JOURNAL OF MAGNETIC RESONANCE IMAGING 21:297–304 (2005)

Original Research

Real-Time Fourier Velocity Encoding: An In Vivo

Evaluation
Christopher K. Macgowan, PhD,1* Christian J. Kellenberger, MD,1 Jay S. Detsky,1
Kevin Roman, MD,1,2 and Shi-Joon Yoo, MD1,2

tant part of the clinical investigation of cardiovascular

Purpose: To compare in vivo real-time Fourier velocity en-
coding (FVE), spectral-Doppler ultrasound, and phase-
disease. Doppler ultrasound is the most commonly
contrast (PC) magnetic-resonance (MR) imaging. used non-invasive tool for this purpose. However, eval-
uation by Doppler ultrasound is impossible when there
Materials and Methods: In vivo velocity spectra were mea- is air, bone, or surgical scar in the ultrasound path, and
sured in the suprarenal and infrarenal aorta and the hepatic
it is inaccurate when the ultrasound beam cannot be
segment of the inferior vena cava of eight normal volunteers
using FVE, and compared to similar measurements using properly aligned with the vessel axis, which is often
Doppler ultrasound and gated PC MR imaging. In vivo wave- encountered when investigating the thoracic vessels.
forms were compared qualitatively according to flow pattern Instead, phase-contrast (PC) cine magnetic-resonance
appearance (number, shape, and position of velocity peaks) (MR) imaging (1,2), also known as phase velocity cine
and quantitatively according to peak velocity. (PVC) MR imaging (3) and velocity encoded cine (VEC)
Results: Good agreement was obtained between peak veloc- MR imaging (4), can be used to measure flow in thoracic
ities measured in vitro using FVE and PC MR imaging (R2 ⫽ vessels, but requires cardiac gating that performs
0.99, P ⫽ 2.10⫺6, slope ⫽ 0.97 ⫾ 0.05). Qualitatively, the FVE poorly in arrhythmic patients. It also requires some
and ultrasound measurements agreed closely in the majority form of respiratory compensation, such as respiratory
of in vivo cases (excellent or good in 21/24 cases) while the PC gating or breath holding, to prevent motion artifacts.
MR method resolved fewer velocity peaks due to the inherent Unfortunately, respiratory gating increases the scan
temporal averaging of cardiac-gated studies (excellent or good time while breath holding may result in hemodynamic
agreement with FVE in 13/24 cases). Quantitatively, the FVE
changes and is impossible for some patients due to age,
measurement of peak velocity correlated strongly with both
ultrasound (R2 ⫽ 0.71, P ⫽ 2.10⫺7, slope ⫽ 0.81 ⫾ 0.08) and
sedation, or shortness of breath.
PC MR (R2 ⫽ 0.85, P ⫽ 2.10⫺10, slope ⫽ 1.04 ⫾ 0.08). Real-time MR methods have been developed to eval-
uate flow and do not require electrocardiogram (ECG)
Conclusion: Real-time MR assessment of blood-flow veloc- gating or respiratory compensation. Real-time PC MR is
ity correlated well with spectral Doppler ultrasound. Such
similar to conventional cardiac-gated PC MR but uses
new methods may allow hemodynamic information to be
acquired in vessels inaccessible to ultrasound or in pa- more efficient data-acquisition methods (e.g., multi-
tients for whom respiratory compensation is not possible. echo or spiral) (5,6). Real-time PC MR is suitable for
qualitative flow visualization and quantitative volumet-
Key Words: blood flow; magnetic resonance imaging; pul- ric flow measurement in large vessels (6); however, its
monary heart disease; stenosis; ultrasonics; valves
spatial resolution (e.g., 2–3 mm in-plane) is insufficient
J. Magn. Reson. Imaging 2005;21:297–304.
© 2005 Wiley-Liss, Inc.
for quantitative measurement of spatial-peak velocity
in pulsatile flow jets, an important component of the
hemodynamic grading of stenoses (6,7), due to intra-
ASSESSMENT OF THE blood-flow velocity, volume, and voxel velocity averaging.
waveform in the major vessels in the chest is an impor- An alternative MR method has been developed to
measure velocity spectra in real-time that is suitable for
stenotic-flow assessment (8,9). High temporal resolu-
1
Department of Medical Imaging, University of Toronto and Hospital for tion is achieved by evaluating the velocity spectrum
Sick Children, Toronto, Canada. within one vessel at a time, similar to spectral Doppler
2
Department of Paediatrics, University of Toronto and Hospital for Sick ultrasound, rather than the spatial distribution of ve-
Children, Toronto, Canada.
locity across an entire anatomic slice. Because this
Contract grant sponsor: Ontario Research and Development Challenge
Fund (ORDCF); Contract grant sponsor: Sun Microsystems Canada. method resolves the velocity spectrum within a desig-
*Address reprint requests to: C.K.M., Department of Diagnostic Imag- nated voxel, it is less sensitive to partial volume effects
ing, The Hospital for Sick Children, 555 University Ave., Toronto, On- than two-point PC MR imaging methods. Larger voxels
tario, M5G 1X8 Canada. E-mail: [email protected]
Received September 2, 2004; Accepted November 18, 2004.
can thus be used to encompass the flow jet and mea-
DOI 10.1002/jmri.20266 sure peak velocity. With this method, a multi-echo ac-
Published online in Wiley InterScience (www.interscience.wiley.com). quisition is used in which successive echoes are in-
© 2005 Wiley-Liss, Inc. 297
298 Macgowan et al.

