ULTRASONIC IMAGING 10, 153-170
THE
MALTeSE CROSS PROCESSOR: SPECKLE REDUCTION FOR CIRCULAR TRANSDUCERS
S.W. Smith"2 and O.T. von Rannn3
'Center for Devices and Radiological Health
Food and Drug Administration
Rockville, MD 20857
*Department of Radiology
Duke University Medical Center
Durham, NC 27706
3Departmentof Biomedical Engineering
Duke University
Durham, NC 27706
new online signal processing technique is described to reduce
speckle noise in ultrasound images. In the imaging system, a focused
piston transducer is divided into thirty-two sectors. In the receive mode,
parallel signal processing arranges the sectors into eight maltese crosses.
The rf signals of the perpendicular arms of each cross are multiplied in a
phase sensitive process. The orthogonal receive mode multiplication is
designed to reduce side lobes resulting from the sector shapes while
maintaining lateral resolution through the use of the full aperture
diameter. The signals from the crosses are then combined via postdetection
sumnation. Six of the eight crosses perform successfully. The six maltese
crosses show decorrelated signals equivalent to four independent samples of
the speckle noise which decreases noise contrast by a factor of two with no
measureable loss of spatial resolution. Post summation compression is
included to retain the conventional signal dynamic range. Parallel signal
processing maintains the normal image line rate. Images of tissue-mimicking
phantoms including speckle targets show improved detectability of simil.ated
lesions. o 1988 Academic Press, Inc.
A
Key words:
Resolution; signal-to-noise ratio; speckle reduction;
ultrasound imaging.
The topic of speckle interference in ultrasound images has received
significant attention in the last decade. In 1978, Burckhardt [l] first
0161-7346/88 $3.00
153
Copyright 0 I988 by Academic Press, Inc.
All rights of reproduction in any form reserved.
Downloaded from uix.sagepub.com at PENNSYLVANIA STATE UNIV on March 5, 2016
SMITH AND VON RAMM
noted its existence in B-mode images and its analog to laser speckle. He
also derived the first order statistics of speckle and described how to
reduce speckle contrast in an ultrasound image through the technique of
spatial compounding. That is, the average of N uncorrelated samples of an
object volume from several independent transducer orientations in both the
transmit and receive modes reduces the contrast of the speckle noise
(increases the speckle signal-to-noise ratio, SNR) by N"* Since then,
several investigators [2-41, including those from our own laboratories
[5-71, have attempted to ameliorate the effects of speckle using spatial
compounding in high speed linear phased arrays.
.
For the case of mechanically scanned circular transducers the topic of
spatial compounding per se has received little attention [El. However,
investigators at the University of Toronto [9,10] have achieved significant
speckle reduction via signal processing techniques which are similar to
spatial compounding. A single orientation of the transmit aperture was
used, but the receive aperture is divided into several subapertures to
obtain independent samples of the target volume. A hybrid transducer was
used consisting of a spherically shaped, focused transducer combined with a
concentric planar transducer disk and two aluminum mirrors. In their most
recent work, both the spherical transducer and the planar disk were
segmented into eight sectors which enabled several options for receive mode
signal processing. In each of these techniques, a full circular aperture
was used in the transmit mode with either the spherical transducer or the
planar di-sk. In the receive m d e , the signal outputs from the eight
sectors of either the spherical transducer or the planar disk were combined
using multiplicative processing OK summation. The techniques included
"phase insensitive sector addition," i.e.,
s, (t)=CN=8 2, (t)
i=l
where xi(t) are the RF signals from individual sectors of the array, R,(t)
are the signals after envelope detection, and S,(t) is the sum of these
signals for the N=8 sectors of the receive mode transducer. Other speckle
reduction techniques evaluated have included multiplicative processing of
the rf signals, i.e.,
154
Downloaded from uix.sagepub.com at PENNSYLVANIA STATE UNIV on March 5, 2016
MALTESE CROSS PROCESSOR
and "phase insensitive sector multiplication", i.e.,
For each of these signal processing techniques, as well as for the
schemes of spatial compounding with linear phased arrays, one must evaluate
the trade-off of increasing speckle SNR versus loss of lateral resolution.
