanning window to suppress

truncation artifacts and by a Gaussian smoothing function,

since only low spatial frequency banding information is

sought. Despite differences in banding shape for FSE and b-

SSFP profiles, both sequences exhibit similar banding

frequencies (~7 cycles per FOV for an fx of 2.7%). The

identification of the first harmonic peak location from

Fourier transform profiles is slightly more robust for FSE

acquisitions, as this peak exhibits a greater peak height

relative to those from b-SSFP Fourier transform profiles.

Fig. 3A shows the band frequency for FSE acquisitions

(extracted according to the approach described in Fig. 2) as a

function of fx for two representative slices from a phantom.

Data points are only present for fx values (nonzero)

producing a clearly defined band frequency, with the

extracted banding frequency negated for negative fx values.

Excellent linearity and very consistent curves are present for

all slices within a phantom. The optimal fx value needed to

produce no spatial banding in the frequency-encode direc-

tion is obtained from an extrapolation of linear fits for each

slice to zero band frequency. The optimal calibration

parameter for imaging fx(opt) is obtained from the mean of

the optimal fx values derived from each slice, weighted

inversely by the variance of residuals from fits for each slice.

Fig. 3B shows the phase of Fourier transform profiles

at the first harmonic peak location (Fig. 2C) as a function of

fx. The plots are again quite linear, with minimal slopes and

different intercept positions for each slice. Physically, for

FSE, this band phase corresponds to the relative angle

between two signal pools (Me and Mo in Fig. 1C) at the

gradient isocenter (x=0 cm line; Fig. 2A). The band phase is

also the relative angle between Me and Mo that remains after

the readout gradient has been properly balanced through

C.V. Bowen et al. / Magnetic Resonance Imaging 24 (2006) 857–867 863

fractional weighting by fx(opt). The band phase for each slice

at the optimal fx setting (/0) is derived from the linear

extrapolation of each curve to fx(opt). Fig. 3C plots the band

phase as a function of slice position (z). A variance-

normalized linear regression of these data produces the final

two calibration parameters required to correct for slice-

dependent variation in phase angle, as follows:

/0 ¼ P0 þ Pslz ð7Þ

where Psl and P0 are the slope and intercept from Fig. 3C,

respectively. The required correction for slice-dependent

phase variation /cor is applied through phase shifts, as

Fig. 4. Demonstration of robust band frequency extraction using frequency-encode

acquisitions for oblique slices (A), a quadrature surface coil pair (C) and a 3-cm-t

magnitude profiles are shown (B, D and F) with deliberate imbalance in the read pr

gradient area. Note the identical banding frequency for two slices separated by 22

relative to each other (A and C).

described in the Methods section, where /cor=� /0 from

Eq. (7). These three correction factors ( fx(opt), Psl and P0), as

calculated from profile prescan, are employed for subse-

quent acquisition of images, with the resultant image quality

demonstrated in Section 4.2.

4.2. In vivo image quality

The robustness of the readout profile calibration proce-

dure is demonstrated in Fig. 4 under potentially challenging

conditions, such as those involving oblique imaging

(Fig. 4A and B), use of surface coils (Fig. 4C and D) and

3D slab selection (Fig. 4E and F). Both the read profiles

profiles obtained from a human head. Read profiles are shown from b-SSFP

hick 3D slab-selective profile (E). The corresponding 1D Fourier transform

ewinder gradient ( fx) equal to 8% (B) and 13% (D and F) of the full readout

mm (A–D), while the bands have different phases and are spatially shifted

Fig. 5. Comparison of images from the head of a normal volunteer

acquired using optimized calibration parameters (A and C) or nominal

settings (B and D). Images from both T2-weighted FSE (A and B) and

b-SSFP (C and D) illustrate the successful removal of banding for

acquisitions obtained using calibration parameters.

C.V. Bowen et al. / Magnetic Resonance Imaging 24 (2006) 857–867864

(Fig. 4A, C and E) and the corresponding 1D Fourier

transforms (Fig. 4B, D and F) are shown for a representative

slice obtained from the head of a normal human volunteer

using the b-SSFP sequence. As described in the Theory

section and as illustrated in Fig. 2B, b-SSFP signal

modulation in readout profiles does not follow a sinusoidal

behavior, as with FSE profiles, and thus provides the more

challenging application of the readout profile calibration

procedure. In each case, the dominant modulation observed

in the 1D Fourier transform readout profiles (Fig. 4B, D and

F) results from imbalanced read gradients rather than from

any structure within the image or from the sensitivity profile

of the receive coil.