(8). Briefly, a two-dimensional radio-frequency pulse

was used to excite a cylinder of blood in a vessel (11–
13). The diameter (full-width-half maximum) of this cyl-
inder was set interactively to fit within the vessel of
interest. A rapidly oscillating magnetic-field gradient
was then applied during data acquisition to obtain ve-
locity information at high temporal resolution. The pro-
cessed velocity data from a 1-cm segment of the excited
cylinder were then displayed as a spectral waveform in
real-time.
The relevant experimental parameters used in our
implementation of the FVE method were: flip angle ⫽
20°, temporal resolution ⫽ 42 msec, acquisition band-
width ⫽ ⫾ 125 kHz, samples/echo ⫽ 7560, and partial
k-space filling ⫽ 60%. The velocity sensitivity of the FVE
method was controlled by adjusting the amplitude of
the readout gradient, which was measured directly and
used in the velocity calculations (14). The MR signal
from blood moving through a constant magnetic-field
gradient accrues phase in proportion to the blood ve-
Figure 1. FVE pulse sequence showing the amplitudes of the
locity and the first moment of the gradient. Therefore,
magnetic-field gradients (Gx, Gy, Gz) and the radio-frequency the greatest sensitivity was achieved by maximizing the
(RF) pulse for a single measurement of the velocity spectrum. gradient amplitude as shown in Fig. 1, which produced
First, a two-dimensional RF pulse is applied to excite a cylin- an aliasing velocity of ⫾ 53 cm/second and a velocity
der of blood in a vessel. An oscillating gradient (Gz) is then resolution of 3.8 cm/second. Higher aliasing velocities
applied during data acquisition to provide velocity encoding. were achieved by lowering the amplitude of the readout
The aliasing velocity and velocity resolution of the measure- gradient at the cost of a proportionally lower velocity
ment are inversely proportional to the first moment of each resolution. This method to control the aliasing velocity
lobe of the oscillating gradient and the first moment of the
was chosen for practicality: nearly arbitrary scaling of
entire oscillating gradient, respectively. The specific parame-
ters used in the flow study are listed in the text.
the velocity sensitivity could be easily performed and
the k-space trajectory was only linearly scaled, so the
same parameters could be used to process data ac-
creasingly sensitive to velocity. Such Fourier velocity quired at any aliasing velocity. Alternatively, the veloc-
encoding (FVE) complements the considerable cardio- ity sensitivity could have been reduced by narrowing
vascular information already provided by MR imaging each trapezoid in the readout gradient. This second
and has the potential of acquiring hemodynamic infor- approach was not used because, although it makes
mation from vessels inaccessible to ultrasound (10). optimal use of the gradient hardware, it also changes
The primary purpose of this study was to correlate in the k-space trajectory and so would require measure-
vivo measurements of blood-flow velocity acquired us- ment of new processing parameters at each aliasing
ing FVE and spectral Doppler ultrasound. Both quali- velocity. Once the FVE scan began, data were acquired
tative and quantitative comparisons were made be- for approximately 5.4 seconds before the sequence re-
tween waveforms acquired in healthy volunteers. turned to anatomic imaging. The acquired flow data
Velocity measurements from the two methods were also were gridded to a Cartesian matrix (size: kx ⫻ kv ⫽ 16 ⫻
compared to those obtained using conventional cardi- 28) using a 2 ⫻ 2 sinc-weighted kernel, Kaiser win-
ac-gated PC MR imaging. dowed (␤ ⫽ 3.5) and homodyne reconstructed (15).

