High frame-rate and high resolution medical
imaging using adaptive beamforming
Johan-Fredrik Synnevåg
Andreas Austeng
Sverre Holm
Department of Informatics, University of Oslo
P.O. Box 1080, N-0316 Oslo, Norway
Abstract— Recently, significant improvement in image
quality has been demonstrated by applying the minimum
variance (MV) beamformer to medical ultrasound imaging.
In this paper we show that the low sidelobe level and
narrow beamwidth of the MV beamformer can be utilized,
not only to increase resolution, but to enhance image quality
in several ways: Fewer and wider transmit beams combined
with increased number of parallel receive beams may be used
to increase frame rate. We demonstrate that the resolution
obtained using an MV beamformer is equal to a conventionally beamformed image of wire targets, using 1/4th of the
number of transmit beams and four parallel receive beams,
potentially increasing the frame-rate by a factor four. The
enhanced resolution of the MV beamformer may also be
utilized to decrease the number of focal zones required
to form a high resolution image. We demonstrate better
resolution using one focal zone with an MV beamformer,
compared to full dynamic focus using DAS. Finally, we
show that by lowering the frequency of the transmitted
beam and beamforming the received data with the adaptive
beamformer, increased depth of penetration is achieved
without sacrificing lateral resolution.
I. I NTRODUCTION
Several authors have applied the minimum variance
(MV) adaptive beamformer to medical ultrasound imaging and shown significant improvement in image quality [1], [2], [3], [4], [5]. We have previously shown
reduction in mainlobe width and decrease in sidelobe
level and demonstrated increased resolution in medical
images using the MV beamformer [4].
In this paper we show how the properties of the
MV beamformer can be utilized to increase frame-rate
and increase depth of penetration in medical imaging
without sacrificing resolution compared to delay-andsum (DAS). We demonstrate that higher frame-rate can
be achieved in two ways: By using fewer and wider
transmit beams combined with an increased number of
parallel receive beams processed with the MV beamformer, increased frame-rate can be achieved without
sacrificing image quality. Also, the high resolution of
the MV beamformer allows fewer focal zones, giving a
potential increase in frame-rate.
Due to the narrow beamwidth of the MV beamformer, the center frequency of the transmit beam can
be lowered without reducing the lateral resolution. This
gives increased depth of penetration in attenuating media without sacrificing image quality.
In the next section we present how we applied the MV
beamformer to medical ultrasound imaging, and present
three ways to utilize its properties to enhance image
quality. In Section III we show results from each of the
proposed enhancements, and draw conclusions from our
findings in Section IV.
II. M ETHOD
A. Minimum variance beamformer
We consider a transducer consisting of M channels,
each recording a signal xm (t) per scan-line. We assume
that each channel has been delayed to focus dynamically
at points along the scan-line. The M observations are
ordered in a vector


x0 (t)
 x1 (t) 


X(t) = 
(1)
.
..


.
x M−1 ( t )
The output of the beamformer is a weighted sum of the
delayed measurements,
M−1
z(t) =
∑
wm (t) xm (t) = w(t) H X(t),
(2)
m=0
where wm (t) is the aperture weight for sensor m at
time t, and w(t) = [w0 (t) w1 (t) · · · w M−1 (t)] H . The MV
beamformer minimizes the variance of the output of the
beamformer [6], [7],
P(t) = E[| z(t)|2 ],
(3)
while maintaining unit gain in the focal point. This gives
the optimization problem
min w(t) H R(t)w(t)
(4)
subject to w(t) H a = 1,
(5)
w(t)
where
i
h
R(t) = E X(t)X(t) H
(6)
is the spatial covariance matrix and a is the equivalent of
the steering vector in narrowband applications. Since the
data already have been delayed to focus at the point of
interest, a is simply a vector of ones. The solution to (5)
is [7]
R( t )−1 a
w(t) = H
.
(7)
a R( t )−1 a
Applying the weights in (7) to (2) gives the MV amplitude estimate.
In practice the covariance matrix, R(t), in (7) is replaced by the sample covariance matrix, R̂(t). To obtain
a good estimate, we use spatial smoothing [8], where the
array is divided into overlapping subarrays, and the
covariance matrices for all subarrays are averaged. This
technique is also used to avoid signal cancellation when
signals are correlated. The estimated covariance matrix
at time t then becomes
R̂(t) =
1
M−L+1
where



X̄l (t) = 

M− L
∑
l =0
X̄l (t)X̄lH (t),
xl (t)
xl +1 ( t )
..
.
xl + L−1 ( t )
(8)



