Reducing susceptibility artifacts in fMRI using volume-selective z-shim compensation.

Abstract

Susceptibility-induced magnetic field gradients (SFGs) can result in severe signal loss in the orbitofrontal cortex (OFC) in gradient-echo-based functional MRI (fMRI) studies. Although conventional z-shim techniques can effectively recover the MRI signal in this region, the substantial penalty in imaging time hampers their use in routine fMRI studies. A modified z-shim technique with high imaging efficiency is presented in this study. In this technique, z-shim compensations are applied only to a selective volume where the susceptibility artifact is severe. The results of an fMRI study (N=6) demonstrate the feasibility of detecting the OFC activation with z-shim in whole-brain fMRI studies at a temporal resolution of 2 s.

Full text

Page 1

Reducing Susceptibility Artifacts in fMRI Using Volume-
Selective z-Shim Compensation
Yiping P. Du,1*Manish Dalwani,1Korey Wylie,1Eric Claus,2and Jason R. Tregellas1
Susceptibility-induced magnetic field gradients (SFGs) can re-
sult in severe signal loss in the orbitofrontal cortex (OFC) in
gradient-echo-based functional MRI (fMRI) studies. Although
conventional z-shim techniques can effectively recover the MRI
signal in this region, the substantial penalty in imaging time
hampers their use in routine fMRI studies. A modified z-shim
technique with high imaging efficiency is presented in this
study. In this technique, z-shim compensations are applied only
to a selective volume where the susceptibility artifact is severe.
The results of an fMRI study (N ? 6) demonstrate the feasibility
of detecting the OFC activation with z-shim in whole-brain fMRI
studies at a temporal resolution of 2 s.
396–404, 2007. © 2007 Wiley-Liss, Inc.
Key words: fMRI; echo-planar imaging; susceptibility artifact;
z-shim compensation; orbitofrontal cortex
Magn Reson Med 57:
In fMRI experiments signal loss can occur in regions of
brain adjacent to air cavities where susceptibility-induced
magnetic field gradients (SFGs) perpendicular to the slice
plane are the primary source of intravoxel dephasing. This
signal loss often precludes the acquisition of fMRI data in
regions that are of considerable interest for functional
studies, including the orbitofrontal cortex (OFC) and infe-
rior temporal regions (1,2). In addition, this susceptibility
artifact becomes more severe at high magnetic field
strengths (3T or higher). Tailored RF pulses (3), spin-echo-
based fMRI (4), spiral-in/out (5–7), and passive shim (8–
10) methods have been used to reduce the signal loss in
regions compromised by SFGs.
A technique known as “z-shim” was developed to effec-
tively compensate for this susceptibility artifact in fMRI
(1,11–13). In the z-shim technique, additional image vol-
umes are acquired with a compensation gradient applied
to the slice direction. These additional image volumes are
then combined with the original image volume acquired
without z-shim to form a composite 3D fMRI data set (14).
Previous studies demonstrated the effectiveness of the z-
shim technique in recovering fMRI signal in the OFC and
inferior temporal regions (15–18). However, the need to
acquire multiple volumes prolongs the acquisition time for
each time frame, or effective repetition time (eTR). Several
variations of the z-shim technique with improved imaging
efficiency have been developed recently. These include a
3D z-shim method with extended kzcoverage (19), a sin-
gle-shot double-echo echo-planar imaging (EPI) scan with
z-shim applied to one of the echoes (20), and a single-shot
z-shim method with hybrid gradient-spin-echo acquisition
(21).
In whole-brain fMRI experiments the spatial extension
of the regions with severe susceptibility artifact is typi-
cally in the range of 20 mm along the superior/inferior
(S/I) direction (19,22,23), which is a small fraction of the
size of the brain. The benefit of applying z-shim to other
regions with mild field inhomogeneity is usually out-
weighed by the penalty in imaging time. In this study we
developed a fast scheme for the z-shim technique in which
the z-shim compensation is applied only to the selective
volume where severe susceptibility-induced signal loss
occurs (Fig. 1) (24). A scheme for reordering of data acqui-
sition in compensated and noncompensated slice loca-
tions is presented to ensure a steady state in the compen-
sated slices. We conducted fMRI experiments with a re-
ward-punishment task to demonstrate the feasibility of
this technique for detecting fMRI signal in the OFC.
THEORY
The presence of SFGs in the slice selection, Gs, introduces
a signal weighting function:
wt?t,Gs? ?
1
?z?
??z/2
?z/2
e?i?G,ztdz ? sinc?
?
2Gs?zt?, [1]
where t is the time starting from the isocenter of the exci-
tation RF pulse, ?z is the slice thickness, and ? is the
gyromagnetic ratio. The first null point of this sinc func-
tion in an image acquired with an echo time (TE) is at:
?Gs?zTE ? 2?. The corresponding gradient area, Anull, is
then
Anull? GsTE ? 2?/??z. [2]
Note that Anullis solely dependent on the slice thick-
ness. Anullis equal to 5.87 ms ? mT/m at a slice thickness
of 4 mm with ?/2? ? 42.58 MHz/T. The amplitude of Gsin
the OFC is on the order of 0.1 mT/m at 1.5T (15) and 0.2
mT/m at 3T. Such Gsvalues are sufficient to cause nearly
complete local signal loss at a TE of 30 ms.
In an EPI acquisition the spatial frequency in the phase-
encodingdirection isrelated
? TE?/Tacq?Y, where Tacqis the length of the echo train,
396
tot byky
?
?t
1Department of Psychiatry, University of Colorado at Denver and Health
Sciences Center, Aurora, Colorado, USA.
2Department of Psychology, University of Colorado at Boulder, Boulder,
Colorado, USA.
Grant sponsor: Office of National Drug Control Policy (ONDCP), Counterdrug
Technology Assessment Center. Grant sponsor: National Institutes of Health;
Grant numbers: 5 P50 MH068582; 5 R01 MH38321; R01 MH070037; 5 R37
DA09842; VISN19/MIRECC.
*Correspondence to: Yiping P. Du, Ph.D., Brain Imaging Center, University of
Colorado at Denver and Health Sciences Center, Building 400, 12469 East
17th Place, Aurora, CO 80010-7155. E-mail: Yiping.Du@UCHSC.edu
Received 2 August 2006; revised 12 October 2006; accepted 31 October
2006.
DOI 10.1002/mrm.21150
Published online in Wiley InterScience (www.interscience.wiley.com).
Magnetic Resonance in Medicine 57:396–404 (2007)
© 2007 Wiley-Liss, Inc.