MATERIAL AND METHODS Real-Time Graphical Interface

The FVE method was implemented on a clinical MR
A real-time interface to the MR scanner was developed
scanner and experimental velocity data were obtained
that enabled the data acquisition and processing to be
and compared to those from spectral Doppler ultra-
controlled interactively (see Fig. 2). This interface was
sound and conventional PC MR. The implementation
developed from an existing application (16) but was
and testing of the method are described below.
redesigned specifically for real-time flow assessment.
All data processing was performed by this application
Real-Time Velocimetry directly on the MR console, an SGI Octane (SGI, Moun-
An MR pulse sequence able to acquire anatomic images tain View, CA).
and velocity spectra in real-time was implemented on a The upper portion of the interface provided control
1.5-T MR system (Signa CV/i, GE Medical Systems, over image-plane navigation and basic real-time imag-
Milwaukee, WI). ing parameters including flip angle, slice thickness, and
Blood-flow velocity spectra were acquired using the spatial resolution. Scan-plane navigation was possible
FVE sequence shown in Fig. 1. This method was first by graphically prescribing oblique slices on succes-
introduced by Hu et al (9) and later refined by Pat et al sively displayed anatomic images.
In Vivo Fourier Velocity Encoding 299

Figure 2. Real-time interface to the MR

scanner, developed for FVE scanning.
The upper portion of the interface pro-
vides basic controls for image naviga-
tion. An axial image of a volunteer at the
level of the hepatic segment of the infe-
rior vena cava (IVC) is shown, with a
yellow circle drawn around the vessel to
prescribe the FVE sample volume. The
bottom portion of the interface provides
basic control of the FVE scan and dis-
plays the resulting flow waveform.

Once a cross-sectional image through a vessel of in- The maximum velocity of each FVE spectrum was
terest was obtained, a circular region of interest (ROI) defined as the largest velocity component above a
could be graphically prescribed in the vessel to specify threshold spectral amplitude (25% of maximum), as
the sample volume for the FVE scan (see Fig. 2). The shown in Fig. 3. This 25% value was chosen empirically
application of the cylindrical excitation as a spatial- to be greater than observed background spectral arti-
saturation pulse during imaging could also be toggled fact. The maximum velocity was measured at each ser-
on and off. This allowed the true position and size of the vo-motor speed, with an uncertainty equal to the veloc-
sample volume to be visualized as a darkened region of ity resolution of each acquisition.
the image. High resolution axial PC MR images were obtained
The lower portion of the interface provided control using an ungated sequence with the following parame-
over the duration, aliasing velocity, and flip angle of the ters: flip angle ⫽ 30°, slice thickness ⫽ 10 mm, field of
real-time flow measurement. The aliasing velocity was view ⫽ 20 cm, matrix ⫽ 512 ⫻ 512, number of aver-
set manually based on the expected peak velocity in ages ⫽ 8, bandwidth ⫽ ⫾ 15.625 kHz, and TE/TR ⫽ 6
each vessel, and increased if aliasing was observed. msec/18 msec. Magnitude masking of the phase im-
Imaging paused automatically during the acquisition of ages was not performed. Spatial-maximum velocities
velocity data but resumed immediately after the velocity were measured in each PC image by averaging the ve-
spectra were displayed. locities from a small ROI (25% of tube area) positioned