Table 1 shows the mean and the standard deviation of the

three correction factors ( fx(opt), Psl and P0) obtained from five

repeat measures of the profiles shown in Fig. 4. In the case of

oblique imaging, surface coil imaging and 3D imaging, the

standard deviation of parameters is sufficiently small to

ensure that avoidable signal loss resulting from imprecise

calibration is less than 2.7%, 3.5% and 3.7%, respectively,

for any spatial position within the imaging volume. This

estimate assumes one standard deviation in each parameter

and a more pessimistic sinusoidal modulation of signal

intensity, given by Eq. (4), for FSE acquisitions.

Fig. 5 shows oblique 2D FSE and b-SSFP images from

a normal human volunteer obtained with optimized (Fig. 5A

and C) or nominal (Fig. 5B and D) calibration factors

( fx(opt), Psl and P0). Oblique images were acquired parallel

to the anterior commissure–posterior commissure line. The

effective removal of banding for oblique images acquired

using optimized parameters is clearly apparent for both late

effective-echo FSE imaging and b-SSFP imaging.

Fig. 6 shows fat-suppressed fourth effective-echo

FSE images (Fig. 6A and B) and 2D b-SSFP images

(Fig. 6C and D) for two sagittal slices through the knee of a

normal human volunteer. All images were acquired after

the optimization of calibration factors using readout profile

technique. An excellent contrast between synovial fluid and

cartilage is observed for both 2D b-SSFP and T2-weighted

FSE images. The minimization of phase-encode direction

blurring is accomplished by preservation of the signal

intensity along the refocusing pulse trains through satisfac-

tion of the conditions given by (Eqs. (3a), (3b), (6a), and

(6b) for FSE and b-SSFP imaging, respectively. The large

degree of signal heterogeneity within knee images presents

challenges that illustrate the robustness of the readout

profile calibration procedure where identification of spatial

frequencies produced by readout gradient imbalance might

Table 1

Calibration parameters for b-SSFP acquisitions in the human head

fx(opt) (%) P0 (8) Psl (8/cm)

Oblique slice 2.188F0.008a 36.5F2.9 �119.3F0.5

Surface coil 1.842F0.008 28.9F2.4 �128.7F1.0

3D slab 0.022F0.016 �1.5F0.5 N/A

a Uncertainties indicate standard deviation from five repeat measures.

be problematic. Banding patterns far dominate any intrinsic

modulation present even within complex in vivo images, as

demonstrated by the quality of sagittal knee images shown

in Fig. 6.

5. Discussion

Both the importance and the reliability of the proposed

readout profile calibration technique can be assessed from

calibration parameter estimates obtained with acquisitions

from various anatomic locations. The peak fractional signal

loss expected relative to acquisitions using only nominal

settings is an indicator of the importance of implementing

calibration and can be derived from the mean values of

calibration parameter estimates. For the example shown in

Fig. 3A using axial slice FSE acquisitions obtained from a

phantom, fx(opt) is estimated to be �0.042%. Although

apparently small in this ideal case, failure to account for this

deviation results in 0.11 band cycles across the FOV

(Nb=Nrofx) or a maximum signal loss of 22%, assuming

Fig. 6. Comparison of images from the knee of a normal volunteer acquired using T2-weighted FSE (A and B) or b-SSFP (C and D). Matched sagittal slices that

are medial (A and C) and midline (B and D) are shown. Excellent contrast between cartilage and synovial fluid is apparent for all images (white arrow in medial

images), while trabecular bone contrast is particularly striking in b-SSFP. All images were obtained using optimized calibration parameters and show the

absence of banding artifacts.

C.V. Bowen et al. / Magnetic Resonance Imaging 24 (2006) 857–867 865

that sinusoidal modulation was predicted from Eq. (4). A

similar calculation based on the mean parameter estimates

listed in Table 1 for b-SSFP profiles obtained from the head

indicates that one or more bands of signal loss would be

expected across the imaging volume for acquisitions using

nominal settings rather than calibrated settings (see Fig. 5).