Page 2

and ?Y is the pixel dimension. The weighting function in
kycan then be written as:
wk?ky,Gs? ? sinc?
?
2Gs?z?Tacq?Yky? TE??, [3]
which is the Fourier transform of a rectangular function,
rect(?y), with a width of ?y:
?y ?
?
2?Gs?zTacq?Y. [4]
Multiplying the rectangular data acquisition window
with wk?ky,Gs? in the ky-direction is equivalent to convolv-
ing the image with a rectangular function, rect(?y), and
results in blurring along the phase-encoding direction (25).
Because the image intensity is primarily determined by the
signal at ky? 0, the weighting function in the image
domain is given by:
wi?TE,Gs? ? sinc?
?
2Gs?zTE?. [5]
When a z compensation gradient area, Acomp, is added
prior to data acquisition, the weighting function becomes
(see Fig. 2):
wk?ky,Gs,Acomp?
? sinc?
?
2?z?GsTacq?Yky? GsTE ? Acomp??. [6]
Note that wk?ky,Gs,Acomp? is the same function as
wk?ky,Gs?, except for a shift of the center to ky ? ?Acomp
? GsTE?/GsTacq?Y. The degree of blurring caused by the
sinc weighting function remains the same even with this
compensation. With the compensation, the weighting
function in the image domain becomes:
wi?Gs,Acomp? ? sinc?
?
2?z?GsTE ? Acomp??. [7]
Ideally, Acomp ? GsTE is the optimum selection of the
z-shim compensation because the central echo (ky? 0)
will be completely compensated. It should be noted that
even with the optimum compensation, the echoes prior to
the central echo are overcompensated and the echoes after
the central echo are undercompensated (see Fig. 2). In
practice, the z-shim compensation is suboptimal due to
the spatial variation of Gs. For a specific compensation
with Acomp, the regions with Gs? Acomp/TE are overcom-
pensated and the regions with Gs? Acomp/TE are under-
compensated. Furthermore, the exact range of Gsis usually
unknown during the scan, partly because of the differ-
ences among individual subjects and positioning of the
subjects (26). A test run would help to empirically deter-
mine the compensation factor: fc? Acomp/Anull.
There are several ways to form a composite image using
the compensated images with the original noncompen-
sated image (14). Using the sum of squares (SSQ) ap-
proach, the intensity in the composite image is given by:
w?TE,Gs?
? ME?sinc2?
?
2?zGsTE???
k?1
M
sinc2?
?
2?z?GsTE ? Acomp,k??,
[8]
where MEis the steady-state magnetization with the flip
angle selected at the Ernst angle, M is the number of
compensation steps, and Acomp,kis the kth z-shim com-
pensation gradient area, which is usually selected to have
an equal step size: Acomp,k? kAcomp. The weighting func-
tions as a function of fcat M ? 1 and 2 are plotted in Fig.
3. As expected, the range of compensation increases, and
the height of the weighting function decreases, with a
FIG. 1. A volume-selective z-shim scheme shows the z-shim com-
pensation applied to a selected volume labeled as Region-C. The
rest of the imaging volumes, labeled as Region-NC, are not com-
pensated. In the example shown in this figure, five of the 29 slice
locations are z-shim compensated. The z-shim compensated slices,
from the 7th slice to the 11th slice (marked as thick lines), covers a
region immediately above the OFC.
FIG. 2. In the z-shim EPI pulse sequence, a compensation gradient
with an area Acompis applied on the slice-selection gradient after
slice selection and before the echo train (top panel). The lower panel
shows the weighting functions corresponding to no compensation,
undercompensation, optimal compensation, and overcompensa-
tion.
Volume-Selective z-Shim in fMRI397

