Ultrasound in Med. & Biol., Vol. 27, No. 4, pp.

551–564, 2001
Copyright © 2001 World Federation for Ultrasound in Medicine & Biology
Printed in the USA. All rights reserved
0301-5629/01/$–see front matter

PII: S0301-5629(01)00357-X

● Original Contribution

EFFECTS OF TRANSDUCER, VELOCITY, DOPPLER ANGLE, AND

INSTRUMENT SETTINGS ON THE ACCURACY OF COLOR DOPPLER
ULTRASOUND

SANDY F. C. STEWART
Hydrodynamics and Acoustics Branch, Center for Devices and Radiological Health, Food and Drug Administration,
Rockville, MD, USA

(Received 7 August 2000; in final form 22 January 2001)

Abstract—The accuracy of a commercial color Doppler ultrasound (US) system was assessed in vitro using a
rotating torus phantom. The phantom consisted of a thin rubber tube filled with a blood-mimicking fluid, joined
at the ends to form a torus. The torus was mounted on a disk suspended in water, and rotated at constant speeds
by a motor. The torus fluid was shown in a previous study to rotate as a solid body, so that the actual fluid velocity
was dependent only on the motor speed and sample volume radius. The fluid velocity could, thus, be easily
compared to the color Doppler-derived velocity. The effects of instrument settings, velocity and the Doppler angle
was assessed in four transducers: a 2.0-MHz phased-array transducer designed for cardiac use, a 4.0-MHz
curved-array transducer designed for general thoracic use, and two linear transducers designed for vascular use
(one 4.0 MHz and one 6.0 MHz). The color Doppler accuracy was found to be significantly dependent on the
transducer used, the pulse-repetition frequency and wall-filter frequency, the actual fluid velocity and the
Doppler angle (p < 0.001 by analysis of variance). In particular, the phased array and curved array were
observed to be significantly more accurate than the two linear arrays. The torus phantom was found to provide
a sensitive measure of color Doppler accuracy. Clinicians need to be aware of these effects when performing color
Doppler US exams. (E-mail: sxs@cdrh.fda.gov) © 2001 World Federation for Ultrasound in Medicine &
Biology.

Key Words: Color Doppler accuracy, Doppler ultrasound phantom, Color Doppler ultrasound.

INTRODUCTION assume that color Doppler systems are accurate, or may

The assessment of color Doppler accuracy of clinical
ultrasound (US) systems is an important, but neglected,
topic. Given the enormous number of exams performed
annually worldwide, color Doppler accuracy should be
given more attention than it is. One of the impediments
to the assessment of Doppler accuracy is the lack of
generally accepted Doppler accuracy phantoms. Color
Doppler, in particular, suffers from the lack of a suitable
technique for assessing accuracy. Because color Doppler
is generally assumed to be used qualitatively, a common
belief may be that it doesn’t need to be assessed for
accuracy. However, there are at least two arguments in
favor of the development and use of color Doppler ac-
curacy standards: 1. numerous quantitative uses for color
Doppler US have been developed, 2. clinicians may
Address correspondence to: Sandy F. C. Stewart, Ph.D., Hydro-
dynamics and Acoustics Branch, Food and Drug Administration, 9200
Corporate Blvd. (HFZ-132), Rockville, MD 20850 USA. E-mail:
sxs@cdrh.fda.gov
interpret color images in a quantitative or semiquantita-
tive fashion, either consciously or subconsciously, pos-
sibly leading to diagnostic errors. This latter effect may
be more likely to show up in serial exams, where data
from one examination are compared against that from
another. Although reproducibility is often more impor-
tant than absolute accuracy in serial exams, differences
in color images due to variations with instrument settings
or transducer used may, nevertheless, be interpreted as
having substantive diagnostic meaning.
Numerous methods have been previously described
for assessing the accuracy of Doppler in commercial US
systems. String phantoms have been used extensively for
calibrating pulsed Doppler (Walker et al. 1982; Phillips
et al. 1990). A phantom using a moving belt of foam was
developed for assessing color Doppler (Rickey et al.
1992). A rotating rubber disk has also been used to assess
Doppler accuracy (McDicken et al. 1983; Fleming et al.
1994; Schwarz et al. 1995). However, string phantoms

