MAGNETIC RESONANCE IN MEDICINE 22,48 1-492 ( 199 I )

On the Nature and Reduction of the Displacement

Artifact in Flow Images

G. NISHIMURA,JOHN1. JACKSON,
DWIGHT AND JOHNM. PAULY

Department of Electrical Engineering, Stanford University, Stanford, Cal$ornia 94305

Received September 25, 1990 revised November 2, 1990

In flow-imaging experiments with 2-D Fourier transform sequences, the time difference
between phase encoding and readout leads to a potentially misleading displacement artifact.
This artifact arises in regions of rapid flow and high shear, and manifests as an intensity
distortion in addition to a bulk shift. We have studied methods of mitigating the artifact,
including offset-echo acquisition, backward-evolvingphase encoding, moment-compensated
phase encoding, and projection-reconstruction imaging. Experiments on flow phantoms
verified the nature and reduction of this displacement artifact. Of the four methods studied,
the projection-reconstruction sequence proved to be the most effective, completely elim-
inating the artifact. 0 1991 Academic Press, Inc.

INTRODUCTION

Imaging sequences based on phase encoding and gradient-recalled echoes predom-

inate in MR flow-imaging techniques. Although the trend toward shorter echo times
(TE) and more compact gradient waveforms has led to improved flow depiction, a
variety of flow artifacts continue to limit the effectiveness and accuracy of these tech-
niques to characterize or visualize flowing material. A prime example is the artifactual
signal loss due to coherence loss in regions of irregular flow. Another class of flow
artifacts involves displacements and distortions in the reconstructed signals due to
spin motion occurring between the selective excitations and the various gradient wave-
forms. For example, spins flowing obliquely through a thinly excited slice move to a
displaced position during the time from the selective excitation to the readout. A
similar type of displacement stems from the time difference between phase encoding
and readout as spins flowing obliquely to the phase-encoding and readout axes recon-
struct at a displaced position (1-3). This paper focuses on this well-known phase-
encoding displacement artifact. Being negligible in many flow-imaging experiments,
this artifact is often ignored we examine situations however where it may become
significant and misleading. In the following sections, we first discuss the nature of the
phase-encoding displacement artifact and then compare potential imaging sequences
aimed at reducing this artifact.

THE DISPLACEMENT ARTIFACT

We first consider the simplest case of a single spin positioned at (xa, yo) after ex-
citation and flowing at constant velocity with components ( v x , v,,). Using a conven-
48 I 0740-3194/91 $3.00
Copyright 0 1991 by Academic P r q Inc.
All rights of reproduction in any form reserved.
482 NISHIMURA, JACKSON, AND PAULY

tional variable-amplitude phase-encoding y -gradient Gy of duration 7 (applied im-

mediately after the excitation), the phase I # J ( ~of
) the moving spin is

which can be rewritten as

121

The phase behavior of the flowing spin is therefore the same as that of a stationary
+
spin at yo v ~7/2);
( hence the y-component of the spin’s reconstructed position will
be based on the spin’s location at the center of the phase-encoding interval. While the
y-component depends on the phase-encoding interval, the x-component depends on
+
the spin’s position at the gradient echo-that is, at xo v,TE, where TE is the time
from the excitation to the gradient echo. Here, we assume both the excitation pulse
and readout dephasing interval are brief. We also ignore the blurring effect from motion
during the readout interval ( 4 ) . The displacement artifact arises because the x- and
y-positions are encoded at different times. Given the streaming nature of flow, we can
determine the extent of displacement by referencing time with respect to the center
of the phase encoding, 7/2, and considering the time difference from that point.
In this case then, the time difference is AT = TE - 7 / 2 and the displacement is
(VAT, 0 ) .
We now extend this analysis by considering a collection of flowing spins, as in
the case of a straight vessel oriented at some angle 8 with respect to the readout axis
(Fig. 1 ). Examining the behavior in the primed coordinate system with y’ aligned
with the longitudinal axis of the vessel, a spin positioned at (xb, y b ) at the center
+
of the phase-encoding interval will be reconstructed at (xb V A T cos f? sin 8, yb
+ VATsin f? sin f?), where v is the flow velocity (assumed parallel to the y’ axis) and
AT is the time from the center of the phase-encoding lobe to the gradient echo. Because
of the streaming nature of flow along y’, the artifact can be studied by considering
only the displacement in the x’direction. This x’-displacement is u A T cos 8 sin 8 (or
(VAT/2 ) sin 20), which is clearly a maximum when 8 = 45 O .