Page 3

larger fcor M value. In an fMRI study we may specify a
range of SFG, (0, Gs,max), in which z-shim compensation is
expected to be optimal. Gs,maxcan be empirically deter-
mined by prior knowledge of Gsat the cortical region of
interest (ROI). One can then determine the number of
compensation steps, M, after selecting the compensation
factor fcthrough a testing scan or from the plots in Fig. 3:
M ? Gs,maxTE/?fcAnull?. [9]
MATERIALS AND METHODS
Volume Selective z-Shim
In the present technique, z-shim compensation is applied
to the regions with severe SFGs, referred to as Region-C.
The rest of the regions, referred to as Region-NC, are sam-
pled only once per repetition time (TR) (see Fig. 1). The
eTR is less than TR in Region-C and equal to TR in Region-
NC. This difference of eTR results in lower signal intensity
in Region-C than in Region-NC due to less complete T1
relaxation in Region-C.
In the rest of this paper we refer to the images acquired
without z-shim compensation in Region-C as NZScimages.
The images acquired with z-shim compensation are re-
ferred to as ZS images. The number of slices in Region-C is
referred to as Nc, and the number of slices in Region-NC is
Nnc. The total number of slices is referred to as Ntot(? Nc
? Nnc). Because the image acquisition in Region-C occurs
multiple times in a TR, the eTR between the NZScand ZS
acquisitions in Region-C may vary, resulting in different
degrees of partial saturation in these images. In extreme
situations, some of the NZScand ZS images may have
substantially reduced signal due to severe partial satura-
tion. In this study we reordered the data acquisitions to
ensure the steady state (or constant eTR) of the NZSc and
ZS acquisitions, and thus reduced the signal variations
between these images.
Slice Reordering
To ensure the steady state of the z-shim slices, the NZSc
and ZS images should be acquired with an eTR of, or as
close as possible to, TR/(M ? 1). Several possible imple-
mentations can be used to achieve the steady state for the
z-shim slices. In this study we 1) interleaved the data
acquisition order of the NZScand ZS images in Region-C,
2) interleaved the data acquisition order of the rest of the
slices in Region-NC, and 3) inserted the interleaved data
acquisition order in step 1 into the acquisition order of the
rest of the slices (M ? 1) times with equal time interval, at
time tmdetermined by:
tm? ?m ? 1?Nnc/?M ? 1? ? mNc? 1, [9]
where m ? 0, 1,. . ., M, and Nncis a multiple of (M ? 1). A
different amount of z-shim compensation was applied at
m ? 1, . . ., M.
The following examples show the reordering of slice
acquisition with z-shim. We can acquire up to 34 slices in
a TR of 2 s with our fMRI protocol without z-shim (ma-
trix ? 64 ? 64, TE ? 26 ms, with ramp sampling). If we
select one z-shim compensation (M ? 1) in five slice loca-
tions (Nc? 5, starting at the seventh slice), the total num-
ber of slice locations is reduced to 29 (Ntot? 29). The
z-shim slices are acquired at time points 13–17 and 30–34
in a TR, respectively. The eTR of the compensated slices is
1 s. The acquisition order is shown in the second row in
Table 1. If we select two z-shim compensations (M ? 2) in
four slice locations (Nc? 4, starting at the seventh slice),
FIG. 3. The steady-state magnetization Mss(relative to M0) of the
composite image is plotted as a function of GsTE at fc? 0.5, 0.6,
0.7, 0.8, 0.9, and 1.0 when M ? 1 (upper plot) and M ? 2 (lower plot)
at TR ? 2 s. The sinc function centered at GsTE ? 0 is also plotted
as a reference. The Ernst angle was used as the flip angle for
excitation in the simulation. [Color figure can be viewed in the online
issue, which is available at www.interscience.wiley.com.]
Table 1
The Order of Slice Locations in Data Acquisition in two Examples: (M ? 1, NA? 5, Mtot? 29) and (M ? 2, NA? 4, Mtot? 25)*
1
1
1
2
3
3
3
5
5
45678910
25
11
27
10
12
29
19
13
7
21
14
9
23
15
11
25
16
8
17
10
18
2
19
4
20
6
21
12
22
14
10
23
16
12
24
18
14
25
20
16
26
22
18
27
24
20
28
26
22
29
28
24
30
7
7
31
9
9
32
11
33
8
10
34
1013
11
15
13
17
15
19
17
21 23
7982467988
*Top row: the time order of data acquisition. Middle row: the order of slice locations in z-shim data acquisition with M ? 1, NA? 5, Mtot?
29. Z-shim compensation is applied to the 7th–11th slices. The interleaved order of the five z-shim slices (i.e., 7, 9, 11, 8, 10) are inserted
at time points t0? 13 and t1? 30. The effective TR of the z-shim data acquisition is TR/2. Bottom row: the order of slice location in z-shim
data acquisition with M ? 2, NA? 4, Mtot? 25. Z-shim compensation is applied to the 7th–10th slices. The interleaved order of the 4 z-shim
slices (i.e., 7, 9, 8, 10) are inserted at time points t0? 8, t1? 19, and t2? 30. The effective TR of the z-shim data acquisition is TR/3.
398Du et al.