551
552 Ultrasound in Medicine and Biology Volume 27, Number 4, 2001

are not particularly suitable for 2-D color imaging, be-

cause the moving string forms a stationary, narrow color
image that cannot be easily interpreted. Furthermore,
these devices use moving solids that may not produce a
realistic backscatter signal.
Phantoms that use parabolic fluid flow in a tube
(McDicken 1986) or between parallel plates (Boote and
Zagzebski 1988) have some advantages. These phantoms
use a blood-mimicking fluid instead of a solid and pro-
vide a more physiologic model of arterial flow. These
have been used to calibrate volume flow measurements
(Picot et al. 1995; Forsberg et al. 1995; Holland et al.
1996) and to derive the velocity profile (Hein and
O’Brien 1992). The associated velocity gradients are
high, however, so that the fluid velocity is a strong
function of position. The calibration of Doppler US Fig. 1. Torus phantom for assessing color Doppler accuracy.
requires precise positioning of the US sample volume,
combined with an accurate evaluation of the velocity
profile. Errors in measuring the volume flow rate, corre- effects of instrument settings, Doppler angle and velocity
lating the volume flow to the velocity profile, and ne- magnitude were also investigated.
glecting entrance length effects limit the usefulness of
tube phantoms. In addition, turbulence alters the velocity METHODS
profile, even at moderate flow rates; thus, invalidating
Torus phantom
the known relationship between velocity and radius and
The torus phantom used was described previously
limiting the useful velocity range of the phantom.
(Stewart 1999). Briefly, the phantom was made from a
An accuracy phantom designed for color Doppler
bicycle inner tube measuring 0.064 cm in thickness and
was recently described that overcomes some of the lim-
3.0 cm in diameter. The tube was filled with a blood-
itations of previously available phantoms (Stewart 1999).
mimicking fluid optimized for US studies that had an
The torus phantom moves a blood-mimicking fluid at a
acoustic velocity equal to 1548 ⫾ 5 m/s (R.G. Shelley
known rate with a low radial velocity gradient, making it
Ltd., North York, Ontario, Canada) (Rickey et al. 1995).
ideal for color Doppler tests. At a velocity of 115.2 cm/s The ends were clamped onto a short cylinder to form a
(equivalent to 40 revolutions/min), the velocity gradient torus, which was then glued onto the rim of a 52-cm
perpendicular to the plane of the color Doppler image is diameter disk (Fig. 1). The torus had a minor radius of
approximately 4.2 s⫺1 (Stewart 1999). This is substan- 3.0 cm and a major radius of approximately 28 cm. The
tially less than the velocity gradient at the wall of a 1-cm torus was submerged in degassed water with the axis
diameter tube phantom at a peak velocity of 115.2 cm/s, aligned vertically, and rotated by a gearmotor controlled
which is calculated to be 460.8 s⫺1. Spectral broadening by a digital velocity feedback system. Numerical simu-
of the Doppler signal due to the velocity gradient is, thus, lations and flow visualization studies have shown that the
minimized. The torus phantom also benefits from a high fluid within such a rotating torus spins as a solid body at
maximum velocity, around 300 cm/s in the current de- the same angular velocity as the torus, after about 5 to
sign. The phantom was shown to provide a sensitive 10 s acceleration time (Stewart 1999). The circumferen-
measure of color Doppler accuracy in tests of a commer- tial velocity of the fluid (calculated as the sample volume
cial US system, using a single transducer designed for radius times the torus angular velocity) was the standard
vascular use. Significant effects of some of the system’s against which the Doppler-derived velocities were com-
instrument settings on Doppler accuracy were reported pared. In this paper, “accuracy” is defined as the ratio of
(Stewart 1999). the Doppler-derived velocity to the actual fluid velocity.
Manufacturers of US systems typically provide a
variety of transducers optimized for different functions, Doppler ultrasound system
such as vascular, cardiac and obstetrical examinations. An ATL Ultramark 9 HDI US system was used for
The current investigation was designed to assess the all laboratory experiments. The four different transducers
effect of the choice of transducer on the color Doppler tested are listed in Table 1. The transducer under test was
accuracy. Four transducers were examined, including mounted in a four-axis positioning system (three linear
two linear arrays, a curved array, and a phased array. The and one rotational, the rotational axis being parallel to
 Color Doppler accuracy ● S. F. C. STEWART 553