For a given imaging task in the body, one must ask whether it is better to
use the full transducer aperture with optimum lateral resolution in a
simple scan, or to divide the aperture into N subapertures to achieve N
independent samples of the speckle noise and increase the speckle SNR by
(N)l'* while losing some lateral resolution. To our knowledge, there have
been no published clinical studies to confirm the effectiveness of speckle
reduction via spatial compounding. Instead, investigators have relied on
various figures of merit in studies of phantoms to evaluate success of
previous speckle reduction techniques. These figures of merit have
included: 1) the point spread response of the imaging system, 2 ) the ratio
of mean to standard deviation, ,u/u, the so-called signal-to-noise ratio of
the speckle probability density function, 3 ) the crosscorrelation
coefficient between samples of the speckle target volume, 4 ) measurements
of the detectability of lesions in tissue mimicking materials such as
contrast-detail diagrams and the area-wise signal-to-noise ratio [lll or
the contrast-to-speckle ratio [91. In previous studies, there have been
consistent losses of lateral resolution or image contrast as measured by
the point spread response in exchange for the improvement in speckle SNR.
In the present study, our goal was to develop an online signal
processing technique for speckle reduction in conventional mechanical
scanners which use focused piston transducers. We wished to minimize
degradation of the transducer point spread response by the signal
processor. McKechnie [12] had analyzed laser speckle reduction for a
circular lens and recommended the time average signal of a rotating
aperture shaped in the form of a Maltese cross to achieve speckle
reduction. Since the length of each orthogonal a m of the cross is equal
to the lens diameter, there is no apparent loss of lateral resolution as
determined by the main lobe of the point response function. However, this
geometry causes a significant loss of image contrast due to the increased
side lobes of the response function. We believed that multiplicative
processing of the orthogonal arms of the Maltese cross would maintain the
point response function of the imaging system.
155
Downloaded from uix.sagepub.com at PENNSYLVANIA STATE UNIV on March 5, 2016
SMITH AND VON RAMM
In this paper, we describe the design and implementation of the
Maltese cross processor which includes a multiplicative processing scheme
and which has been incorporated into a commercial static B-scan system. We
describe our preliminary measurements of the performance of the system in
terms of its point spread response, the improvement in the speckle SNR
( p / u ) and improved images of speckle targets in tissue mimicking phantoms.
We also discuss briefly its performance in terms of the more complex
measures of lesion detection.
Figure 1 illustrates the principles of the Maltese cross processor. A
conventional focused piston transducer is segmented into 32 equal sectors.
In the transmit mode, the 32 sectors are fired simultaneously by a single
transmitter. Each sector and its opposite member, separated by 180°, is
wired together at the transducer to form 16 "bar-tie" elements. The length
of each element is equal to the transducer diameter. The element can be
approximated by a long thin rectangle whose length is approximately 10
times its width. The receive mode rf signals from the horizontal element
are multiplied by the signal from the orthogonal vertical element.
The results of this product are illustrated qualitatively in figure 1
in terms of the contour map of the recei e mode point spread function. The
main lobe of the diffraction pattern P ( y 6 ) of each element is approximated
by
sin B
B
Fig. 1
(4)
conventional focused piston transducer is segmented into 16
"bar-tie" elements. The rf output of each element is multiplied
by that of its orthogonal element to produce the Maltese cross.
The product of the point spread responses of the horizontal and
vertical elements (shaded areas) yields a response peaked on the
origin with side lobes only on the x and y axis.
A
156
Downloaded from uix.sagepub.com at PENNSYLVANIA STATE UNIV on March 5, 2016
MALTESE CROSS PROCESSOR
where a = nDsiny/X, y is the angle parallel to the length of the element
relative to the axis of the transducer, f3 = n2Dsind/32X, and d is the angle
parallel to the element width relative to the axis of the transducer. The
side lobe amplitude of the receive mode diffraction pattern is determined
by the "b-tie" apodization of each element and is significantly larger
than that predicted by a sin X/X function.