Page 4

the total number of slice locations is then reduced to 25
(Ntot? 25). The total number of acquisitions is reduced to
33 in order to satisfy the condition that Nncis a multiple of
(M ? 1). The z-shim slices are acquired at time points
8–11, 19–22, and 30–33 in a TR, respectively. The eTR of
the compensated slices is 0.666 s. The acquisition order is
shown in the third row in Table 1.
Partial Signal Saturation in the z-Shim Slices
Using the Ernst angle of gray matter (GM) maximizes the
signal intensity in cortical areas in fMRI studies. The Ernst
angle is determined by eTR and T1:
?E? cos?1?e?eTR/T1?, [10]
The corresponding steady-state magnetization is:
ME? M0??1 ? e?eTR/T1?/?1 ? e?eTR/T1?, [11]
where M0is the fully relaxed magnetization. Note that
eTR ? TR/(M ? 1) in Region-C, and eTR ? TR in Region-
NC. A different Ernst angle should be applied to each of
the regions in order to maximize the MR signal. In our
experiment a flip angle of ?E,ncis prescribed on the scan-
ner and used in the acquisition of the images in Region-
NC. The flip angle ?E,cis calculated internally by the pulse
sequence software for the acquisition of the images in
Region-C.
The Ernst angle for GM at 3T (T1? 1330 ms) is 52.7° at
an eTR of 666 ms, 61.9° at 1000 ms, and 77.2° at 2000 ms.
The corresponding steady-state magnetization MEis equal
to 0.50 M0, 0.60 M0, and 0.80 M0, respectively, at these
angles. The intensity discontinuity at the boundaries of
these regions can cause errors in subsequent motion cor-
rection. A multiplication factor, ME,nc/?E,c, is applied to
the composite images to reduce the signal discontinuity at
the boundaries between Region-C and Region-NC, and
hence reduce the possible local errors that could arise
during motion correction. Table 2 shows the dependency
of the signal intensity and the Ernst angle on TR at differ-
ent M values. Note that the relative SNR reduction in a
compensated image is greater at larger M and shorter TR.
This suggests that it is preferable to use a small M (e.g.,
M ? 1) when the TR is relatively short.
fMRI Paradigm
In our fMRI experiment we used a reward-punishment
task similar to the one used by Cusack et al. (27) in their
passive shim study. Prior to the fMRI scans the subjects
were instructed to press a button as fast as possible when
colored squares were presented centrally for 1500 ms. In
the task, green squares indicated “reward” blocks in which
fast responses resulted in a monetary reward, which was
displayed for 1000 ms, 500 ms after the trial, along with a
happy face and an encouraging tone. Responses slower
than the threshold resulted in a sad face but no penalty.
The maximum reward amount was $10, in $0.50 incre-
ments. Red squares indicated the “punishment” blocks, in
which slow responses resulted in a monetary penalty
(–$0.50) accompanied by a sad face and harsh tone. During
all conditions a reward bar was presented at the side of the
screen, which represented cumulative rewards with a
maximum worth of $10. Unknown to the subjects, the
reaction time threshold was adaptively set such that 67%
of reward trials were rewarded, and 67% of punish trials
were punished. The experiment started with 14 s of fixa-
tion, followed by six pseudo-randomized 36-s blocks in
each of reward, punishment, and fixation. Prior to the
fMRI scan the subjects participated in practice trials to
familiarize themselves with the task during a T1-weighted
3D anatomical scan. The average response time for each
subject was recorded near the end of the practice trial, and
was used to set the task for the initial 67% correct rate at
the beginning of the task. The task was repeated in a
second run with a different pseudo-random order of re-
ward, punishment, and fixation. Each run lasted around
11 min.
fMRI Data Acquisition
Six normal subjects (age range ? 24–45 years, mean age ?
30.1 years) participated in the fMRI experiment under a
protocol approved by the UCDHSC institutional review
board. Written informed consent was obtained from all
subjects. The experiment was conducted on a 3T whole-
body scanner (General Electric Healthcare, Milwaukee,
WI, USA). In the experiment, high-order shimming (28)
was applied. A trial scan of a whole-brain EPI was ac-
quired, followed by a 10-min acquisition of a high-resolu-
tion anatomical scan using a 3D inversion-recovery (IR)
spoiled gradient-echo (SPGR) sequence. During the ana-
Table 2
The Ernst Angles (?E, in Degrees) and the Corresponding Steady-state Magnetization ME(Relative to M0) at Different TRs (in Seconds)
and the Number of z-shim Compensations, M
TR
M
0123
?E
71.1
77.2
81.2
84.0
85.9
87.2
ME
0.71
0.80
0.86
0.90
0.93
0.95
?E
55.3
61.9
67.0
71.1
74.4
77.2
ME
0.52
0.60
0.66
0.71
0.76
0.80
?E
46.6
52.7
57.7
61.9
65.4
68.5
ME
0.43
0.50
0.55
0.60
0.64
0.68
?E
41.0
46.6
51.3
55.3
58.8
61.9
ME
0.37
0.43
0.48
0.52
0.56
0.60
1.5
2.0
2.5
3.0
3.5
4.0
Volume-Selective z-Shim in fMRI399