FIG.I . Oblique vessel oriented at angle 0 to the readout (x) and phase-encoding ( y ) axes. Flow is assumed
to be along the y’ axis.
 DISPLACEMENT ARTIFACT IN FLOW IMAGES 483

In projective vessel imaging, the resultant projection profile will depend on the
distribution of velocities across the lumen because the profile is based on the line
integral through the lumen. With plug flow, the uniform velocity distribution leads
to the same amount of displacement for all spins in the vessel, and therefore the
displacement causes a bulk shift of the vessel with no distortion of the projection
profile.
With parabolic flow, v ( x ’ , z) = 2u,, ( 1 - (x” + z 2 ) / R 2 )where
, R is the vessel
radius and v,, is the average velocity. In this case, the distribution of velocities leads
to a distribution of displacements, creating a more subtle effect on the resultant line
integrals. Figure 2 plots the computed vessel projection profile (along x‘) for the case
of parabolic flow through a circular lumen. The ideal semiellipse profile when v,,, = 0
becomes distorted when v,, = 20 and 40 cm/s ( A T = 4 ms, 0 = 45”, R = 0.32 cm
( 8 in.)). The asymmetry of the profile with increasing velocity is a result of a “piling-
up” effect in the direction of the shift as faster flowing material, displaced by a greater
amount, reconstructs on top of adjacent slower flowing material ( 5 ) . The vessel profile
boundaries do not necessarily shift because of the lower velocities near the vessel wall.
Clearly, the distortion becomes appreciable when the displacement reaches the order
of a pixel width.
More generally, the effect seen in Fig. 2-in which the signal piles up-becomes
more pronounced in regions of high shear. Where a steep velocity gradient exists,
regions of relatively high velocities closely border regions of low velocities, resulting
in a “discrete” displacement. As an extreme example, we consider a 1-D velocity
profile consisting of three regions of plug flow (Fig. 3). The middle region will be
shifted by an amount greater than that of the two outer regions; therefore, the projection

FIG.2. Simulated vessel projection profiles: Ideal semiellipse profile of circular lumen when v, = 0 cm/
s (solid line) becomes distorted for v,,
= 20 cm/s (dashed line) and u,, = 40 cm/s (dotted line). Parabolic
velocity distriblltion assumed with AT = 4 ms, 0 = 45”, and R = 0.32 cm ( in.).
484 NISHIMURA, JACKSON, AND PAULY

velocity
profile

projection profile
FIG.3. Extreme shear example: Given a velocity distribution comprised of three regions of plug flow, the
middle region displaces more than the outer regions, creating a sharp peak and a void in the resultant
projection profile.

profile will show a sharp peak in the right transition region where the shift has resulted
in an overlap, and a void in the left transition region where the shift has rendered a
gap. This effect is similar to a water/fat chemical shift artifact in which fat shifts with
respect to water, leaving a void and creating a bright region of overlap.

ARTIFACT-REDUCTION METHODS

To minimize the phase-encoding displacement artifact, 2-D Fourier transform (FT)

sequences typically use a brief phase-encoding lobe and abut the phase-encoding in-
terval to the readout interval. In this section, we discuss four strategies to mitigate this
artifact further: ( 1 ) offset gradient-echo acquisition, ( 2 ) backward-evolving phase en-
coding, ( 3) moment-compensated phase encoding, and (4) projection-reconstruction
imaging. Figure 4 compares the gradient waveforms for these four sequences.

Oflset Gradient Echo Acquisition

Acquisition of offset gradient echoes (Fig. 4a) has proven effective at reducing flow-
dependent dephasing artifacts by shortening the gradient intervals before the gradient
echo, thereby minimizing the gradient moments (6). The offset also reduces TE and
AT, the time interval between phase encoding and the readout. Because of the asym-
metric nature of the data acquisition, partial k-space reconstruction techniques can
be applied to sharpen an otherwise blurred image.