In practice, much greater fx(opt) values are encountered in

vivo and for high-resolution acquisitions that employ larger

gradient strengths. In vivo acquisitions, particularly at high-

field strength, present difficult conditions for shimming

that may result in uncorrected first-order B0 variation that

can be compensated for by a proposed calibration technique.

Although improved shim algorithms are helpful [32], MSE

artifacts from problems associated with gradient fidelity,

such as gradient amplifier nonlinearity, unaccounted gradi-

ent delays and imperfect eddy current compensation, are not

uncommon in many systems regardless of field strength

[14–17] and cannot be removed through shimming. These

artifacts are most prevalent for high-resolution acquisitions

since the larger gradient strengths employed magnify

gradient fidelity problems, while the increased RF pulse

spacing required increases the sensitivity of MSE acquis-

itions to banding artifacts. The relatively long RF pulse

refocusing intervals used to produce the images in Figs. 5

and 6 demon strate the robust remova l of bandin g artifacts

under these conditions.

The reliability of the proposed technique can be assessed

from the variance in parameter estimates obtained for a

given anatomic location under varied acquisition conditions.

Table 1 lists the variance of parameter estimates obtained

using b-SSFP profiles measured in the head of volunteers

for oblique, surface coil and 3D acquisitions. In all cases

studied, adequate reproducibility was achieved to ensure that

calibrations limited avoidable signal loss to less than 4% for

any spatial location within the imaging volume. Nonlinear

B0 field variation from imperfect higher order shimming

results in destructive MSE signal coherence that is not

avoidable, since this cannot be compensated as a result

of linearity assumptions intrinsic to the proposed prescan

technique. Equivalent measurements for FSE acquisitions

(not shown) were more reliable than those of b-SSFP and

produced parameter estimates limiting avoidable signal loss

to 1.5% throughout the imaging volume. The superior

reliability of parameter estimation for FSE profiles results

from both the sinusoidal modulation described in Eqs. (4),

(6a) and (6b) and the near 100% signal loss in the readout

C.V. Bowen et al. / Magnetic Resonance Imaging 24 (2006) 857–867866

profile stopbands by the fourth echo [20,24,25]. Together,

these properties produce easily extracted spatial frequency

peaks using Fourier transform (see Fig. 2C and D). The

robustness of parameter estimation for b-SSFP acquisitions

was improved by collecting profiles with flip angles (708)higher than the Ernst angle (258 in gray matter at 4 T). The

use of higher flip angles produces signal modulation that is

more rounded in the passband [7,11] and results in more

easily extracted spatial frequencies.

Surface coil acquisitions present the greatest challenge,

particularly for smaller surface coils, because the signal

intensity modulation resulting from the falloff of the RF

coil sensitivity profile may occur over spatial scales similar

to those of the signal banding induced by the readout profile

prescan technique. This problem could be addressed by

a modification of the range of fx values used in profile

acquisition to produce a higher range of spatial frequencies

that are more separable from spatial frequencies present in

the coil sensitivity modulation along the readout direction.

All 3D profile arrays involved a single, thick, slab-

selective excitation (32 mm for Fig. 4E and F), with

excellent precision of both estimated calibration parameters

( fx(opt) and P0). An intrinsic limitation of 3D profile acqui-

sitions is an inability to correct for more than the DC off-

resonance, rather than any linear field variation present in

the slice direction (as compensated for by Psl for 2D

multislice acquisitions). This could potentially be problem-

atic for very thick slab selections, with 3D acquisitions

having a large number of slices. Indeed, a limitation of

any readout profile technique is an inherent inability to

compensate for effects that are intrinsically 3D and vary

spatially in the phase-encode direction. However, we

employ this prescan calibration technique daily within our

laboratory and have observed robust artifact suppression in

the phase-encode direction, as well as the slice direction for

3D acquisitions. The application of symmetric gradient

waveforms with opposite polarity for both phase encoding

and rewinding tends to cancel (over a refocusing pulse

interval) any undesired residual gradient encoding that

results from imperfect gradient fidelity. Mathematically,

this indicates that phase-encode gradients do not disturb the

conditions for coherent signal formation along the echo

train, as given by Eqs. (3a) and (6a) for FSE and b-SSFP,

respectively. Despite their asymmetry, the gradient wave-

forms along the slice and the readout directions are identical

for each refocusing period and, thus, only two coherent

signal pools may form through imperfect gradient fidelity.