Page 5

tomical scan we visually examined the echo-planar images
from the trial scan to empirically determine the number
and location of the z-shim slices in which the OFC showed
intermediate or severe SFG signal loss. We noticed that
five slice locations were typically sufficient to cover the
region affected by the susceptibility artifacts. No notice-
able susceptibility artifacts caused by foreign magnetic
materials, such as dental fillings, were observed in these
subjects.
A volume-selective z-shim pulse sequence based on a
gradient-echo EPI sequence was used for fMRI data acqui-
sition. The imaging parameters were as follows: matrix ?
64 ? 64, TR/TE/flip angle ? 2 s/26 ms/77°, slice thick-
ness ? 4 mm with no gap, field of view (FOV) ? 22 cm.
MRI data were acquired on the flat-top and both ramps of
the readout gradients to minimize echo spacing (0.52 ms).
One z-shim compensation was applied to five of 29 axial
slices (i.e., M ? 1, Nc? 5, Nnc? 24, Ntot? 29). The eTR
of the NZScand ZS images was 1 s. A flip angle of 62° (i.e.,
the Ernst angle of GM at eTR ? 1 s) was applied to the
NZScand ZS images. A compensation gradient with an
Acompof 4.11 ms ? mT/m (? 0.70 Anull, a trapezoid with an
amplitude of 13.35 mT/m and a base of 0.608 ms) was
used. A total of 34 (? Nnc? MNc) images were acquired in
a TR of 2 s. The fMRI scan consisted of two runs, with 331
repetitions in each run. Additionally, one IR echo-planar
image (TI ? 505 ms) volume was acquired from each
subject to improve coregistration between the echo-planar
and IR-SPGR images for the fMRI data analysis.
fMRI Data Analysis
We combined the NZScand ZS images pixel by pixel using
SSQ to form the composite images prior to other image
preprocessing procedures. The intensity in the composite
images was multiplied by a factor of 1.33 to reduce signal
discontinuity between Region-C and Region-NC.
Functional data were analyzed with the use of SPM2
(Wellcome Department of Imaging Neuroscience, London,
UK). Data from each subject were realigned to the first
volume, normalized to the Montreal Neurological Institute
template, using a GM-segmented IR-EPI as an intermediate
to improve registration between the echo-planar and IR-
SPGR images, and smoothed with an 8-mm full width at
half maximum (FWHM) Gaussian kernel. The blood oxy-
gen level-dependent (BOLD) response was modeled with
the canonical hemodynamic function convolved with the
stimulus function, using the general linear model in
SPM2. A 128-s high-pass filter was applied to remove
low-frequency fluctuation in the BOLD signal.
To account for both within-group and within-subject
variance, a random effects analysis was implemented for
the group analysis. Parameter estimates for each individ-
ual’s first level analysis (SPM contrast images) were en-
tered into second-level t-tests for the contrast of interest,
reward-punishment. The mean response for all voxels
within an anatomically defined OFC ROI (29) was calcu-
lated using the Marsbar Toolbox in SPM (30).
RESULTS
We conducted a test scan on one of the subjects using the
same parameters as in the fMRI experiment to determine
the optimum fcfor the fMRI scans. Different amounts of
z-shim compensation were applied to the five slice loca-
tions above the inferior boundary of the OFC with fc? 0.5,
0.6, 0.7, 0.8, and 1.0, respectively. In the images without
z-shim we observed severe signal loss in the OFC at slice
locations immediately above the inferior boundary of the
OFC (slices 1–4 in Fig. 4) and intermediate signal loss at
superior slice locations (slice 5 in Fig. 4). In slice 1 we
observed that the effect of z-shim compensation was mod-
est and the compensation was slightly more effective with
fc? 0.5. In slice 5 the signal drop-off was less severe than
in the other slices, and the z-shim compensation was more
effective at smaller fcvalues. In slices 2–4, effective z-shim
compensation was seen at fc? 0.6–0.8. We also observed
that the region in which MR signal was effectively recov-
ered became more focal at fc? 1.0, resulting in undercom-
pensation in the neighboring region (such as the region
indicated by an arrow in slice 3).
Figure 5 shows the signal intensity profile along a line in
the OFC, as marked by a dotted line in slice 3 in Fig. 4. In
this slice, fc? 0.5 yielded a larger area of compensation
but lower recovered signal. On the other hand, fc? 1.0
yielded a higher maximal signal recovery but in a much
smaller spatial extension. An adequate balance between
the maximum signal recovery and the spatial extension of
the signal recovery appeared to be at fc? 0.6–0.8. Based
on these observations, we used fc? 0.7 in our fMRI exper-
iment.
The effect of z-shim compensation in the fMRI experi-
ment with six subjects is shown in Fig. 6. A comparison of
activation with and without z-shim in five slice locations
in one of the subjects is shown in Fig. 6a. The t-score maps
with z-shim were superimposed on the composite images
(upper panel), and the t-score maps without z-shim, ob-
tained using the NZScdata, were superimposed on the
NZScimages (lower panel). Severe signal loss is observed
in the OFC in the NZScimages. The signal loss was effec-
tively recovered in the images acquired with z-shim, re-
sulting in a substantial increase in observed task-related
activation. In at least four of the lower slices the detected
activations are located in regions with severe signal loss in
the NZScimages. The same comparison with five other
subjects in a selected slice location is shown in Fig. 6b. A
relatively low t-score threshold, 25% of the maximum
t-score in the five z-shim slices in each subject, was se-
lected in the data analysis to demonstrate the effect of
z-shim in improving the t-score in the OFC in each indi-
vidual subject. Substantial MRI signal recovery and a con-
siderable increase in observed activation are obtained in
the OFC in the images acquired with z-shim in all the
subjects. Although this study focuses on activation of the
OFC, and the statistical power is relatively low in each
individual data set, we observed in five of the six subjects
an inferior temporoparietal and fusiform gyrus activation
that is reasonable in the context of the task performed.
Inferior temporal activation is consistent with increased
attention to the visual properties of the stimulus, i.e., it is
possible that the “reward” condition elicited enhanced
attention to the colored squares. Also, the asymmetrical
hemispheric occipitotemporal and temperoparietal activa-
tions observed in our study are inconsistent with previ-
ously reported results from visual presentation tasks (31).
400Du et al.