,

(9)
1
M−L+1
M− L
∑
w(t) H X̄l (t).
We achieve increased depth of penetration in attenuating media if the frequency of the transmitted beam is
reduced. This is because high frequencies are attenuated
more than low. Lowering the transmit frequency will
however affect resolution in the image, so by using a
DAS beamformer penetration depth must be traded off
for resolution. In the following section we show that the
lateral resolution lost by lowering the transmit frequency
can be compensated for using an MV beamformer on
reception. The axial resolution is however also affected,
but is usually much better than the lateral resolution, so
a reduction can be afforded.
III. R ESULTS
We have simulated a 96 element, 4MHz, 18.5 mm
transducer, imaging wire targets using Field II [9]. All
transmitter and receiver combinations were simulated,
allowing dynamic focus on both transmission and reception. The depth of the wire targets ranged from 30
mm to 80 mm, and the wires were separated by 3 mm.
The following results show simulations from the three
proposed enhancements using the MV beamformer.
A. Parallel receive beams
and L is the subarray length. After computation of the
optimal aperture weights using the sample covariance
matrix, we obtain the amplitude estimate as
ẑ(t) =
C. Increasing depth of penetration
(10)
l =0
The choice of L affects the performance the MV beamformer. A smaller L gives poorer resolution, but an
increasingly robust amplitude estimate.
B. Increasing frame-rate
We propose to utilize the high resolution of the MV
beamformer to increase frame-rate without sacrificing
resolution. Fewer and wider transmit beams combined
with multiple parallel receive beams can be used to
increase frame-rate in medical imaging. However, using conventional beamforming, there exists a trade-off
between image quality and frame-rate, as wider transmit
beams leads to lower resolution. We will show in Section III that by using an MV beamformer on reception,
the resolution lost from the wide transmit beams can be
compensated for.
A high-quality medical ultrasound image may be obtained using many focal zones. Using DAS beamforming
we must trade off frame-rate for image quality, as increasing the number of focal zones leads to a reduction
in frame-rate. In Section III we show that targets that
are out of focus are may be equally well resolved using
an MV beamformer, demonstrating that the number of
focal zones can be decreased.
Fig. 1 shows images of the wire targets obtained
with DAS and MV beamforming. The MV image was
formed with 1/4th of the number of transmission lines
as the DAS image, using a wider transmit beam and
four parallel receive beams. We chose the number of
scan lines to be twice the number required from the
Rayleigh criterion, to avoid potential amplitude loss if
targets appear between scan lines. The wider transmit
beam was achieved using only 1/4th of the available
transducer elements. The loss in amplitude due to fewer
transmit elements was compensated for. We see that
the wires are resolved in both images, and that the
resolution is approximately the same. Fig. 2 shows the
lateral resolution at depths 50 and 80 mm. We see that
the mainlobe width is approximately the same for the
two beamformers. The sidelobe level is lower for the
MV beamformer at 80 mm. The lateral resolution of DAS
using four receive beams has been included for reference.
B. Decreasing the number of focal zones
Fig. 3 compares a DAS image using full dynamic focus
to an MV image using one focal zone. The focal depth
was 50 mm. We see that the resolution is better for
the MV beamformer, even though only a single focal
zone was used. Fig. 4 shows the lateral resolution at
50 and 70 mm. We see that the mainlobe is narrower
and the sidelobes are lower for the MV beamformer at
both depths. The amplitude is however lower, since the
targets are out of focus of the transmission beam.
0
30
−5
−10
−10
−10
−15
−20
−20
−40
50
−25
−30
60
−40
70
−50
80
−55
0
Amplitude [dB]
−40
70
80
0
30
−5
−10
40
−15
−20
50
−25
−30
60
−35
−40
70
−45
−50
80
−55
0
10
20
[mm]
−55
0
( a)
(b)
10
20
[mm]
(b)
Fig. 3. Simulated wire targets using an 18.5 mm, 96 element, 4 MHz
transducer. (a) DAS using full dynamic focus, (b) MV using one focal
zone (focal depth 50 mm).
C. Depth of penetration
−10
−20
−30
−40
−50
DAS
MV, L=32 (4 Rx beams)
DAS (4 Rx beams)
10
15
20
Angle [degrees]
( a)
0
Fig. 5 compares the images obtained when we included attenuation in the simulations. The attenuation
was 0.5 dB/cm/MHz. Figure (a) shows the image obtained using DAS with a 4 MHz transducer. Figure (b)
shows the corresponding MV image using a 2 MHz
transducer with the same size and number of elements.