Backward-Evolving Phase Encoding

To reduce the time difference AT, this approach keeps the phase-encoding interval
as short and as close as possible to the readout for each phase encode. This results in
a variable-duration phase-encoding lobe similar to that proposed in ( 7). However, in
contrast with the forward-evolving, variable-TE approach of ( 7), this approach uses
a phase-encoding lobe that terminates next to the readout interval and lengthens toward
 DISPLACEMENT ARTIFACT IN FLOW IMAGES 485

FIG.4. Comparison of test sequences: (a) Offset echo, (b) backward-evolvingphase encodin&(c) moment-
compensated phase encoding, (d) projection reconstruction.

the RF pulse with each higher-order phase encode, maintaining a fixed TE (Fig. 4b).
To analyze the impulse response in this case, the focus is on the interval, parameterized
by T ~which
, vanes with each phase encode. Thus, the phase upon completion of the
phase-encoding interval is

t 31

= Y.o[
YO + %( 7 - ?)I7. 141

for the positive phase encodes ( -Gyo applies for the negative phase encodes). The
impulse response amounts to a blur in this variable-duration scheme, not a simple
shift as in the variable-amplitude scheme. However, one simple interpretation of Eq.
[4] is that the effective y-position, yo -t vY(7 - ~ ~ / 2 although
), varying with each
phase-encoding step, is based nearer the end of the phase-encoding interval for the
lower-order phase encodes.

Moment-Compensated Phase Encoding

In this approach, a modified phase-encoding gradient waveform is applied in which
its first moment effectively extrapolates the reconstructed y-position to a time coin-
cident with the gradient echo (2). The waveform that achieves this first-order extrap-
olation (Fig. 4c) produces a phase at the end of the interval (time 7 ) of
486 NISHIMURA, JACKSON, AND PAULY

#J = YGJYO+ VYTE)~. [51

Hence the two equations that specify the phase-encoding waveform relate to the gra-
dient's area and first moment. Given a bipolar waveform with rectangular segments,
the relationships are
+
G ~ ( Y T G2( 1 - ( Y ) T = Gyr [61
for the area and

* + G,( 1 - a') -
T L TL
G ~ ( Y- G,TTE [71
2 2
for the first moment. Thus moment compensation for displacement reduction cor-
responds to making the first moment large, not zero. To determine the waveform
parameter values, we typically know Gyr = A (the desired phase-encoding area) and
G,, (the maximum gradient amplitude). Letting G2 = -GI = G,, and rearranging
Eqs. [ 61 and [ 71, expressions for r and (Y are

and
1 A
ff=--- [91
2 2Gma,r'
Note that LY < i , which implies that the second lobe will be longer than the first.
A trade-off with this type of moment compensation is the long gradient interval r
involved. To examine how severe this is, we let /3 = (TE - r ) be the time from the
end of phase encoding until the gradient echo. Therefore /3 represents how far forward
+
in time the phase-encoding lobes must compensate. Substituting /3 r for TE in Eq.
[ 8 1, it can be derived that

A
Gmax
For comparison, conventional uncompensated single-lobe phase encoding would use
a gradient of amplitude Gm,, and duration rc to achieve the desired gradient area A .
Thus,

Substituting this expression into Eq. [lo], we obtain an expression relating T to 7,:

As an example, if p = 1.0 ms, then r GS 3rc and a GZ 0.33. To make /3 approach 0

(easing the compensation), shorter readouts (or greater echo offsets) must be used.
 DISPLACEMENT ARTIFACT IN FLOW IMAGES 487

However, even in the easiest case in which p = 0, the moment-compensated lobes are
2.4 1 times longer than the conventional uncompensated lobe.