For this reason, the proposed projection prescan technique

has been successfully employed to constructively align the

coherent signal pools for each spatial direction.

6. Conclusions

In this work, we have proposed a readout-projection-

based prescan technique that automatically adjusts read

gradient–slice gradient balance to ensure coherent signal

formation in MSE sequences to first order. This technique

utilizes banding artifacts that form through a deliberate

adjustment of readout gradient balance from a nominal

value in order to calculate calibration parameters for use

in subsequent image acquisition. The robustness of this

approach was demonstrated for FSE and b-SSFP images

obtained from the knees of human volunteers. The cali-

bration technique is very reliable even within complex

in vivo images such as the knee, since banding patterns

far dominate any intrinsic modulation in typical images.

Although, as a projection technique, calibration compen-

sates for gradient imperfections in the slice direction and in

the read direction to first order only, we employ this prescan

technique daily within our laboratory in multiple anatomic

locations and have observed robust behavior for all oblique

imaging, surface coil and 3D scans. An extension of this

prescan calibration approach to diffusion-weighted MSE

sequences [13,14,22], or those requiring precise phase-

cycling patterns for producing T2 quantification maps [33],

might compensate for inherent technical difficulties and

might facilitate their implementation. The correction is

expected to be of particular importance for high-field

strength acquisitions with extended RF pulse spacing in

the development of fMRI techniques with high BOLD

sensitivity and improved suppression of large draining veins

[34]. We believe that application of the proposed prescan

technique will greatly compensate for many of the technical

challenges present at higher fields and will permit the

implementation of many MSE sequences, including the FSE

and b-SSFP sequences demonstrated within this work.

Acknowledgments

This work was supported, in part, by Canadian Institutes

of Health Research grant MOP-64399 and, in part, by a

BRAIN grant from the Ontario R&D Challenge Fund.

References

[1] Chen CN, Sank VJ, Cohen SM, Hoult DI. The field dependence of

NMR imaging: I. Laboratory assessment of signal-to-noise ratio and

power deposition. Magn Reson Med 1986;3:722–9.

[2] Kennan RP, Zhong J, Gore JC. Intravascular susceptibility contrast

mechanisms in tissues. Magn Reson Med 1994;31:9–21.

[3] Ogawa S, Lee TM, Kay AR, Tank DW. Brain magnetic resonance

imaging with contrast dependent on blood oxygenation. Proc Natl

Acad Sci U S A 1990;87:9868–72.

[4] Menon RS, Ogawa S, Kim SG, Ellermann JM, Merkle H, Tank DW,

et al. Functional brain mapping using magnetic resonance imaging.

Signal changes accompanying visual stimulation. Invest Radiol

1992;27(Suppl 2):S47–S53.

[5] Hoult DI, Phil D. Sensitivity and power deposition in a high-field

imaging experiment. J Magn Reson Imaging 2000;12:46–67.

[6] Norris DG, Bornert P, Reese T, Leibfritz D. On the application of

ultra-fast RARE experiments. Magn Reson Med 1992;27:142–64.

[7] Freeman R, Hill HDW. Phase and intensity anomalies in Fourier

transform NMR. J Magn Reson 1971;4:366–83.

[8] Hennig J, Nauerth A, Friedburg H. RARE imaging: a fast imaging

method for clinical MR. Magn Reson Med 1986;3:823–33.

C.V. Bowen et al. / Magnetic Resonance Imaging 24 (2006) 857–867 867

[9] Mulkern RV, Wong ST, Winalski C, Jolesz FA. Contrast manipulation

and artifact assessment of 2D and 3D RARE sequences. Magn Reson

Imaging 1990;8:557–66.

[10] Zur Y, Wood ML, Neuringer LJ. Motion-insensitive, steady-state free

precession imaging. Magn Reson Med 1990;16:444–59.

[11] Scheffler K, Hennig J. Is TrueFISP a gradient-echo or a spin-echo

sequence? Magn Reson Med 2003;49:395–7.

[12] Thomas DL, De Vita E, Roberts S, Turner R, Yousry TA, Ordidge RJ.