The contour map of the diffraction pattern of the shaded horizontal
element is shown narrow in the x direction with two side lobes and broad in
the y direction. Conversely, the diffraction response of the shaded
vertical element is narrow in the y direction with two side lobes and broad
in the x direction. The receive mode product of these two orthogonal
diffraction patterns is strongly peaked at the origin with side lobe
contributions only on the x and y axes. The same product is performed in
parallel for each orthogonal pair of the sixteen elements resulting in rf
signals from eight Maltese crosses. The signal from each cross is envelope
detected in parallel and summed with the detected signal from the other
crosses. The overall signal process is described by
i=l
where the bar now represents the envelope detection process.
The side lobes for each cross occur at different angles without
significant overlap. Because of the summation in Q. (5) and the lack of
overlapping side lobes, the main lobe amplitude relative to the side lobe
amplitude is increased by approximately a factor of eight compared to that
of a single cross. The resulting overall receive mode response pattern is
circularly symnetric with minimal side lobe contributions. The overall
transmit-receive point response is further improved by the Airy disk
pattern of the full circular transmit aperture.
Figure 2 s h m , in detail, our hardware implementation of the Maltese
cross processor and compares the conventional signal processing of the
control case versus that of the Maltese cross for the same focused piston
transducer. The signal processing techniques were implemented in parallel
to enable simultaneous comparison of the resulting B-mode images. A 19 mm,
8 an radius of curvature, PZT-SA crystal, with nominal frequency of 2.25
MHz, was used to fabricate the transducer array. After dicing and backing,
the array operated at a center frequency of 1.8 MHz. The receive mode rf
157
Downloaded from uix.sagepub.com at PENNSYLVANIA STATE UNIV on March 5, 2016
SMITH AND VON RAMM
8 Crosses
32 Sectors
Amplifiers
Amplifiers
Detectors/
LPF
Summer
Detector/
LPF
Fig. 2
(A) Conventional signal processing for the control case. The rf
output from each transducer element is summed before envelope
detection and B-mode display. (B) Signal processing for the
Maltese cross. The rf output of orthogonal arms of each cross
is multiplied together. The output of each multiplier is
envelope detected and summed. The sum undergoes logarithmic
compression before B-mode display.
signals from each of the 16 elements pass in parallel through one of 16
matched FET amplfiers contained in the transducer handle for the purpose of
impedance matching. The signals then pass through one of 16 matched
preamplifiers. The 16 preamplifiers were retrieved from a Grumman model
RT400 phased array system no longer in clinical use. The Grumman
preamplifiers with n;C produce 90 dB of gain over a dynamic range of 120 dB
and a minimum noise level of 10 UV.
shown in figure 2A, to achieve the equivalent of conventional
processing for a single focused piston with no speckle reduction, we summed
the 16 rf outputs from the preamplifiers in an Analog Devices model 465
operational amplifier. The summed signal then is envelope detected in t h e
conventional manner. It should be noted that, other than envelope
detection, the signal has undergone no nonlinear amplification such as
logarithmic compression or reject up to this point. The detected signal is
then sent to the analog scan converter of a Picker 8OL static B-scanner.
The transducer array is mechanically connected to the articulated arm of
the Picker device, but the arm and transducer are constrained to a simple
linear scan format by a ball-bearing slide assembly which travels on rails
above the water tank.
As
The Maltese cross processing of figure 2B is performed in parallel
with the signal processing of the control case of figure 2A. The rf
multiplication of the shaded horizontal arm times the shaded vertical arm
is performed by an Analog Devices 429B multiplier. The output signal is
158
Downloaded from uix.sagepub.com at PENNSYLVANIA STATE UNIV on March 5, 2016
MALTESE CROSS PROCESSOR
then centered at 3.6 MHz. The 1.8 MHz feedthrough is reduced by 23 dB by a
high pass filter following the multiplication. The output signal from the
multiplier undergoes conventional envelope detection. The signals from
the other seven crosses are obtained from parallel multipliers and envelope
detectors. The detected signals from the eight crosses are then summed in
an operational amplifier. The multiplicative operation of the Maltese
cross signal processor doubles the signal dynamic range. We restored the
original dynamic range by a postsumation logarithmic compression amplifier
(30 dB to 20 dB) which approximates a square root operation.