2-0 Projection-Reconstruction Imaging

An entirely different means of reducing the artifact is a projection-reconstruction
(PR) sequence ( 8 ) .One approach to 2-D PR imaging is with FID signal acquisition
(Fig. 4d), that is, beginning the data sampling at the base of the gradient ramp and
omitting the preparatory dephasing lobe that conventionally exists prior to readout.
Therefore the first data sample is consistent with each readout direction and the k-
space trajectory consists of radial lines emanating from the origin. The immediate
sampling of the k-space origin without any preparatory readout gradient lobe( s) also
minimizes flow-dependent phase shifts. An additional feature of a 2-D PR sequence
is the short TE made possible by the absence of a phase-encoding lobe.
Compared to an FT sequence, the symmetry of the radial k-space trajectory in a
PR sequence leads to a different impulse response in the presence of flow. The impulse
response h( Y , 8 ) of the 2-D PR sequence (ignoring gradient ramps) can be derived by
considering a spin flowing from the origin with velocity and angle $. With a readout
gradient of amplitude G, at a variable angle 8, the resultant signal S( t ) is
S( t ) = e - i ~ G r ~ c o s ( ~ - J . ) ( t 2 ,/ 2 ) t131
where y is the gyromagnetic ratio. Time maps to radial k-space position according to
the relationship
7
p = - G,t,
2?r
where p is the k-space radial variable in polar coordinates. Thus the impulse response
is the inverse 2-D Fourier transform given by
h ( r , 6) = 9;;
{ e - i 2 r b ( ~ 2 / 2 COS(0
) - It,}, ~ 5 1
where b = 2a/yG,. An example of the impulse response for a spin flowing in the x-
direction (It = 0) is presented in Fig. 5 . The impulse response shows a shift of energy
in the flow direction, and blur extending asymmetrically in all directions, although
primarily in the flow direction. Given the radial symmetry of the acquisition, this
response rotates accordingly with the flow direction It. Note that the shift observed in
the impulse response, since it occurs in the flow direction, will be masked when imaging
a continuum of flowing spins. This analysis lends some indication that the displacement
artifact does not apply to a 2-D PR sequence.

EXPERIMENTAL RESULTS

We conducted flow-phantom experiments on a GE 1.5-T Signa system to dem-

onstrate the nature of the displacement artifact and to study the effectiveness of the
artifact-reduction methods. We implemented and compared five sequences-one
“reference” sequence in addition to the four test sequences. All sequences produced
projective images with in-plane spatial resolution of 0.7 by 0.7 mm. For the 2-D FT
sequences, phase encoding was along the vertical axis and readout along the horizontal
488 NISHIMURA, JACKSON, AND PAULY

FIG.5. 2-D Projection-reconstruction impulse response to a flowing spin: The flowing spin is assumed to
have velocity in the x-direction corresponding to motion of about 1.5 pixels during readout. The response
reveals asymmetric blur extending in all directions with energy shifted in the x-direction. Magnitude response
is shown next to an ideal impulse at x = 0 for comparison.

axis. Partial k-space reconstruction ( 9 ) was applied for the highly offset gradient-echo
acquisitions. To restrict the field-of-view along the primary flow direction, all sequences
used the same mildly selective asymmetric RF pulse with its main lobe offset toward
the readout to shorten the TE and to reduce phase shifts (6). Table 1 lists the TE,
echo offset, and AT for the five sequences.
The reference sequence was a 2-D FT sequence, velocity-compensated along the
readout direction, with a 2.15-ms trapezoidal phase-encoding lobe adjacent to the 6.2-
ms readout interval (TE = 6.8 ms, echo offset = 25%, AT = 2.6 ms).
A highly offset gradient-echo 2-D F T sequence served as one of the test sequences
(TE = 3.8 ms, echo offset = 42%, AT = 1.5 ms). With an echo offset of 42%, partial
k-space reconstruction becomes desirable for improving the detail of smaller structures.
The backward-evolving phase-encoding sequence (TE = 3.8 ms, echo offset = 42%,
AT = 1.0 ms) was approximated in the following manner. For the lower-order phase

TABLE 1
Test Sequence Parameters

Sequence TE (ms) Echo offset AT(ms)

Reference (velocity-compensated) 6.8 25% 2.6

Highly offset echo 3.8 42% 1.5
Backward-evolving phase encoding 3.8 42% 1.0
Moment-compensated phase encoding 6.6 42% 0
Projection reconstruction 1.o 50% 0

Note. AT = time difference between phase encoding and readout.

 DISPLACEMENT ARTIFACT IN FLOW IMAGES 489

encodes, the amplitude of a 1.0-ms triangular lobe adjacent to the readout interval
was incremented until the maximum amplitude of 1 G/cm was attained. For the
higher-order phase encodes, the lobe became trapezoidal at the maximum amplitude
as the additional area under the lobe was gained by increasing the interval toward the
RF pulse. The readout gradient was the same as in the highly offset echo sequence.
The moment-compensated sequence (TE = 6.6 ms, echo offset = 42%, AT = 0
ms) possessed a longer TE because of the additional gradient action needed to create
the large first moment. The phase-encoding gradient consisted of two lobes of equal
amplitude, the first of 1.7 ms and the second of 3.3 ms. This gave an effective AT of
0 ms, assuming constant velocity during the sequence. The same readout gradient
was used as in the previous two sequences.
For the PR sequence (TE = 1.0 ms, echo offset = 50%, AT = 0 ms), 256 radial
lines spaced over a 27r angular extent were acquired and reconstruction was performed
by interpolating the data onto a 2-D Cartesian grid prior to a 2-D FFT.Data acquisition
commenced at the base of the gradient ramp to minimize phase shifts and TE. The
readout interval for each radial k-space “spoke” was 3.1 ms.
Figures 6 and 7 show images obtained on straight and branching flow phantoms.
All of the images shown are magnified by 3X. In Fig. 6 , we can compare the results