Page 6

A group analysis of the six subjects was conducted to
detect the activation of reward-punishment in the OFC,
and the results are shown in Fig. 7. With z-shim, a signif-
icant activation of the OFC was observed in the reward
compared to the punishment condition (t ? 2.23, df ? 5,
P ? 0.042). For the same contrast without z-shim, no
activation above threshold was observed (t ? 1.66, df ? 5,
P ? 0.08). Figure 7 shows significant activation (t ? 3.91,
df ? 5, P ? 0.011) in the OFC for the reward vs. punish-
ment conditions with z-shim compared to the condition
without z-shim (task ? z-shim condition interaction). The
activation map was thresholded at P ? 0.01 for visualiza-
tion, overlaid on the T1-weighted anatomical image of a
single subject.
DISCUSSION
The present technique provides effective z-shim compen-
sation with a relatively small penalty in imaging time.
With M compensation steps, the present technique in-
creases the temporal resolution from TR to (1 ? MNnc/
Ntot)TR, whereas the conventional z-shim techniques
would increase temporal resolution from TR to (1 ? M)TR.
In the z-shim protocol used in our experiment, the penalty
in temporal resolution is only 17.2% (i.e., 34 vs. 29) for
M ? 1. This technique may be particularly useful for
detecting BOLD activations in the OFC or amygdala-hip-
pocampal region in whole-brain event-related fMRI stud-
ies, which require short TRs.
In this study we introduced a parameter, Anull, which is
independent of Gs, TE, and even B0, as a reference for
determining the step size of the compensations, Acomp,
and the number of the compensations, M. When Acomp?
Anull, there is a 10% reduction of signal at brain regions
FIG. 4. The frontal portion of the echo-planar im-
ages acquired with different fcvalues in the five
z-shim slices. In each small panel the image ac-
quired without z-shim compensation is shown at
the top, the image acquired with z-shim compen-
sation is shown in the middle, and the composite
image is shown at the bottom.
FIG. 5. Signal intensity profiles of the z-shim images acquired with
different fcvalues in the third z-shim slice at a location marked by a
dotted line across the OFC in Fig. 4. The signal intensity profile of
the image acquired without z-shim compensation (dotted line in
brown) is also shown for comparison. [Color figure can be viewed in
the online issue, which is available at www.interscience.wiley.com.]
Volume-Selective z-Shim in fMRI401

Page 7

where Gsis near Anull/2TE (see Fig. 3 with M ? 1 and fc?
1.0). This signal reduction disappears when Acomp? 0.9
Anull. It should be noted that only the z component of SFGs
is considered in the calculation of Anull. The SFG compo-
nent in the readout direction can cause local image distor-
tion and hence signal enhancement or reduction. The SFG
component in the phase-encoding direction can cause lo-
cal image “shear,” which also introduces signal variation.
In this study we selected Acomp? 0.7 Anullbased on the
results of a test scan. This test run could be automated in
a regional shim procedure in a prescan, such as the previ-
ously described technique that employs automated brain
segmentation prior to B0mapping (32). Such an automated
procedure would allow the optimum selection of fcin each
individual z-shim slice, as well as the selection of the
slices for z-shim compensation.
Because the z-shim technique can only compensate for
the linear component of the susceptibility-induced B0in-
homogeneity, any uncompensated higher-order B0inho-
mogeneity could cause noticeable residual artifacts. These
residual artifacts can be severe in the inferior region of the
OFC, although the uncompensated in-plane SFGs will
likely be the dominant source of artifacts in that region.
The range of the compensation is related to TE and B0.
Because SFGs are proportional to B0, one must increase the
range of compensation to obtain a similar degree of com-
pensation at higher field strengths. Also, the range of com-
pensation has to be increased at longer TEs. The number of
FIG. 6. a: Comparison of activation with and without z-shim in five slice locations in one subject. The t-score map with z-shim is
superimposed on the composite images (upper row), and the t-score map without z-shim is superimposed on the echo-planar images
(lower row). b: The same comparison with five other subjects in a selected slice location. The signal loss was effectively recovered and the
t-score in the OFC is substantially increased with z-shim in all subjects.
FIG. 7. A t-score map obtained from a group anal-
ysis of six subjects shows significant activation in
the OFC in reward vs. punishment conditions, and
z-shim vs. no z-shim conditions. The activation
map is overlaid on the T1-weighted image of a
single subject.
402 Du et al.