Table 1. Transducers used in study. varied considerably with PRF. However, an earlier study
showed that color Doppler accuracy did not vary sub-
Transmit frequency
Transducer Scanhead type (MHz) stantially with the power output, beyond a threshold level
(Stewart 1999).
P3-2 Phased array 2.0 In the second series of experiments, the study was
C5-40R Curved array 4.0
L5 Linear array 4.0 repeated with the Doppler angle set to 70°. PRF and
L10-5 Linear array 6.0 WFF settings had to be adjusted to take into account the
lower perceived velocity at this higher Doppler angle. In
the third series of experiments, the Doppler angle was set
to 55°. All the velocities were used with the L10-5
the torus major radius). The active face was suspended transducer at 55°. Only one fluid velocity was used for
approximately 1.0 to 2.0 cm above the torus, with the US the P3-2, C5-40R, and L5 transducers, corresponding to
beam focused on the center of torus fluid. The degassed an intermediate torus speed of 40 revolutions/min.
water bath in which the torus was immersed provided a The color box was centered within the torus tube
path for the US beam between the transducer and the (visible in the 2-D imaging B-mode) and positioned 4 ⫾
torus. The circumferential velocity of the moving fluid 2 cm from the transducer face. A single focal zone was
was measured by Doppler and compared to the velocity always used, and was positioned as close as possible to
calculated by the motor speed multiplied by the radial the middle of the color Doppler box. In all cases, the
position of the sample volume. The radial position was color baseline was maximally shifted downward, which
calibrated by focusing the transducer on the gap between effectively doubled the Nyquist limit and allowed a
two sound-absorbing rectangles glued to the torus higher maximum velocity to be measured.
mounting disk (Stewart 1999), which was a known dis-
tance from the center of the torus. By using the position- Color Doppler image processing and data analysis
ing system, the radial position of the sample volume Images from the US system’s color video output
within the torus could be accurately set. were captured using a Flashpoint 128 video frame grab-
ber (Integral Technologies, Indianapolis, IN) mounted in
Color Doppler experiments a personal computer running Windows 95. Images cap-
Three series of experiments were performed for tured with a resolution of 640 ⫻ 480 pixels and color
each of the four transducers. In the first series, the effects depth of 24 bits (i.e., 8 bits per color red, green and blue)
of the pulse-repetition frequency (PRF), wall-filter fre- were saved onto hard disk in Windows bitmap (BMP)
quency (WFF), and fluid velocity on Doppler accuracy image file format. A custom color/velocity map on the
were investigated at a Doppler angle of 40°. Tests were US system was used to encode velocities in the red
run at fluid velocities equal to 28.8, 57.6, 115.2, 172.8, image plane, using a linear velocity-to-color translation
230.4 and 288.0 cm/s, corresponding to torus rotational scheme. Red byte values varied from 0 to 255, with 0
speeds of 10, 20, 40, 60, 80 and 100 revolutions/min. The corresponding to zero velocity and 255 corresponding to
other color instrument settings, including color gain, the maximum velocity. Green was added to make the
output power, color sensitivity, color persistence and color maps more visible, with pixel values of 0 and 197
dynamic motion differentiation, were fixed between ex- corresponding to half the maximum velocity and the
periments (Table 2). The spatial peak temporal average maximum velocity, respectively. The green pixel values
intensity (derated), or ISPTA.3, averaged 20 mW/cm2, but were ignored when calibrating the velocity/color map or
making measurements.
After an image was acquired, the velocity-color
Table 2. ATL Ultramark 9 HDI nominal instrument settings calibration was done. A simple linear regression was
used in color Doppler accuracy tests. performed between the color and velocity, using the
pixels within the color bar. Two assumptions were made:
Instrument parameter Setting used 1. the velocity was linear with position down the color
HDI/high frequency on/on bar, and 2. the minimum velocity corresponded to black
Variance off and maximum velocity corresponded to the maximum
Output power ISPTA.3 ⫽ 20 mW/cm2, MI ⫽ 0.5 color in the color bar (so that the y-intercept was near
Color gain 94%
Color sensitivity Maximum ⫽ 16 zero). Each image was separately calibrated.
Color persistence Minimum ⫽ 0 Image files stored on disk were analyzed by an
Dynamic motion differentiationTM Off analysis program written in Borland Delphi version 3.0
ISPTA.3 ⫽ spatial peak temporal average intensity, derated; MI ⫽ (Inprise Corp., Scotts Valley, CA). After an image was
mechanical index. read into this program, the color bar was calibrated
554 Ultrasound in Medicine and Biology
Fig. 2. Average of 520 color Doppler images using P3-2
phased-array transducer at a fluid velocity of 115.2 cm/s and
Doppler angle of 70°. The upper and lower red lines have been
fitted to upper and lower walls of the torus tube. The middle red
line (arrow) is the average of the two and is the assumed
direction of fluid flow.
against the maximum velocity corresponding to the PRF.
The velocities were then corrected for the Doppler angle
␾ between the fluid velocity and the transducer beam, by
dividing the velocity by cos(␾).
In most cases, the transducers were set at an oblique
angle to the torus tube (as in Fig. 2), so that the direction
of the fluid velocity had to be determined from the
B-mode image of the tube walls. All the images for a
Fig. 3. Color Doppler image using P3-2 phased-array trans-
ducer. Yellow cross labeled “O” at the top indicates center
defined by radii of ultrasound beam. Straight yellow line is
direction of fluid flow (equivalent to middle red line in Fig. 2).
Green sector outline delimits pixels to be processed. Only those
pixels pi tinted in purple have a Doppler angle ␾i ⫽ ␾e ⫾ 1°,
where ␾e ⫽ 70° in this case. Fluid velocity ⫽ 115.2 cm/s.