FIG.6. Oblique tube phantom: (a) Reference Sequence, flow OFF; ( b ) reference sequence, flow ON from
bottom (uavc = 60 cm/s). Signal displacement causes signal pile up toward the right. Test sequences: (c)
Offset echo, (d) backward-evolving phase encoding, (e) moment-compensated phase encoding, (f) projection
reconstruction. Images (c) and (d) show progressively reduced artifact. Images (e) and ( f ) show no artifact.
490 NISHIMURA, JACKSON, AND PAULY

FIG. 7. Branching phantom: ( a ) Reference sequence, flow OFF, ( b ) reference sequence, flow ON from
bottom (nave= 40 cm/s). Artifact appears as bright oblique line beginning at the origin of the left branch
and as a diminished signal at the inner wall of the left branch. Test sequences: (c) Offset echo, ( d ) backward-
evolving phase encoding, (e) moment-compensated phase encoding, ( f ) projection reconstruction. Images
(c), ( d ) , (e) show progressively reduced artifact with better “filling” of the inner wall of the left branch.
Image ( f ) shows no artifact.

of constant flow through a straight segment of a 0.635-cm (i in.)-diameter Tygon

tube oriented at about 45”. Figure 6a is from the reference 2-D FT sequence with no
flow while Fig. 6b is from the same 2-D FT sequence with flow (average velocity of
60 cm/s) entering from the bottom. Given the oblique orientation of the tube, we
can clearly see the shift in the signal intensity toward the right side of the tube, the
nature of which corresponds to the simulation results of Fig. 2. Because of the distri-
bution of flow velocities present, there exists a distribution of displacements, which
results in a bunching up of the signals. With an average velocity of 60 cm/ s, an oblique
angle of about 45 O , and a time differenceof 2.6 ms between readout and phase encoding,
the average shift (perpendicular to the tube’s longitudinal axis) is 0.8 mm, correspond-
ing to about one pixel for the image of Fig. 6b. The results of the four test sequences
are given in Figs. 6c-6f. The highly offset echo sequence (Fig. 6c) shows a slight
reduction in shift versus the reference sequence. There remains some asymmetry in
the intensity distribution as indicated by the off-centered position of the brightest
region and the darker signal on the left side of the tube. An additional slight improve-
ment in the intensity distribution can be seen for the backward-evolving case (Fig.
 DISPLACEMENT ARTIFACT IN FLOW IMAGES 49 1

6d) as the signal on the tube’s left side is nearly comparable with that on the right
side. Both the moment-compensated phase-encoding sequence (Fig. 6e) and the 2-D
PR sequence (Fig. 6f) reveal no shift artifact.
The second experiment is of a branching phantom (Fig. 7) with a diameter of 0.47
cm ( h in.) in the main segment. Compared to the reference 2-D F T image with no
flow (Fig. 7a), the reference 2-D FT image with flow (average velocity of 40 cm/s)
entering the main segment from below is shown in Fig. 7b. The artifact in Fig. 7b
appears as an unusually bright line beginning near the origin of and extending into
the left branch. Correspondingly, there is signal loss along the inner wall of the left
branch. Such an artifact is attributable to the high shear near the inner walls of the
branch (10)-the distortion manifests as a region of diminished signal next to a dis-
tinctly bright region as discussed earlier. We observe some reduction in this artifact
in the test sequence images of Figs. 7c-7f. Both the highly offset echo image (Fig. 7c)
and the backward-evolving phase-encoding image (Fig. 7d) show a small reduction
in the intensity and extent of the artifact. A more marked reduction is seen in the
moment-compensated image (Fig. 7e). The artifact is completely absent in the 2-D
PR image (Fig. 7f) as this image compares favorably with the reference image obtained
without flow.