During fabrication of the transducer array, two adjacent sector
elements (#31 and #32) failed so that in our preliminary evaluations we
were left with six operating crosses. To ensure equivalent operation for
the control case, we also disconnected the remaining transducer elements of
those two failed crosses in receive mode.
The point spread response of the Maltese cross signal processor was
compared to that of the control for the lateral dimension by scanning the
tip of a 1 mm stainless steel wire in a two dimensional rectilinear pattern
in the transducer focal zone at a range of approximately 8 cm. The output
signal from the tip of the wire target was sent to a Panametrics Model
5052UA peak detector, and the resulting dc level was digitzed and stored in
a Hewlett-Packard model 21MX computer. We took care not to saturate the
Panametrics peak detector or the logarithmic compressor of the Maltese
cross processor.
Because of the multiplicative operation of the Maltese cross, we were
concerned about potential image artifacts resulting from cross product
terms from multiple closely spaced specular targets. Therefore, we also
compared the B-scans of an immersible MUM standard 100 m wire phantom
[13] for the control case versus the Maltese cross.
To evaluate the speckle reduction capabilities of the Maltese cross
processor, we scanned tissue mimicking material. We used a slurry of 1 mm
agar spheres in a 5 percent solution described by Madsen et al. [14] and
obtained from RMI, Inc. Each sphere contained 20 pm graphite scatterers.
We had recently used this material to develop a new version of the
ultrasound contrast-detail phantom [15]. The slurry exhibits Rayleigh
statistics and images of the material using clinical scanners operating
without logarithmic compression or reject (linear mode) show a fully
developed speckle pattern whose speckle SNR was approximately 1.91.
We determined the cross-correlation coefficients between speckle
signals from each of the six crosses for all possible nonredundant
combinations of the Maltese cross signals. These include nearest
159
Downloaded from uix.sagepub.com at PENNSYLVANIA STATE UNIV on March 5, 2016
SMITH AND VON RAMM
neighbors, i.e., (N,N+1), N = 1,5, next nearest neighbors, (N,N+2), N =
1,4, etc. Note combinations with crosses #7 and #8 are omitted because of
the dead elements. Because of the circular symmetry of the crosses, the
combination (N,N+3) is equivalent to (N,N+5); (N,N+2) is equivalent to
(N,N+6), etc. In making these measurements for each combination of
crosses, we digitized a 4 cm range of envelope detected speckle signal
centered about the transducer focal zone and stored the signal in a
Tektronix 7854 digital oscilloscope. Crosscorrelation coefficients were
determined using the waveform calculator of the oscilloscope and repeated
for each pair of crosses.
The coefficients were calculated using
xj
where
and-Zj are the envelope detected echo at location j for line Y and
Z,and Y and Z are the mean data values of line Y and 2 . Coefficient data
were averaged for four independent regions of the tissue mimicking phantom.
For each of these four data sets, we also calculated the speckle SNR
for the control case and the Maltese cross processor. The SNR data were
obtained for the speckle signals (A-mode lines) rather than from regions of
interest in the corresponding B-mode images because we wished to exclude
any nonlinear processing associated with the scan converter of the Picker
system. To assess the effects of the logarithmic compressor in the Maltese
cross processor, we compared the SNR data after compression versus that
obtained using only the square root function in the Tektronix waveform
calculator.
Finally, we inserted into the slurry, negative contrast conical
targets fabricated from lower concentrations of the agar sphere material.
The cones had known object contrasts of -16 dB, -10 dB, and -3 dB
previously measured on clinical scanners w i t h no logarithmic compression or
reject, i.e., linear mode [15]. We obtained images of each cone in long
axis to compare the image quality at the tip of the cone for conventional
processing versus that of the Maltese cross. We also obtained short axis
images of the cones to simulate a disk shaped lesion.