Page 8

compensation steps, M, can be determined by the range of
compensation and the step size, as shown in Fig. 3. A
drawback of present pulse sequence is that SNR is reduced
in the compensated slices because of the reduced eTR.
Using a larger M will further reduce SNR in these slices.
An additional potential drawback of using a larger M is the
possible contamination of the in-flow effect at eTR ? 1 s
(33,34). In that case, the use of a flip angle lower than the
Ernst angle is advised to lessen the in-flow effect.
The amygdala-hippocampal region, which is of great
interest in fMRI studies of emotion and memory, is also
severely affected by SFGs. These regions are adjacent to
the OFC region along the S/I direction. By properly tilting
the slice orientation, one can cover these regions along
with the OFC region by the same z-shim slices. In addition,
the present technique can readily be applied to multiple
compensation regions if necessary.
The compensation factor fcdoes not have to be the same
in all of the z-shim slices, as implemented in this study. A
smaller fccan be used in a more superior z-shim slice,
where Gsis smaller, to increase the SNR in the composite
image (as indicated in Fig. 3). The present technique can
also be used as volume-selective oversampling without
z-shim (i.e., fc? 0 in the selected volume). In some whole-
brain fMRI applications, the activation in selected regions
of the brain is of particular interest (e.g., hippocampal
activation in memory studies). The present slice-reorder-
ing approach can be used to double or triple the sampling
rate in these regions during an fMRI scan. When we select
Acomp? 0, (1 ? M) images are acquired in a TR, leading to
a shortened temporal resolution of TR/(M ? 1). These extra
images can increase the statistical power in the selected
volume, despite the reduced spatial SNR in each image
(35). The oversampled images can also be combined to
form an SSQ image with a higher SNR. For example, the
composite images in the selected volume can have a
24% higher SNR than the other images when TR ? 4 s
and M ? 2.
The present z-shim approach of adding a compensation
gradient lobe in the slice selection is not optimal in the
compensation of Gs. First, as pointed out in the Theory
section, the early echoes prior to ky? 0 are overcompen-
sated and the latter echoes are undercompensated. As a
result, image blurring along the phase-encoding direction
caused by Gsis not compensated (25). Second, the added
compensation gradient lobe can only compensate (at best)
the zeroth-order moment, but not the first-order moment,
induced by Gs. Such suboptimal compensation can have
implications for image artifacts related to physiological
motion. One could implement a more effective z-shim
compensation scheme by applying a small continuous z-
gradient throughout the entire length of the sequence,
because this z-gradient might exactly counterbalance the
effect of Gs.
In summary, this study presents a volume-selective z-
shim technique to increase the efficiency of data acquisi-
tion. Using this technique, one can effectively compensate
for the susceptibility-induced signal loss in the OFC in
whole-brain fMRI scans while keeping the TR at 2 s. The
BOLD activation was effectively detected in the OFC in a
reward-punishment task.
ACKNOWLEDGMENTS
The authors thank Drs. Jody Tanabe, Marie Banich, Diet-
mar Cordes, Brendan Depue, and Tom Crowley for helpful
discussions, and Debra Singel for acquiring the fMRI data.
REFERENCES
1. Frahm J, Merboldt KD, Hanicke W. Direct FLASH MR imaging of
magnetic field inhomogeneities by gradient compensation. Magn Reson
Med 1988;6:474–480.
2. Ojemann JG, Akbudak E, Snyder AZ, McKinstry RC, Raichle ME, Con-
turo TE. Anatomic localization and quantitative analysis of gradient
refocused echo-planar fMRI susceptibility artifacts. Neuroimage 1997;
6:156–167.
3. Stenger VA, Boada FE, Noll DC. Three-dimensional tailored RF pulses
for the reduction of susceptibility artifacts in T2*-weighted functional
MRI. Magn Reson Med 2000;44:525–531.
4. Norris DG, Zysset S, Mildner T, Wiggins CJ. An investigation of the
value of spin-echo-based fMRI using a Stroop color-word matching task
and EPI at 3 T. Neuroimage 2002;15:719–726.
5. Glover GH, Law CS. Spiral-in/out BOLD fMRI for increased SNR and
reduced susceptibility artifacts. Magn Reson Med 2001;46:515–522.
6. Yang Y, Gu H, Zhan W, Xu S, Silbersweig DA, Stern E. Simultaneous
perfusion and BOLD imaging using reverse spiral scanning at 3T:
characterization of functional contrast and susceptibility artifacts.
Magn Reson Med 2002;48:278–289.
7. Glover GH, Thomason ME. Improved combination of spiral-in/out im-
ages for BOLD fMRI. Magn Reson Med 2004;51:863–868.
8. Wilson JL, Jenkinson M, Jezzard P. Optimization of static field homo-
geneity in human brain using diamagnetic passive shims. Magn Reson
Med 2002;48:906–914.
9. Wilson JL, Jezzard P. Utilization of an intra-oral diamagnetic passive
shim in functional MRI of the inferior frontal cortex. Magn Reson Med
2003;50:1089–1094.
10. Hsu JJ, Glover GH. Mitigation of susceptibility-induced signal loss in
neuroimaging using localized shim coils. Magn Reson Med 2005;53:
243–248.
11. Constable RT. Functional MR imaging using gradient-echo echo-planar
imaging in the presence of large static field inhomogeneities. J Magn
Reson Imaging 1995;5:746–752.
12. Yang QX, Dardzinski BJ, Li S, Eslinger PJ, Smith MB. Multi-gradient
echo with susceptibility inhomogeneity compensation (MGESIC): dem-
onstration of fMRI in the olfactory cortex at 3.0 T. Magn Reson Med
1997;37:331–335.
13. Yang QX, Williams GD, Demeure RJ, Mosher TJ, Smith MB. Removal of
local field gradient artifacts in T2*-weighted images at high fields by
gradient-echo slice excitation profile imaging. Magn Reson Med 1998;
39:402–409.
14. Constable RT, Spencer DD. Composite image formation in z-shimmed
functional MR imaging. Magn Reson Med 1999;42:110–117.
15. Constable RT, Carpentier A, Pugh K, Westerveld M, Oszunar Y, Spen-
cer DD. Investigation of the human hippocampal formation using a
randomized event-related paradigm and z-shimmed functional MRI.
Neuroimage 2000;12:55–62.
16. Cordes D, Turski PA, Sorenson JA. Compensation of susceptibility-
induced signal loss in echo-planar imaging for functional applications.
Magn Reson Imaging 2000;18:1055–1068.
17. Li Z, Wu G, Zhao X, Luo F, Li SJ. Multiecho segmented EPI with
z-shimmed background gradient compensation (MESBAC) pulse se-
quence for fMRI. Magn Reson Med 2002;48:312–321.
18. Deichmann R, Gottfried JA, Hutton C, Turner R. Optimized EPI for
fMRI studies of the orbitofrontal cortex. Neuroimage 2003;19:430–441.
19. Glover GH. 3D z-shim method for reduction of susceptibility effects in
BOLD fMRI, Magn Reson Med 1999;42:290–299.
20. Song AW. Single-shot EPI with signal recovery from the susceptibility-
induced losses. Magn Reson Med 2001;46:407–411.
21. Heberlein KA, Hu X. Simultaneous acquisition of gradient-echo and
asymmetric spin-echo for single-shot z-shim: Z-SAGA. Magn Reson
Med 2004;51:212–216.
22. Li S, Dardzinski BJ, Collins CM, Yang QX, Smith MB. Three-dimen-
sional mapping of the static magnetic field inside the human head.
Magn Reson Med 1996;36:705–714.
23. Collins CM, Yang B, Yang QX, Smith MB. Numerical calculations of the
static magnetic field in three-dimensional multi-tissue models of the
human head. Magn Reson Imaging 2002;20:413–424.
Volume-Selective z-Shim in fMRI403