Volume 27, Number 4, 2001
Fig. 4. Color Doppler image using C5-40R curved array. Green
line is an arc of the circle fitted to curved face of transducer.
The center of the circle defines the radii along which the US is
radiated. Straight yellow line is direction of fluid flow (equiv-
alent to middle red line in Fig. 2). Green sector outline delimits
pixels to be processed. Only those pixels pi tinted in purple
have a Doppler angle ␾i ⫽ ␾e ⫾ 1°, where ␾e ⫽ 70° in this
case. Fluid velocity ⫽ 115.2 cm/s.
given transducer, Doppler angle and velocity were aver-
aged, and a line was fitted to the grey-scale images of the
averaged top tube wall (Fig. 2) as follows. A narrow
rectangle (yellow box in Fig. 2) was drawn manually
around the image of the walls to tell the program which
pixels to include. Then linear regression was performed
between the x and y positions of each pixel within the
yellow rectangle, with a weighting factor equal to the
intensity of the pixel at that point. This gave a straight
line that was the best fit to the tube wall. The procedure
Fig. 5. Color Doppler image using L5 linear array. Direction of
fluid flow is horizontal. Doppler angle ⫽ 70°. Fluid velocity ⫽
115.2 cm/s.
 Color Doppler accuracy ● S. F. C. STEWART
was then repeated for the bottom wall. The top and
bottom red lines in Fig. 2 are the fitted lines, and the
middle red line (arrow) has the average slope and inter-
cept of the top and bottom line, and is assumed to be the
direction of the fluid velocity.
The curved-array and phased-array transducers
emitted the US beam along radii of a circle, so that the
Doppler angle ␾ was a function of position (an effect that
is readily visible in Fig. 3). For the phased-array trans-
ducer, the center C defined by the US radii was marked
manually by a yellow cross (labeled “O” at the top of
Fig. 3). For the curved array, the center C was found by
fitting a circle to the arc indicating the transducer surface
(Fig. 4). The fitted circle is marked in green in Fig. 4; in
this case the calculated center C is beyond the visible part
of the image. In Figs. 3 and 4, the yellow line between
cross marks is the center of the tube (and estimated
direction of the fluid velocity) found by the curve fitting
procedure described above (Fig. 2). For the curved- and
phased-array transducers, the beam steering angle was
always set within the range of 5 to 20.
After the center of radii was found in the phased-
array and curved-array transducers, a sector of a circle
was drawn within the color box (defined by two radii and
two arcs, seen as green lines in Figs. 3 and 4). The
direction of the US beam for each pixel pi within the
green box was determined by first computing the line
between pi and the center C. The Doppler angle ␾i for
pixel pi was then calculated from the angle between the
line defining the US beam for that pixel and the line
defining the fluid velocity. Only those pixels pi with
angles ␾i ⫽ ␾e ⫾ 1 were analyzed further, where ␾e ⫽
angle for a particular experiment (␾e ⫽ 40°, 55° or 70°).
The velocity at pi was divided by cos(␾i), and the cor-
rected velocities in the selected range of angles were then
averaged. In Figs. 3 and 4, the locus of points with 69.0°
ⱕ ␾i ⱕ 71.0° are indicated by the purple tint.
For the linear arrays, the Doppler angle ␾ was
constant over the entire color image, because the US
beam was assumed to be emitted along parallel lines
(Fig. 5). The direction of the US beam was determined
from the angled side of the color box. When testing the
linear arrays at 70°, the transducers were placed perpen-
dicular to the torus wall, and the US beam was steered 20
from vertical (as in Fig. 5). The direction of the fluid
velocity, in this case, was assumed to be horizontal. In
the case of the other two Doppler angles (55° and 40°),
the beam was steered 20 from vertical, and the probe was
then tilted an extra 15° or 30° with respect to the torus
tube. In these latter two cases, the direction of fluid flow
was calculated by fitting lines to the walls of the torus
tube in an averaged image (as in Fig. 2). The Doppler
angle ␾ between the fluid velocity and the direction of
the US beam was calculated. A graphic parallelogram
555
Fig. 6. Pulsed Doppler spectrum for P3-2 transducer at a
Doppler angle of 40°. Fluid velocity ⫽ 115.2 cm/s. ROI is
outlined by rectangular box.
was drawn within the color box, and the color-encoded
velocities within the parallelogram were divided by
cos(␾) and averaged.
The Doppler-derived velocity was also dependent
on the variable angle between the plane of the US beam
and the torus itself, due to the mild curve of the torus
major radius. However, this effect was assumed to be
negligible (Gill 1985; Stewart 1999).
Pulsed Doppler comparisons
Measurements were also taken using the P3-2 and
L10-5 transducers in pulsed Doppler mode, to provide a
comparison to the color Doppler accuracy. The Doppler
angle was set to 40° or 70°, using the same transducer tilt
and beam steering angles as were used in the color
Doppler experiments. The ATL’s internal angle correc-
tion was used. The pulsed Doppler sample volume was
set to 2.5 mm, and the velocity, PRF and WFF were
varied as in the color Doppler studies. The output and
gain were set to avoid saturating the image. Spectral
images were downloaded to the personal computer and
the mean velocity was calculated off-line. The velocity
was first calibrated to the vertical direction. Then, at each
point in time ti in the spectrum, a mean velocity vi was
calculated, using the pixel intensity as a weight factor.
Finally an overall mean (⫾ SD) velocity Vm was calcu-
lated by averaging the vi over time. A typical spectrum
for the P3-2 transducer at a Doppler angle of 40° and
velocity of 115.2 cm/s is shown in Fig. 6. The region-
of-interest (ROI) is outlined by the rectangular box. The
spikes are reflections from the short cylinder onto which
the ends of the torus tube are clamped, and were ex-
cluded from the velocity measurements.
556 Ultrasound in Medicine and Biology Volume 27, Number 4, 2001