High-resolution fast spin echo imaging of the human brain at 4.7 T:

implementation and sequence characteristics. Magn Reson Med

2004;51:1254–64.

[13] Alsop DC. Phase insensitive preparation of single-shot RARE:

application to diffusion imaging in humans. Magn Reson Med 1997;

38:527–33.

[14] Seifert MH, Jakob PM, Jellus V, Haase A, Hillenbrand C. High-

resolution diffusion imaging using a radial turbo-spin-echo sequence:

implementation, eddy current compensation, and self-navigation.

J Magn Reson 2000;144:243–54.

[15] Peters DC, Derbyshire JA, McVeigh ER. Centering the projection

reconstruction trajectory: reducing gradient delay errors. Magn Reson

Med 2003;50:1–6.

[16] Vasilic B, Song HK, Wehrli FW. Coherence-induced artifacts in large-

flip-angle steady-state spin-echo imaging. Magn Reson Med 2004;

52:346–53.

[17] Bieri O, Markl M, Scheffler K. Analysis and compensation of eddy

currents in balanced SSFP. Magn Reson Med 2005;54:129–37.

[18] Reeder SB, Atalar E, Faranesh AZ, McVeigh ER. Referenceless

interleaved echo-planar imaging. Magn Reson Med 1999;41:87–94.

[19] Beaulieu CF, Zhou X, Cofer GP, Johnson GA. Diffusion-weighted

MR microscopy with fast spin-echo. Magn Reson Med 1993;30:

201–6.

[20] Poon CS, Henkelman RM. Practical T2 quantitation for clinical

applications. J Magn Reson Imaging 1992;2:541–53.

[21] Le Roux P. Non-CPMG fast spin echo with full signal. J Magn Reson

2002;155:278–92.

[22] Bastin ME, Le Roux P. On the application of a non-CPMG single-shot

fast spin-echo sequence to diffusion tensor MRI of the human brain.

Magn Reson Med 2002;48:6–14.

[23] Bangerter NK, Hargreaves BA, Vasanawala SS, Pauly JM, Gold GE,

Nishimura DG. Analysis of multiple-acquisition SSFP. Magn Reson

Med 2004;51:1038–47.

[24] Crawley AP, Henkelman RM. Errors in T2 estimation using multislice

multiple-echo imaging. Magn Reson Med 1987;4:34–47.

[25] Wan X, Parker DL, Lee JN, Buswell HR, Gullberg GT. Reduction of

phase error ghosting artifacts in thin slice fast spin-echo imaging.

Magn Reson Med 1995;34:632–8.

[26] Deshpande VS, Shea SM, Laub G, Simonetti OP, Finn JP, Li D. 3D

magnetization-prepared true-FISP: a new technique for imaging

coronary arteries. Magn Reson Med 2001;46:494–502.

[27] Pauly J, Le Roux P, Nishimura DG, Macovski A. Parameter relations

for the Shinnar–LeRoux selective excitation pulse design algorithm.

IEEE Trans Med Imaging 1991;10:53–65.

[28] Deshpande VS, Chung YC, Zhang Q, Shea SM, Li D. Reduction of

transient signal oscillations in true-FISP using a linear flip angle series

magnetization preparation. Magn Reson Med 2003;49:151–7.

[29] Zwanenburg J, Kuijer J, Marcus J, Heethaar R. Startup method for

magnetization-prepared SSFP cine imaging. ISMRM Proc 2003:978.

[30] Foxall DL. Frequency-modulated steady-state free precession imag-

ing. Magn Reson Med 2002;48:502–8.

[31] Scheffler K, Hennig J. Eddy current optimized phase encoding

schemes to reduce artifacts in balanced SSFP imaging. ISMRM Proc

2003:294.

[32] Klassen LM, Menon RS. Robust automated shimming technique

using arbitrary mapping acquisition parameters (RASTAMAP). Magn

Reson Med 2004;51:881–7.

[33] Foltz WD, Stainsby JA, Wright GA. T2 accuracy on a whole-body

imager. Magn Reson Med 1997;38:759–68.

[34] Constable RT, Kennan RP, Puce A, McCarthy G, Gore JC. Functional

NMR imaging using fast spin echo at 1.5 T. Magn Reson Med 1994;

31:686–90.