In Vitro Experiments
The FVE method was first evaluated using an experi-
mental flow apparatus. This was constructed from a
gear pump (AMBV1A, Hypro Corporation, New Brigh-
ton, MN) and computer-controlled servo motor
(BSM80N-250AA, Baldor Electric Company, Fort
Smith, AR), which pumped water through reinforced
nylon tubing into the scan room. To ensure a symmetric
and developed parabolic flow profile over the measure-
ment volume, a rigid Plexiglas tube of inner-diameter
9.5 mm and length 2 m carried the pumped water
through the bore of the magnet.
Velocity data were acquired using the FVE method
and compared to the data from PC MR. Data acquisition
was performed under steady flow conditions using the Figure 3. Experimental FVE spectrum of constant flow
body coil, with the speed of the servo motor increased through the in vitro apparatus (flow waveform inset). To char-
incrementally between acquisitions. All data were saved acterize maximum flow velocities reproducibly in the presence
for offline analysis in MATLAB (Mathworks, Natick, of spectral artifact, the velocity at which the spectrum dropped
MA). to 25% of its maximum amplitude was measured (vertical line).
300 Macgowan et al.

at the point of highest flow. The correlation and velocity profile with an approximate area 25% that of
weighted linear fit between the FVE and PC MR mea- the vessel.
surements were then calculated. The FVE scans were conducted using the parameters
described above (see In Vitro Experiments section). The
MR sample volume was intentionally extended beyond
In Vivo Experiments the walls of the studied vessel during initial scans to
include signal from the surrounding static tissue. This
In vivo studies were performed in eight normal adult
static tissue appeared as a horizontal band in the wave-
volunteers using spectral Doppler ultrasound, a seg-
form that could act as a zero-velocity marker for sub-
mented PC MR imaging method, and the FVE method to
sequent acquisitions. As with the ultrasound measure-
produce 24 measurements of pulsatile flow. Approval of
ments, three FVE measurements of peak velocity were
all procedures was obtained from our institutional re-
averaged from each vessel for analysis.
view committee, and informed consent was obtained
An experienced radiologist (C.J.K.) qualitatively eval-
from each subject prior to examination. Abdominal ves-
uated the flow waveforms measured by each method
sel locations were chosen for this study because they according to 1) the number of waves shown, 2) the
were easy to access by ultrasound and easy to land- relative width and timing of the waves, and 3) the rela-
mark using either ultrasound or MR imaging. Flow tive amplitude of each wave for each vessel. Based on
waveforms were obtained in the suprarenal and infra- the number of characteristics that were equivalent be-
renal aorta and the hepatic segment of the inferior vena tween two waveforms, the agreement between FVE, ul-
cava of each subject. Waveforms were compared quali- trasound, and PC MR was rated as excellent (three
tatively according to the flow pattern and quantitatively matching characteristics), good (two matching charac-
according to the peak velocity. teristics), fair (one matching characteristic), or poor (0
Ultrasound measurements were performed by an ex- matching characteristics). Quantitatively, the correla-
perienced radiologist (C.J.K.) using a commercial du- tion and linear fit between the real-time and PC mea-
plex ultrasound system (Acuson Sequoia, Mountain surements of peak velocity were calculated and a
View, CA) with a 3-MHz transducer. All subjects had Bland-Altman analysis of the data was performed. The
fasted for over 4 hours prior to being scanned to reduce velocities measured using the two MR methods were
artifacts from intestinal gas. Pulsed-wave Doppler re- also correlated and fitted to values measured using
cordings were performed during quiet-breathing. The Doppler ultrasound.
site of sampling was guided by grey scale imaging to
position the sample gate (4 mm) in the center of the
vessel and to minimize the angle of insonation between RESULTS
the targeted vessel and the Doppler beam (less than In Vitro Evaluation of the FVE Method
60°). Recordings were made in the hepatic segment of
Presented in Fig. 4 is a comparison between the FVE
the inferior vena cava (2–3 cm inferior to the junction
and PC MR measurements of velocity obtained using
with the hepatic veins), in the suprarenal aorta (1–2 cm
the experimental flow apparatus. These measurements
proximal to the celiac trunk), and in the infrarenal aorta
were highly correlated (R2 ⫽ 0.99, P ⫽ 2.10⫺6) and had
(1 cm above the aortic bifurcation) with the subject
a unitary slope (slope ⫽ 0.97 ⫾ 0.05) over a wide veloc-
supine on an MR trolley bed outside the MR scan room.
ity range (0 –3.3 m/second).
The maximum velocity in each vessel was measured
after angle correction using built-in calipers. Three
measurements from each vessel were averaged for data Qualitative Comparison of Velocity Waveforms
analysis. Velocity waveforms were also stored for later The validated FVE method was then applied to human
qualitative analysis. volunteers. Representative velocity waveforms acquired
Each subject was then transferred into the MR scan- using spectral Doppler ultrasound and FVE are pre-
ner and velocity measurements were performed using sented in Figs. 5 and 6. These figures correspond to
the PC and FVE methods. Imaging planes were pre- measurements in the suprarenal aorta and inferior
scribed to target the same vessel locations previously vena cava, respectively.
targeted by Doppler ultrasound. Data were acquired Qualitatively, the velocity waveforms acquired using
during quiet breathing to ensure the MR and ultra- the two methods were similar. In the suprarenal aorta,
sound measurements were performed under similar waveforms from both methods exhibited systolic and
physiologic conditions. The body coil was used for all diastolic forward flows separated by a brief period of
MR measurements. retrograde flow (Fig. 5). The relative timing and ampli-
The PC measurements were performed using a seg- tude of the peaks were also consistent between these
mented k-space acquisition (17) with the following pa- waveforms. The main difference between the two meth-
rameters: flip angle ⫽ 20°, slice thickness ⫽ 6 mm, ods was that the spectrum of flow velocities displayed
matrix ⫽ 256 ⫻ 160, number of averages ⫽ 2, band- on FVE was consistently narrower than on ultrasound,
width ⫽ ⫾ 15.625 kHz, TE/TR ⫽ 6.3 msec/18 msec, resulting in clearer definition of peaks by real-time MR.
views-per-segment ⫽ 2, and cardiac phases recon- A similar agreement was obtained between measure-
structed ⫽ 20. Scans lasted approximately 160 sec- ments of venous flow in the inferior vena cava; however,
onds. Magnitude masking of the phase images was not the deflections of the triphasic flow pattern appeared
performed. Peak velocities were obtained by averaging better defined by FVE. As shown in Fig. 6, both methods
the velocities from an ROI centered manually over the depicted characteristic features of the venous waveform
In Vivo Fourier Velocity Encoding 301