Fig. 7. Color Doppler-derived velocity as a function of transducer at a Doppler angle of 40°. (a) P3-2 phased array; (b)
L5 linear array; (c) C5-40R curved array; (d) L10-5 linear array. Fluid velocity ⫽ 115.2 cm/s. Effects of PRF and WFF
can also be seen.

Statistical analysis A sample size of 10 images were analyzed for each

The color Doppler accuracy was evaluated statisti-
cally by analysis of variance, using a general linear
model approach that automatically accounted for the
unbalanced experimental design. The dependent variable
was the accuracy (i.e., the ratio of the color Doppler-
derived velocity to the actual fluid velocity). The inde-
pendent variables were the transducer, Doppler angle,
PRF, WFF and actual fluid velocity. All independent
variables had fixed levels, so no covariates were used.
Interaction terms among certain of the independent vari-
ables were included in the model: (transducer ⫻ Doppler
angle), (transducer ⫻ WFF), (transducer ⫻ actual fluid
velocity), (Doppler angle ⫻ WFF), and (Doppler an-
gle ⫻ actual fluid velocity). Higher interaction terms and
interaction terms involving PRF could not be used, be-
cause of disconnects within the data set (e.g., not all
levels of PRF could be used with all levels of the other
variables). Bonferroni pair-wise comparisons were made
among the levels of the five independent variables to
determine the significance of differences among levels.
combination of transducer, velocity, Doppler angle, WFF
and PRF. Statistical analysis was performed on a per-
sonal computer using Minitab release 12 (Minitab, State
College, PA).
RESULTS
Effects of transducer
The color Doppler-derived velocity as a function of
transducer at a fluid velocity of 115.2 cm/s and a Doppler
angle of 40° is shown in Fig. 7. The P3-2 phased-array
and C5-40R curved-array transducers had average accu-
racies (0.911 ⫾ 0.066 and 0.905 ⫾ 0.044, respectively;
mean ⫾ SD) that were higher than that of the two linear
arrays (0.842 ⫾ 0.046 and 0.828 ⫾ 0.065 for the L5 and
L10-5 transducer, respectively) as indicated in Table 3.
Differences in the range of PRF used reflect the different
transmission frequencies among the four transducers
used (Table 1).
The color Doppler-derived velocity as a function of
 Color Doppler accuracy ● S. F. C. STEWART 557

transducer at the same velocity, but at 70° is shown in

0.915 ⫾ 0.136

0.907 ⫾ 0.079

0.808 ⫾ 0.077

0.826 ⫾ 0.092
All velocities
and angles
Fig. 8. Again, the P3-2 phased-array and C5-40R curved-
array transducers had higher average accuracies
(0.911 ⫾ 0.116 and 0.905 ⫾ 0.093, respectively) than the
two linear arrays (0.771 ⫾ 0.068 and 0.817 ⫾ 0.078,
respectively, Table 3).
The effect of transducer was statistically significant,

0.915 ⫾ 0.086
0.940 ⫾ 0.059
0.912 ⫾ 0.169
0.904 ⫾ 0.058
0.860 ⫾ 0.091
0.918 ⫾ 0.085
0.849 ⫾ 0.062
0.812 ⫾ 0.064
0.763 ⫾ 0.055
0.829 ⫾ 0.072
0.838 ⫾ 0.073
0.815 ⫾ 0.111
All velocities
at p ⬍ 0.001. The Bonferroni pair-wise comparisons
showed that each transducer was significantly different
from all the others, with an adjusted p ⬍ 0.0001.