Page 9

24. Du YP. Whole brain fMRI with steady state z-shimming at selected
slices. In: Proceedings of the 9th Annual Meeting of ISMRM, Glasgow,
Scotland, 2001 (Abstract 1231).
25. Du YP. Image blurring in fMRI with z-shim. In: Proceedings of the 9th
Annual Meeting of ISMRM, Glasgow, Scotland, 2001 (Abstract 1232).
26. Heberlein K, Hu X. Improved shim by subject head positioning. In:
Proceedings of the 9th Annual Meeting of ISMRM, Glasgow, Scotland,
2001 (Abstract 1157).
27. Cusack R, Russell B, Cox SM, De Panfilis C, Schwarzbauer C, Ansorge
R. An evaluation of the use of passive shimming to improve frontal
sensitivity in fMRI. Neuroimage 2005;24:82–91.
28. Kim DH, Adalsteinsson E, Glover GH, Spielman DM. Regularized high-
er-order in vivo shimming. Magn Reson Med 2002;48:715–722.
29. Tzourio-Mazoyer N, Landeau B, Papathanassiou D, Crivello F, Etard O,
Delcroix N, Mazoyer B, Joliot M. Automated anatomical labeling of
activations in SPM using a macroscopic anatomical parcellation of the
MNI MRI single-subject brain. Neuroimage 2002;15:273–289.
30. Brett M, Anton J, Valabregue R, Poline J. Region of interest analysis
using an SPM toolbox. NeuroImage 2002;16(Suppl 1):372–373.
31. Simons JS, Koutstaal W, Prince S, Wagner AD, Schacter DL. Neural
mechanisms of visual object priming: evidence for perceptual and
semantic distinctions in fusiform cortex. Neuroimage 2003;19:613–
626.
32. Wilson JL, Jenkinson M, de Araujo I, Kringelbach ML, Rolls ET,
Jezzard P. Fast, fully automated global and local magnetic field
optimization for fMRI of the human brain. Neuroimage 2002;17:967–
976.
33. Frahm J, Merboldt KD, Hanicke W, Kleinschmidt A, Boecker H. Brain
or vein—oxygenation or flow? On signal physiology in functional MRI
of human brain activation. NMR Biomed 1994;7:45–53.
34. Jezzard P, Song AW. Technical foundations and pitfalls of clinical
fMRI. Neuroimage 1996;4:S63–S75.
35. Constable RT, Spencer DD. Repetition time in echo-planar functional
MRI. Magn Reson Med 2001;46:748–755.
404Du et al.