Agreement between the waveforms from ultrasound

and PC MR was excellent in only three cases, good in
10, fair in five, and poor in six.

Quantitative Comparison of Peak Velocities

Presented in Fig. 7a are the in vivo peak velocities mea-
sured using FVE and PC MR. Similar to the in vitro
experiments, these measurements were strongly corre-
lated with the weighted fit having a unitary slope (R2 ⫽
0.85, P ⫽ 2.10⫺10, slope ⫽ 1.04 ⫾ 0.08). The Bland-
Altman analysis of these data presented in Fig. 7b
shows the lack of bias between the two measurements.
A second comparison is presented in Fig. 8 between
the MR and ultrasound velocity measurements. A
strong correlation was found between FVE and Doppler
ultrasound (R2 ⫽ 0.71, P ⫽ 2.10⫺7) and between PC MR
and Doppler ultrasound (R2 ⫽ 0.78, P ⫽ 1.10⫺8). The
peak velocities measured using ultrasound were
greater than those measured using either FVE or PC
MR, with weighted fitted slopes of 0.81 ⫾ 0.08 and
0.70 ⫾ 0.08, respectively. An average ultrasound in-
sonation angle of 51 ⫾ 10 (range 39 – 60°) was achieved
Figure 4. Comparison of peak velocities measured by FVE
during this study. With the accuracy of angular correc-
and PC MR using the in vitro flow apparatus. A strong corre-
lation between the two methods was observed (R2 ⫽ 0.99, P ⫽
tion approximately ⫾5°, a relative error of ⫾16% was
2.10⫺6, slope ⫽ 0.97 ⫾ 0.05). The uncertainty of the measure- present in the ultrasound measurement of peak veloc-
ments increased in proportion to the aliasing velocity chosen ity. These errors were included in the calculation of the
for each acquisition. weighted fits in Fig. 8 but are not displayed.