DISCUSSION

These results demonstrate the potentially misleading nature of the displacement

artifact, particularly in regions where high flow velocities and high shear exist. The
shift creates a distorted flow signal with regions of both enhanced and diminished
intensity. This artifact may therefore be relevant when imaging or measuring flow in
the carotid bifurcation, aorta, and regions of extreme curvature and stenoses. In these
cases, care must be taken to avoid misinterpretation of the artifact as a pathology, a
dephasing artifact, or inadequate in-flow. The artifact will become more significant
in high-resolution imaging because of the smaller voxel size and longer gradient intervals
involved.
To reduce this displacement artifact in 2-D F T sequences, standard measures include
shortening the phase-encoding lobe and abutting it to the readout interval. The results
indicate that the artifact can be reduced further by acquiring an echo offset toward
the phase-encoding lobe. The backward-evolving phase-encoding scheme, while less
convenient to implement, provides additional immunity without introducing con-
straints on the sequence. The moment-compensated phase-encoding method reduces
the displacement artifact in certain cases; however, the additional lobe and increased
gradient power that is required significantly increases the minimum TE. The large
first moment may also pose problems when flow varies from view to view such as in
ungated scans of pulsatile regions. All of these measures for 2-D FT imaging reduce
but do not eliminate the artifact.
Perhaps the most intriguing of these results is that PR imaging with FID measure-
ments avoids the artifact altogether. In addition to eliminating the displacement artifact,
a significant advantage of PR imaging is the highly coherent flow signals it produces
because of the short TE and minimal flow-inducedphase shifts near the k-space origin.
However, other considerations arise with PR imaging. Compared to conventional FT
492 NISHIMURA, JACKSON, AND PAULY

sequences, the PR sequence requires more measurements because of the nonuniform

sampling density and acquisition of half-lines in k-space. Possible remedies for scan
time reduction include partial k-space acquisition tailored for PR imaging ( 11 ) and
modification of the PR gradient waveforms for more efficient k-space sampling ( 12).
Another concern with PR imaging is its off-resonance behavior, characterized by a
blur instead of a shift. However, this problem is less severe because a single component
(i.e., blood) is being imaged. Also shorter readout intervals can be used to reduce the
amount of blur, and the demodulation frequency during image reconstruction can be
vaned to “focus” appropriately. Despite these concerns, the results indicate that PR
imaging, with its excellent immunity to flow artifacts, may be a viable alternative to
conventional 2-D l T imaging for flow applications.
ACKNOWLEDGMENTS
The authors gratefully acknowledge the support of the GE Medical Systems Group and the National
Institutes of Health Grants HL-34962, HL-39478, and HL-39297. The authors also thank Dr. Gary Glover
for use of his PR reconstruction program.

REFERENCES
I. G. K. SCHULTHESS AND C. B. HIGGINS, Radiology 157, 687 (1985).
2. D. G. NISHIMURA, A. MACOVSKI, AND J. M. PAULY,IEEE Trans. Med. Imaging MI-5, 140 ( 1986).
3. D. N. FIRMIN, G. L. NAYLER, P. J. KILNER,AND D. B. LONGMORE, Magn. Reson. Med. 14,230 ( 1990).
4. V. J. WEDEEN,R. E. WENDT,AND M. JEROSCH-HEROLD, Magn. Reson. Med. 11, 114 ( 1989).
5 . D. 0. KUETHEAND R. J. HERFKENS, Magn. Reson. Med. 10,57 (1989).
6. D. G. NISHIMURA, A. MACOVSKI, J. I. JACKSON,B. S. Hu, C. A. STEVICK, AND L. AXEL,Magn. Reson.
Med. 8,96 (1988).
7. A. KUMAR,D. WELTI,AND R. R. ERNST,J. Magn. Reson. 18,69 (1975).
8. P. C. LAUTERBUR, Nature 242, 190 (1973).
9. D. C. NOLL,D. G. NISHIMURA, AND A. MACOVSKI, IEEE Trans. Med. Imaging MI-10, 154 ( I 99 1).
10 F. J. WALBURN AND P. D. STEIN,J. Biornech. 14(9), 601 (1981).
11. D. C. NOLL,J. M. PAULY,D. G. NISHIMURA, AND A. MACOVSKI, in “Proceedings, 9th SMRM-WIP,”
p. 1314, 1990.
12. J. I. JACKSON, C. H. MEYER,D. G. NISHIMURA, AND A. MACOVSKI, in “Proceedings. 9th SMRM,” p.
428. 1990.