Effects of Doppler angle

0.976 ⫾ 0.120

0.760 ⫾ 0.094

0.961 ⫾ 0.068
0.892 ⫾ 0.039

0.768 ⫾ 0.053

0.871 ⫾ 0.029
0.845 ⫾ 0.052
The effect of the Doppler angle is shown for the

288.0
P3-2 phased-array and the L10-5 linear-array transducer

–
†

†
at 115.2 cm/s in Fig. 9. For the phased-array transducer,
the average accuracies at 40°, 55° and 70° were equal to
0.911 ⫾ 0.066, 0.940 ⫾ 0.059, and 0.911 ⫾ 0.116,
respectively. For the L10-5 linear array, the average

0.972 ⫾ 0.054

0.864 ⫾ 0.105

0.938 ⫾ 0.065
0.852 ⫾ 0.030

0.773 ⫾ 0.060
0.873 ⫾ 0.041
0.822 ⫾ 0.092
0.835 ⫾ 0.092
Table 3. Summary of color Doppler accuracy tests.
accuracies at 40°, 55° and 70° were equal to 0.828 ⫾

230.4
0.065, 0.842 ⫾ 0.074, and 0.817 ⫾ 0.078, respectively.

–
†
The other two transducers showed effects of the Doppler
angle that were similar in magnitude (Table 3).
Note that, at 40°, the angled configuration of the
transducer occasionally did not allow the sample volume

0.922 ⫾ 0.061

0.906 ⫾ 0.111
0.918 ⫾ 0.043

0.897 ⫾ 0.069
0.828 ⫾ 0.060

0.782 ⫾ 0.059
0.856 ⫾ 0.070
0.847 ⫾ 0.052
0.840 ⫾ 0.062
depth to be short enough for a PRF high enough to
172.8
measure the higher velocities without aliasing (Table 3).

–
Fluid velocity (cm/s)

Therefore, some of the higher velocities were missing in

the C5-40R and L10-5 transducer experiments at 40°.
The effect of Doppler angle on the overall accuracy
was statistically significant, at p ⬍ 0.001, although there
0.911 ⫾ 0.066
0.940 ⫾ 0.059
0.911 ⫾ 0.116
0.905 ⫾ 0.044
0.860 ⫾ 0.091
0.905 ⫾ 0.093
0.842 ⫾ 0.046
0.812 ⫾ 0.064
0.771 ⫾ 0.068
0.828 ⫾ 0.065
0.842 ⫾ 0.074
0.817 ⫾ 0.078

* Color Doppler-derived velocity divided by the actual fluid velocity (mean ⫾ SD).
were no clear trends. An increase in the effect of PRF
115.2

and WFF with higher angles is evident, however (Fig. 9).

The pair-wise comparisons showed that each Doppler

PRF could not be set high enough to measure velocity without aliasing.
angle was significantly different from all the others, with
an adjusted p ⬍ 0.0001.
0.910 ⫾ 0.060

0.960 ⫾ 0.157
0.901 ⫾ 0.068

0.933 ⫾ 0.072
0.852 ⫾ 0.074

0.796 ⫾ 0.078
0.814 ⫾ 0.065
0.835 ⫾ 0.069
0.792 ⫾ 0.101

Effects of PRF and WFF

57.6

The color Doppler-derived velocity as a function of

velocity, PRF and WFF for the P3-2 phased-array trans-
ducer at a Doppler angle of 40° is shown in Fig. 10. At
low PRFs, there was a tendency for the Doppler-derived
velocity to decrease with increase in WFF and, at high
0.858 ⫾ 0.103*

0.857 ⫾ 0.067
1.005 ⫾ 0.271
0.899 ⫾ 0.060

0.896 ⫾ 0.107

0.763 ⫾ 0.110
0.831 ⫾ 0.081
0.832 ⫾ 0.082
0.801 ⫾ 0.177

PRFs, the tendency was reversed. At an angle of 70°

28.8

(Fig. 11), the increase in Doppler-derived velocity with

WFF was enhanced at high PRF, but the decrease in
Doppler-derived velocity with WFF was suppressed at
low PRF. Overall, the Doppler accuracy was found to be
0.915 ⫾ 0.086 and 0.912 ⫾ 0.169 at Doppler angles of
Doppler
angle, °

40
55
70
40
55
70
40
55
70
40
55
70

40° and 70°, respectively (Table 3). Combined effects of

PRF and WFF were observed in all transducers, and have
been reported previously (Stewart 1999).
Transducer

The effect of WFF was statistically significant (p ⬍

C5-40R

L10-5

0.001). Pair-wise comparisons showed that each level of

P3-2

L-5

WFF was statistically significantly different from all of

558 Ultrasound in Medicine and Biology Volume 27, Number 4, 2001

Fig. 8. Color Doppler-derived velocity as a function of transducer at a Doppler angle of 70°. (a) P3-2 phased array; (b)
L5 linear array; (c) C5-40R curved array; (d) L10-5 linear array. Fluid velocity ⫽ 115.2 cm/s. Effects of PRF and WFF
can also be seen.