DISCUSSION
including the S-wave, D-wave, and a-wave. The respi-
ratory variation of the venous flow pattern was also This study presents the first quantitative in vivo evalu-
apparent. ation of real-time FVE. The strong agreement obtained
The PC MR flow waveform resolved fewer peaks than between these MR measurements and those of Doppler
either Doppler ultrasound or FVE, and those resolved ultrasound demonstrates the clinical potential of real-
peaks exhibited different relative amplitudes and tim- time MR for hemodynamic assessment and provides
ings compared with the other two methods. For this justification for future studies of pathologic flow using
reason, the flow curves produced by ultrasound and FVE, particularly in vessels difficult to access using
FVE compared more closely with each other than they ultrasound such as the pulmonary arteries and veins.
did with the curves produced by PC MR. The qualitative Such hemodynamic measurements will complement
agreement of the ultrasound and FVE waveforms was the anatomic and other physiologic information already
excellent in 14 cases, good in seven, and fair in three. provided by MR imaging.

Figure 5. Representative comparison

of the supra-renal aortic velocity wave-
form acquired in one subject using FVE
(a) and Doppler ultrasound (b). Wave-
forms exhibited similar shapes with
systolic and diastolic forward flows sep-
arated by a period of retrograde flow.
Note that data were collected separately
using each modality and then qualita-
tively aligned.
302 Macgowan et al.

Figure 6. Representative comparison of

the IVC velocity waveform acquired in a
volunteer using FVE (a) and Doppler ul-
trasound (b). Waveforms clearly showed
S-waves (systolic), D-waves (early dia-
stolic), and a-waves (atrial systolic) in
each cardiac cycle. Beat-to-beat fluctua-
tions in flow due to respiration were also
evident.

FVE measurements of peak velocity were shown to scanner and processed using an objective algorithm, as
correlate strongly with those from both ultrasound and was done by Steinman et al (21).
PC MR. The better agreement between the MR methods A sampling rate of 20 samples per heartbeat is gen-
was likely a result of the fact that both methods erally considered sufficient to visualize cardiac rhythms
scanned the same vessel location and the cardiovascu- (22). Therefore, the acquisition window used in this
lar physiology changed little between the two sequential study (30 msec per TR) should not introduce dramatic
MR measurements. The lower correlation between FVE temporal averaging for heart rates of at least 100 beats/
and Doppler ultrasound was attributed to differences in minute. At higher heart-rates, the aliasing velocity can
the anatomic locations of the measurements, potential be increased by narrowing each trapezoid in the read-
changes in hemodynamic state, and the different out gradient instead. As explained in the Methods sec-
sources of random error between the two modalities tion, this makes optimal use of the gradient hardware
(e.g., constrained insonation angle affects ultrasound and improves temporal resolution but requires the
accuracy). post-processing parameters to be measured separately
The shape of the sample volume from which ultra- at each aliasing velocity. This complication has moti-
sound velocity measurements were acquired also dif- vated the future development of automated methods to
fered from that of the FVE measurements, because of calibrate the data-processing parameters (i.e., k-space
fundamental differences between the two modalities. trajectory) on a per-acquisition basis. However, such
The ultrasound sample volume was defined by the pro- optimization will only improve the method and does not
file of the ultrasound beam and the gating of the re- detract from the presented results.
corded signal, while the MR sample volume was deter- Over the past decade, several advances have been
mined by the radiofrequency pulse and the spatial made in the field of real-time flow assessment by MR.
resolution of the MR acquisition. Because the velocity Real-time flow imaging has been developed by different
measurements depend on the intersection of the sam- groups to visualize flow in large vessels and the heart
ple volume and the velocity profile of the vessel, the two using rapid MR imaging methods at low spatial resolu-
methods could produce differing velocity waveforms tion (e.g., echo-planar or spiral imaging at approxi-
(18). Despite this potential difference, the measured mately 2.5–3.5 mm in-plane resolution) to achieve tem-
waveforms showed excellent qualitative agreement in poral resolutions of approximately 50 –180 msec
most cases. (5,6,23). While these methods may have sufficient res-
Absolute velocities measured using MR were consis- olution to acquire volumetric flows through major ves-
tently lower than those from Doppler ultrasound. This sels, they are limited by intra-voxel velocity averaging
bias was attributed to the known sources of velocity when velocities vary rapidly across a vessel or valve
overestimation by ultrasound related to spectral broad- (e.g., stenotic jet). Because the FVE method is not lim-
ening at large insonation angles and the Doppler gain ited by intra-voxel averaging, it may provide more ac-
settings. For example, a 40% overestimation of velocity curate and reproducible peak velocities under such
has been reported by Winkler and Wu (19) due to in- conditions. The performance of the FVE method under
trinsic spectral broadening at a 60° insonation angle, more complex flow conditions is the subject of on-going
while Hoskins (20) found peak velocity overestimated work at our institution.
by an average of 18% using standard imaging parame- Other real-time MR methods have also been devel-
ters and clinical ultrasound systems. More consistent oped to assess flow in one vessel at a time, similar to
ultrasound values for peak velocity may be expected if FVE, and have been used to measure hemodynamic
the Doppler spectra could be extracted from the clinical properties such as the velocity spectrum, flow acceler-
In Vivo Fourier Velocity Encoding 303