We see that the lateral resolution in the images is about
the same. We also see that the reflections from the
deeper targets are attenuated in the DAS image. From
the steered responses in Fig. 6 we see that amplitude
is about 10 dB lower using 4 MHz transmission at 70
mm, showing that increased depth of penetration is
achieved without sacrificing lateral resolution using an
MV beamformer.
IV. C ONCLUSION
−10
Amplitude [dB]
−35
−50
0
−20
−30
−40
−50
−70
5
60
−50
Fig. 1. Simulated wire targets using an 18.5 mm, 96 element, 4 MHz
transducer. (a) DAS (b) MV with 4 parallel receive beams.
−60
−30
−45
10
20
[mm]
( a)
−70
5
−25
−45
−55
10
20
[mm]
50
−35
−45
−60
40
−15
−35
80
−5
−20
−30
70
−5
−15
−25
60
MV, L=32
0
30
Depth [mm]
50
40
Depth [mm]
Depth [mm]
40
0
DAS (dynamic focus)
MV, L=32 (4 Rx beams)
0
Depth [mm]
DAS
30
DAS
MV, L=32 (4 Rx beams)
DAS (4 Rx beams)
10
15
20
Angle [degrees]
(b)
Fig. 2. Steered responses of DAS (solid) and MV (solid with diamonds)
beamformers using an 18.5 mm, 96 element, 4 MHz transducer. The
MV beamformer used 4 parallel receive beams. DAS using 4 parallel
receive beams has been included for reference (dashed) (a) 50 mm (b)
80 mm depth
We have shown that the MV beamformer may be applied to medical ultrasound imaging to increase framerate or depth of penetration. On simulated images of
wire targets we have demonstrated potential increase in
frame-rate by a factor four without loss of resolution.
We have also shown that the resolution achieved using
one focal zone and the MV beamformer was better than
DAS using full dynamic focus. This also shows potential
for increased frame-rate. Finally, we have demonstrated
increased depth of penetration by lowering the transmit
frequency and applying the MV beamformer on reception. At half the transmit frequency, the lateral resolution
was the same as for DAS, and about 10 dB gain was
achieved for the deeper targets.
R EFERENCES
[1] J. A. Mann and W. F. Walker. A constrained adaptive beamformer
for medical ultrasound: Initial results. Ultrasonics Symposium, 2002.
Proceedings. 2002 IEEE, 2:1807–1810, October 2002.
0
−10
−10
−20
−20
Amplitude [dB]
Amplitude [dB]
0
−30
−40
−30
−40
−50
−50
−60
−60
DAS (dynamic focus)
MV, L=32
−70
5
10
15
−70
5
20
DAS (4 MHz)
MV (2MHz), L=48
10
20
( a)
( a)
0
0
−10
−10
−20
−20
Amplitude [dB]
Amplitude [dB]
15
Angle [degrees]
Angle [degrees]
−30
−40
−30
−40
−50
−50
−60
−60
DAS (dynamic focus)
MV, L=32
−70
5
10
15
20
Angle [degrees]
−70
5
DAS (4 MHz)
MV (2MHz), L=48
10
15
20
Angle [degrees]
(b)
(b)
Fig. 4. Steered responses of DAS and MV beamformers using an 18.5
mm, 96 element, 4 MHz transducer: DAS (solid) using full dynamic
focus and MV (solid with diamonds) using one focal zone (focal depth
50 mm). The reduced amplitude in the MV response is due to targets
being out of focus (a) 40 mm (b) 70 mm depth.
Fig. 6. Steered responses of DAS and MV beamformers using an
18.5 mm, 96 element transducer. DAS (solid) used 4 MHz as transmit
frequency, while MV (solid with diamonds) used 2 MHz (a) 50 mm
(b) 80 mm depth.
DAS (4 MHz)
MV (2MHz), L=48
0
30
0
30
−5
−5
−10
40
−10
40
−20
50
−25
−30
60
−35
−40
70
−15
Depth [mm]
Depth [mm]
−15
−20
50
−25
−30
60
−35
−40
70
−45
−50
80
−45
−50
80
−55
0
10
20
[mm]
( a)
−55
0
10
20
[mm]
(b)
Fig. 5. Simulated wire targets using an 18.5 mm, 96 element transducer. (a) DAS (4 MHz) (b) MV (2 MHz).
[2] Magali Sasso and Claude Cohen-Bacrie. Medical ultrasound imaging using the fully adaptive beamformer. Acoustics, Speech and
Signal Processing, 2005. Proceedings (ICASSP ’05). IEEE International
Conference on, 2, March 2005.
[3] Zhisong Wang, Jian Li, and Renbiao Wu. Time-delay- and timereversal-based robust Capon beamformers for ultrasound imaging.
Transactions on Medical Imaging, 14:1308–1322, Oct. 2005.
[4] J.-F. Synnevåg, A. Austeng, and S. Holm. Minimum variance
adaptive beamforming applied to medical ultrasound imaging.
Proc. IEEE Ultrasonics Symposium, 2:1199–1202, Sept. 2005.
[5] F. Viola and W. F. Walker. Adaptive signal processing in medical ultrasound beamforming. Proc. IEEE Ultrasonics Symposium,
4:1980–1983, Sept. 2005.
[6] F. Bryn. Optimum signal processing of three-dimensional arrays
operating on Gaussian signals and noise. The Journal of the
Acoustical Society of America, 34(3):289–297, March 1962.
[7] J. Capon. High-resolution frequency-wavenumber spectrum analysis. Proc. IEEE, 57:1408–1418, August 1969.
[8] Tie-Jun Shan, Mati Wax, and Thomas Kailath. On spatial smoothing for direction-of-arrival estimation of coherent signals. IEEE
Transactions on Acoustics, Speech, and Signal Processing, 33(4), August
1985.
[9] J. A. Jensen. Field: A program for simulating ultrasound systems.
Medical & Biological Engineering & Computing, 34:351–353, 1996.