the others with an adjusted p ⬍ 0.0001, except for the 50
Hz vs. 100 Hz and 50 Hz vs. 200 Hz comparisons, which
were not significant.
The effect of PRF was also statistically significant
(p ⬍ 0.001). The pair-wise comparisons showed that
most pairs of PRF were significantly different at p ⬍
0.0001, but some of the close pairs at the extremes of the
range (e.g., 0.4 kHz vs. 0.6 kHz at the low end, and 14
kHz vs. 16 kHz at the high end) were not significantly
different from each other. The only close pairs in the
middle of the range that were not significantly different
were the 4.0 kHz vs. 4.5 kHz and 7.0 kHz vs. 8.0 kHz
comparisons.
Effects of velocity
For the P3-2 transducer, the accuracy could be seen
to increase with increasing velocity at 40°, and decrease
with increasing velocity at 70° (Table 3). This effect of
velocity was not a clear trend because it was not repeated
for the other transducers. Nevertheless, the effect of
velocity was statistically significant, at p ⬍ 0.001. Pair-
wise comparisons showed that each level of velocity was
significantly different from all of the others, with an
adjusted p ⬍ 0.0001.
Effects of interactions
The analysis of variance showed that not only were
the effects of all the independent variables significant at
p ⬍ 0.001, but the five tested interaction terms were also
significant at this same p level. Thus, the effect of trans-
ducer was itself significantly dependent on the levels of
Doppler angle, WFF and actual fluid velocity. The effect
of Doppler angle was itself significantly dependent on
the levels of WFF and actual fluid velocity. More com-
plex interactions could not be investigated due to limi-
tations in the data set connectivity.
Pulsed Doppler results
The pulsed Doppler-derived velocity as a function
of velocity, PRF and WFF for the P3-2 transducer at a
Doppler angle of 40° is shown in Fig. 12. The accuracy
for pulsed Doppler at this Doppler angle averaged
 Color Doppler accuracy ● S. F. C. STEWART 559

Fig. 9. Color Doppler-derived velocity as a function of Doppler angle: (a) P3-2 phased array, 40°; (b) P3-2 phased array,
55°; (c) P3-2 phased array, 70°; (d) L10-5 linear array, 40°; (e) L10-5 linear array, 55°; (f) L10-5 linear array, 70°. Fluid
velocity ⫽ 115.2 cm/s.

1.023 ⫾ 0.032, which was closer to 1.0 than the corre-
sponding accuracy (0.911 ⫾ 0.066) for color Doppler
(Fig. 10). At a Doppler angle of 70°, the pulsed Doppler
accuracies averaged 1.012 ⫾ 0.116 for the P3-2 trans-
ducer. For the L10-5 transducer, the accuracies averaged
0.952 ⫾ 0.089 and 0.920 ⫾ 0.057 for Doppler angles of
40° and 70°, respectively. In all cases, the effects of PRF
and WFF were lower in the pulsed Doppler than in the
color Doppler case.
DISCUSSION
The measured color Doppler accuracy has been
shown to be dependent on all of the parameters varied in
this study: the transducer, Doppler angle, WFF, PRF and
velocity. In every case, the main effects were statistically
significant, at a confidence level of p ⬍ 0.001. Three
trends emerge. First, some transducers are markedly
more accurate than others. Although the effects of the
other variables mask the differences somewhat, the pair-
wise comparisons clearly showed the differences among
pairs of transducers to be significant. Second, the effect
of WFF and PRF formed a distinctive pattern. In general,
as the PRF increased, the estimated velocity first de-
creased, then increased with WFF. Third, the variations
in accuracy with velocity and Doppler angle were also
significant, but with much less of an observable trend.
The general linear models approach used in this
study can accommodate an unbalanced experimental de-
sign. This approach was necessary because not all values
of PRF and WFF were available at all velocities, Doppler
angle and transducer. In some cases, aliasing prevented
measurements from being taken at higher velocities. This
560 Ultrasound in Medicine and Biology
Fig. 10. Color Doppler-derived velocity as a function of velocity, PRF and WFF for the P3-2 phased-array transducer
at a Doppler angle of 40°: (a) 28.8 cm/s; (b) 57.6 cm/s; (c) 115.2 cm/s; (d) 172.8 cm/s; (e) 230.4 cm/s; (f) 288.0 cm/s.
flexibility was exploited to reduce the number of mea-
surements required by about 25%. Examination of the
L10-5 transducer data (taken at all three Doppler angles)
suggested that the 55° data could safely be omitted for
most of the velocities in the other three transducers.
Table 3 shows that, for the other transducers, experi-
ments were performed over the full velocity range at 70°
and 40°, but only at a single intermediate velocity at 55°.
The missing velocities were, thus, bracketed on either
side by velocities at 40° and 70°. Any missing data were
accounted for by the model, and the resulting p values
were automatically adjusted. Statistically significant ef-
fects were demonstrated despite the unbalanced design.
Effects of transducer
In color Doppler mode, all four transducers sig-
nificantly underestimated the actual fluid velocity.
This may result from bias in the mean-frequency es-
Volume 27, Number 4, 2001
timator, due to difficulty in filtering a Doppler signal
composed of only a few data points (Rickey and
Fenster 1996; Willemetz et al. 1989). The underesti-
mation has been shown to be less when using pulsed
Doppler (Fig. 12), so is probably not due to errors of
the torus phantom mechanism itself. An assessment of
the torus phantom error demonstrated it to be accurate
to within ⫾ 1.7% (Stewart 1999).
The two linear-array transducers were found to be
significantly less accurate than the phased-array and
curved-array transducers. This may be due to the
higher focusing in the linear arrays, which would
increase the Doppler bandwidth and make them more
susceptible to aliasing; thus, reducing the mean
estimated frequency and lowering the measured veloc-
ity.
Variations in the beam spread may be due to dif-
ferences of ⫾ 2 cm in the range of interest (position of
 Color Doppler accuracy ● S. F. C. STEWART 561