ation, or pulse-wave velocity (24 –26). In practice, these

MR “tagging” methods are less sensitive to flow com-
pared with FVE, but are also less susceptible to velocity
errors resulting from eddy-currents and magnetic-field
inhomogeneities within the MR scanner. To reduce ed-
dy-current and susceptibility errors in the FVE mea-
surement, several preventative steps were taken in our
study: the magnetic field was shimmed before each FVE
measurement and signal from static tissue surround-
ing each vessel was measured to define zero velocity. In
the future, such steps could be performed automati-
cally by the real-time interface at the start of each ac-
quisition.
The FVE measurement has several advantages over
cardiac-gated PC MR. For example, beat-to-beat varia-
tions in flow can be depicted by the real-time method,
while such variations are averaged together and lost by

Figure 8. Quantitative comparison between MR and Doppler

ultrasound velocity measurements in vivo. Peak systolic veloc-
ity measured using the FVE (circles) and PC MR (triangles)
methods plotted vs. the peak systolic velocities measured us-
ing Doppler ultrasound. High correlations were observed be-
tween the MR and ultrasound methods (FVE: R2 ⫽ 0.71, P ⫽
2.10⫺7; PC MR: R2 ⫽ 0.78, P ⫽ 1.10⫺8) with linear-regression
slopes of FVE ⫽ 0.81 ⫾ 0.08 (solid line) and PC MR ⫽ 0.70 ⫾
0.08 (dashed line).

gated PC MR. This feature of FVE allows for the assess-

ment of short-term changes related to interventional or
pharmacologic stress, or during physiologic maneuvers
like the Valsalva maneuver or natural breathing. In
addition, scan times are shorter with the real-time
method and respiratory compensation is not needed.
By allowing the subject to breathe freely during the
study, a more natural cardiovascular state is main-
tained and velocity measurements better represent
their normal values. Because volumetric-flow measure-
ments still require PC MR imaging, however, the real-
time MR method should be considered a component of
a complete MR protocol to assess blood flow.
In summary, real-time MR assessment of blood-flow
velocity correlates well with spectral Doppler ultra-
sound. Unlike conventional PC MR, real-time MR meth-
ods do not require cardiac triggering or respiratory
compensation. Such new methods may allow hemody-
namic information to be acquired in vessels inaccessi-
ble to ultrasound or in patients for whom respiratory
compensation is not possible, and can be used to mon-
itor flow changes during therapy.

REFERENCES
Figure 7. Quantitative comparison between FVE and PC MR
velocity measurements in vivo. a: Peak systolic velocities mea- 1. Lotz J, Meier C, Leppert A, Galanski M. Cardiovascular flow mea-
sured using the FVE method vs. using PC MR. b: The corre- surement with phase-contrast MR imaging: basic facts and imple-
mentation. Radiographics 2002;22:651– 671.
sponding Bland-Altman analysis. As observed during the in
2. Pelc LR, Pelc NJ, Rayhill SC, et al. Arterial and venous blood flow:
vitro evaluation, the velocities measured by the two methods
noninvasive quantitation with MR imaging. Radiology 1992;185:
were highly correlated (R2 ⫽ 0.87, P ⫽ 2.10⫺10) and statisti- 809 – 812.
cally equivalent (fitted slope ⫽ 1.04 ⫾ 0.08). No bias between 3. Powell AJ, Geva T. Blood flow measurement by magnetic resonance
the two measurements was observed in the Bland-Altman imaging in congenital heart disease. Pediatr Cardiol 2000;21:47–
analysis (mean ⫽ – 4 cm/second, 2 䡠 SD ⫽ 23 cm/second). 58.
304 Macgowan et al.

4. Kondo C, Caputo GR, Semelka R, Foster E, Shimakawa A, Higgins 16. Farb RI, McGregor C, Kim JK, et al. Intracranial arteriovenous
CB. Right and left ventricular stroke volume measurements with malformations: real-time auto-triggered elliptic centric-ordered 3D
velocity-encoded cine MR imaging: in vitro and in vivo validation. gadolinium-enhanced MR angiography—initial assessment. Radi-
AJR Am J Roentgenol 1991;157:9 –16. ology 2001;220:244 –251.
5. Wetzel SG, Lee VS, Tan AG, et al. Real-time interactive duplex MR 17. Pelc NJ, Herfkens RJ, Shimakawa A, Enzmann DR. Phase contrast
measurements: application in neurovascular imaging. AJR Am J cine magnetic resonance imaging. Magn Reson Q 1991;7:229 –254.
Roentgenol 2001;177:703–707. 18. Macgowan CK, Wood ML. Fast measurements of the motion and
6. Nayak KS, Pauly JM, Kerr AB, Hu BS, Nishimura DG. Real-time velocity spectrum of blood using MR tagging. Magn Reson Med
color flow MRI. Magn Reson Med 2000;43:251–258. 2001;45:461– 469.
7. Holen J, Aaslid R, Landmark K, Simonsen S. Determination of 19. Winkler AJ, Wu J. Correction of intrinsic spectral broadening er-
pressure gradient in mitral stenosis with a non-invasive ultrasound rors in Doppler peak velocity measurements made with phased
Doppler technique. Acta Med Scand 1976;199:455– 460. sector and linear array transducers. Ultrasound Med Biol 1995;21:
8. Pat GT, Pauly JM, Hu BS, Nishimura DG. One-shot spatially re- 1029 –1035.
solved velocity imaging. Magn Reson Med 1998;40:603– 613. 20. Hoskins PR. Accuracy of maximum velocity estimates made using
9. Hu BS, Pauly JM, Nishimura DG. Localized real-time velocity spec- Doppler ultrasound systems. Br J Radiol 1996;69:172–177.
tra determination. Magn Reson Med 1993;30:393–398. 21. Steinman AH, Tavakkoli J, Myers Jr JG, Cobbold RS, Johnston
10. Earls JP, Ho VB, Foo TK, Castillo E, Flamm SD. Cardiac MRI: KW. Sources of error in maximum velocity estimation using linear
recent progress and continued challenges. J Magn Reson Imaging phased-array Doppler systems with steady flow. Ultrasound Med
2002;16:111–127. Biol 2001;27:655– 664.
11. Hardy CJ, Cline HE. Broadband nuclear magnetic resonance 22. McDonald DA, Nichols WW, O’Rourke MF. McDonald’s blood flow
pulses with two-dimensional spatial selectivity. J AppL Phys 1989; in arteries: theoretic, experimental, and clinical principles. Phila-
66:1513–1516. delphia: Lea & Febiger; 1990. xviii, 456 p.
12. Pauly JM, Nishimura DG, Macovski A. A k-space analysis of small- 23. Nayak KS, Hu BS. Triggered real-time MRI and cardiac applica-
tip-angle excitation. J Magn Reson 1989;81:43–56. tions. Magn Reson Med 2003;49:188 –192.
13. Hashemi, Bradley WG. MRI: the basics. Bradley WG, editor. New 24. Itskovich VV, Kraft KA, Fei DY. Rapid aortic wave velocity measure-
York: Lippincott, Williams & Wilkins; 1997. 307 p. ment with MR imaging. Radiology 2001;219:551–557.
14. Takahashi A, Peters T. Compensation of multi-dimensional selec- 25. Macgowan CK, Wood ML. Motion measurements from individual
tive excitation pulses using measured k-space trajectories. Magn MR signals using volume localization. J Magn Reson Imaging 1999;
Reson Med 1995;34:446 – 456. 9:670 – 678.
15. Noll DC, Nishimura DG, Macovski A. Homodyne detection in mag- 26. Macgowan CK, Henkelman RM, Wood ML. Pulse-wave velocity
netic resonance imaging. IEEE Trans Med Imaging 1991;10:154 – measured in one heartbeat using MR tagging. Magn Reson Med
163. 2002;48:115–121.