Fig. 11. Color Doppler-derived velocity as a function of velocity, PRF and WFF for the P3-2 phased-array transducer
at a Doppler angle of 70°: (a) 28.8 cm/s; (b) 57.6 cm/s; (c) 115.2 cm/s; (d) 172.8 cm/s; (e) 230.4 cm/s; (f) 288.0 cm/s.

the color box) or focal zone, resulting from the necessity
of preventing steric interference between the transducer
and torus tube. Some of the observed variation in accu-
racy may be due to this effect.
Effects of PRF and WFF
The effects of the PRF and WFF can be explained
by a linear signal analysis of the Doppler spectrum. If the
color baseline is centered, the wall filter acts as a notch
filter centered at zero. In this study, however, the color
baseline was maximally shifted, so that wall filter acts
like a broad bandpass filter dropping off at 0 kHz and at
the PRF. When the mean of the Doppler spectrum is near
the PRF (e.g., left end of Fig. 10a, low PRF setting), the
filter removes some of the higher frequencies, thus bias-
ing the mean frequency estimate downward. Increasing
the WFF removes more of the higher frequencies, thus
increasing the bias downward. When the mean of the
Doppler spectrum is near zero (e.g., right end of Fig. 10a,
high PRF setting), the wall filter removes some of the
lower frequencies, thus biasing the mean frequency es-
timate upward. Increasing the WFF removes more of the
low frequencies, so the bias upward increases. This
mechanism explains most of the interactions between the
PRF and WFF (Rickey and Fenster 1996; Willemetz et
al. 1989; Tysoe and Evans 1995). The magnitude of this
effect was observed to be higher for color Doppler than
for pulsed Doppler (Fig. 10 vs. Fig. 12), most likely
because of greater spectral broadening in the former.
Velocity gradients cause spectral broadening of the
Doppler signal, which makes the mean velocity estimate
more sensitive to the wall filter. This effect is also
dependent on the beam width, and may be a contributing
factor in the differences seen among transducers. How-
ever, the velocity gradient in the torus phantom is small,
on the order of 4.2 s⫺1 for a velocity of 115.2 cm/s, so
562 Ultrasound in Medicine and Biology
Fig. 12. Pulsed Doppler-derived velocity as a function of velocity, PRF and WFF for the P3-2 phased array transducer
at a Doppler angle of 40°: (a) 28.8 cm/s; (b) 57.6 cm/s; (c) 115.2 cm/s; (d) 172.8 cm/s; (e) 230.4 cm/s; (f) 288.0 cm/s.
that effects of the velocity gradient are considered to be
minimal. Some of the variation in the effect of the wall
filter on the accuracy may also be due to differences in
the amount of beam spread with differences in the range
of interest and focal position.
Effects of velocity and Doppler angle
It is speculated that the effects of velocity and Doppler
angle are for the most part secondary effects of the PRF,
because the usable range of PRF was highly dependent on
these two variables. Thus, a high velocity at a low Doppler
angle would be expected to have a similar effect as a lower
velocity at a higher Doppler angle; as far as the US system
is concerned, the two are equivalent. However, the velocity
distribution would not necessarily be the same at the two
Doppler angles, even if the perceived mean velocity is the
same. This effect can be seen in Figs. 10a and 11a. The
Volume 27, Number 4, 2001
effect of WFF at the highest PRF is much less in the 40°
case than at an equivalent PRF in the 70° case. This may be
because the Doppler spectrum (when shifted by the cosine
of the Doppler angle) is broader at a high angle than at a low
angle.
In the linear arrays, the beam was steered to a
constant 70°, regardless of Doppler angle, so there were
no differences in accuracy merely due to differences in
beam asymmetry with steering angle. In the phased and
curved arrays, the amount of steering varied somewhat
(between 5° and 20°), so that a small amount of the
observed variation in accuracy may be attributed to dif-
ferences in the beam asymmetry.
Comparison with pulsed Doppler results
Pulsed Doppler was found to be substantially more
accurate than color Doppler for the transducers tested.
 Color Doppler accuracy ● S. F. C. STEWART
Most likely this is due to color Doppler’s mean fre-
quency estimator having to work with fewer points (Wil-
lemetz et al. 1989). The subsequent broader spectrum is
more susceptible to aliasing of frequency components,
which contribute to a lower mean frequency estimate.
This biases the velocity estimate downward to a greater
extent than is seen in the pulsed Doppler case. This effect
also explains the decreased effect of the wall filter in
pulsed Doppler. The relatively narrow spectrum of
pulsed Doppler is less susceptible to components being
filtered out than that of color Doppler, so that the effect
of the wall filter on the mean frequency is less. Never-
theless, the L10-5 linear-array transducer was still less
accurate than the P3-2 phased-array transducer in pulsed
Doppler mode, possibly due to its higher degree of
focusing and concomitant broader bandwidth.
Clinical relevance
The clinical relevance of these effects is difficult to
assess in detail, but general conclusions can be made.
Quantitative use of color Doppler lags behind that of
other US modes such as pulsed- and continuous-wave
Doppler. However, it is possible that even when color
Doppler is used qualitatively, the clinician may make a
conscious or unconscious correlation between color and
velocity. If so, errors may creep in.
The overall underestimation of the true velocity by
the various transducers may not be as much of a problem
as the variations in accuracy from one exam to the next.
The P3-2 phased-array transducer will be preferred for
cardiac exams, so that errors due to variations with
transducer choice will be minimized. However, in eval-
uating vascular disease, if the L10-5 is used for one
exam, and the L5 used for the next, errors in interpreta-
tion may result. In any case, minor changes in the wall
filter or PRF between serial exams may mimic changes
in the measured velocity where none exist or, conversely,
mask a significant clinical change. The extent to which
this occurs on a day-to-day basis, and the significance to
clinical diagnosis, are unknown.
SUMMARY
The conclusions of this study can be summarized as
follows:
1. The torus phantom was found to be a useful tool for
detailed assessments of color Doppler accuracy over a
wide range of velocity, Doppler angle and instrument
settings.
2. The four transducers tested gave estimated velocities
averaging 90% or less of the real velocity. The dif-
ferences among the transducers was statistically sig-
nificant at p ⬍ 0.001, with the accuracy of the phased
array and curved array significantly greater than that
of the two linear array transducers.
563
3. Color Doppler accuracy was also found to be statis-
tically significantly dependent on the pulse-repetition
frequency and wall-filter frequency (p ⬍ 0.001).
These two settings were observed to depend heavily
on each other. In an earlier study (Stewart 1999),
other instrument settings, such as output power, color
gain and box width, were observed to have much less
effect on color Doppler accuracy.
4. The effects of the Doppler angle and the magnitude
of the actual fluid velocity were also found to be
statistically significant at p ⬍ 0.001. However, this
may be an effect of changes in PRF and WFF,
required to keep the velocity in the proper range
when either the actual velocity was changed, or the
perceived velocity was changed due to changes in
the Doppler angle.
5. The interactions among the independent variables
tested complicated the effects considerably.
6. Clinicians need be aware of these effects whether
interpreting color Doppler quantitatively or intu-
itively. These effects are especially important when
evaluating serial exams for changes in the character of
blood flow.
Acknowledgements— The expert machining assistance of James Duff
and Bruce Fleharty in building the phantom and making many excellent
suggestions is gratefully acknowledged. The assistance of Dr. Gerald
R. Harris with the optical motor controller is also appreciated. The
assistance of Dr. Hector Lopez with the theoretical interpretations is
also acknowledged. The mention of commercial products, their source,
or their use in connection with material reported herein is not to be
construed as either an actual or implied endorsement of such products
by the Department of Health and Human Services.
REFERENCES
Boote EJ, Zagzebski JA. Performance tests of Doppler ultrasound
equipment with a tissue and blood-mimicking phantom. J Ultra-
sound Med 1988;7:137–147.
Fleming AD, McDicken WN, Sutherland GR, Hoskins PR. Assessment
of colour Doppler tissue imaging using test-phantoms. Ultrasound
Med Biol 1994;20:937–951.
Forsberg F, Liu JB, Russell KM, Guthrie SL, Goldberg BB. Volume
flow estimation using time domain correlation and ultrasonic flow-
metry. Ultrasound Med Biol 1995;21:1037–1045.
Gill RW. Measurement of blood flow by ultrasound: Accuracy and
sources of error. Ultrasound Med Biol 1985;11:625– 641.
Hein IA, O’Brien WD. A flexible blood flow phantom capable of
independently producing constant and pulsatile flow with a predict-
able spatial flow profile for ultrasound flow measurement valida-
tions. IEEE Trans Biomed Engr 1992;39:1111–1122.
Holland CK, Clancy MJ, Taylor KJW, Alderman JL, Purushothaman
K, McCauley TR. Volumetric flow estimation in vivo and in vitro
using pulsed-Doppler ultrasound. Ultrasound Med Biol 1996;22:
591– 603.
McDicken WN. A versatile test-object for the calibration of ultrasonic
Doppler flow instruments. Ultrasound Med Biol 1986;12:245–249.
McDicken WN, Morrison DC, Smith DSA. A moving tissue-equivalent
phantom for ultrasound real-time scanning and Doppler techniques.
Ultrasound Med Biol 1983;9:L455–L459.
Phillips DJ, Hossack J, Beach KW, Strandness DE Jr. Testing ultra-
sonic pulsed Doppler instruments with a physiologic string phan-
tom. J Ultrasound Med 1990;9:429 – 436.
564 Ultrasound in Medicine and Biology
Picot PA, Fruitman M, Rankin RN, Fenster A. Rapid volume flow rate
estimation using transverse colour Doppler imaging. Ultrasound
Med Biol 1995;21:1199 –1209.
Rickey DW, Fenster A. A Doppler ultrasound clutter phantom. Ultra-
sound Med Biol 1996;22:747–766.
Rickey DW, Picot PA, Christopher DA, Fenster A. A wall-less vessel
phantom for Doppler ultrasound studies. Ultrasound Med Biol
1995;21:1163–1176.
Rickey DW, Rankin R, Fenster A. A velocity evaluation phantom for
colour and pulsed Doppler instruments. Ultrasound Med Biol 1992;
18:479 – 494.
Schwarz KQ, Bezante GP, Chen X. When can Doppler be used in place
Volume 27, Number 4, 2001
of integrated backscatter as a measure of scattered ultrasound
intensity? Ultrasound Med Biol 1995;21:231–242.
Stewart SFC. A rotating torus phantom for assessing color Doppler
accuracy. Ultrasound Med Biol 1999;25:1251–1264.
Tysoe C, Evans DH. Bias in mean frequency estimation of Doppler
signals due to wall clutter filters. Ultrasound Med Biol 1995;21:
671– 677.
Walker AR, Phillips DJ, Powers JE. Evaluating Doppler devices using
a moving string test target. J Clin Ultrasound 1982;10:25–30.
Willemetz JC, Nowicki A, Meister JJ, De Palma F, Pante G. Bias and
variance in the estimate of the Doppler frequency induced by a wall
motion filter. Ultrason Imag 1989;11:215–225.