160
Downloaded from uix.sagepub.com at PENNSYLVANIA STATE UNIV on March 5, 2016
MALTESE CROSS PROCESSOR
111. RESULTS
Figure 3 compares the transmit-receive point response function for the
control case versus the Maltese cross processor using both a 3-D graphic
display and a contour map. The Maltese cross data was taken using the
logarithmic compression to equalize signal dynamic range w i t h the cont r ol .
The main lobe of the Maltese cross is seen to be narrower than that of the
control processing. The -20 dB beam width of the Maltese cross measures
4 . 6 mm compared to 6.6 mm in the control. In the 3-D graphics display,
there is some evidence of low level off-axis signal in the Maltese cross
data. However, this is not seen in the contour maps out to a level of
-20 a. The off-axis energy in the Maltese cross below -20 dB is not
unexpected since the 1.8 MHz feedthrough from the multiplier was suppressed
only 23 dB below the 3 . 6 MHz desired signal as noted above.
Figure 4 compares the image quality of the control processing (left)
versus the Maltese cross processor (right)using the MUM 100 mm wire
phantom. The phantom had been turned on its side to use the lateral
resolution wire targets at a range of 4.5 cm and to scan the axial
resolution targets in the lateral direction at 8 cm range. We wanted to
examine the possibility of crossproduct artifacts in the Maltese cross
processor from closely spaced specular targets. There seems to be no
evidence of such artifacts at least within the dynamic range of the imaging
system. The Maltese cross shows a smaller point size at the 8 cm focal
point as also evidenced by figure 3 and shows a slightly better lateral
resolution of the closely spaced wires at 8 cm in range.
Fig. 3
Comparison of measured
transmit-receive point spread
response functions of control
processing ( A )
versus Maltese cross
processing ( B).
161
Downloaded from uix.sagepub.com at PENNSYLVANIA STATE UNIV on March 5, 2016
SMITH AND VON RAMM
Fig. 4
Comparison of Emode images of AIUN loom test object by
conventional signal processing (left) versus Maltese cross
processing ( right).
Figure 5 plots the crosscorrelation coefficients of speckle signals
obtained for nonredundant combinations of the six Maltese crosses. The
error bars shows 21 standard deviation for the measurement data. If the
speckle signals from the six crosses were completely independent, one would
expect the Maltese cross processor to increase the speckle S m by (6)”2 =
2.45. The partially correlated signals of figure 5 will result in less
improvement.
The increase in speckle SNR (,u/u) as a result of averaging m partially
correlated speckle images has been described by Trahey et al. [7] and can
1.0
Fig. 5
0.0 L-
FL
u
N+l
N+2
N +3 N +4
Average cross-correlation
coefficients of speckle
signals from neighboring
Maltese crosses measured in
tissue mimicking material
162
Downloaded from uix.sagepub.com at PENNSYLVANIA STATE UNIV on March 5, 2016
MALTESE CROSS PROCESSOR
be calculated using
where d (wk ) is the variance of image Wk and p ( w k ,Wj ) is the correlation
coefficient of images Wk and W, [161. Note that the mean N remains
constant in the averaging operation for equal original means.
The data of figure 5 used in Q. ( 7 ) yield a predicted improvement in
SNR by a factor of 1 . 5 9 (+.392/-.220). The values of the speckle SNR ( p / u )
for the control versus the Maltese cross processor were averaged for the
four independent speckle patterns used to obtain the cross-correlation data
above. The p / u data indicates a control value of 1 . 8 5 _+ .088 which is very
close to the value ( 1 . 9 1 ) expected for Rayleigh statistics. The Maltese
cross data show a value of 3.60 k.65 for the square root function of the
Tektronix digital oscilloscope versus a value of 3.61 k . 2 5 for the online
log compression circuit. The data are in close agreement indicating that
the log compressor reasonably approximates a square root for this
experiment
.
These results correspond to improvements in p / a by a factor of 1 . 9 4
which is in reasonable agreement with that predicted by Eq. ( 7 ) above
(within 1 standard deviation). The increase in speckle SNR shows that the
six partially correlated Maltese crosses yield improvement equivalent to
approximately four independent samples of the speckle noise.
Figures 6-9 show our preliminary imaging results comparing
conventional signal processing with the Maltese cross signal processing for
the six operating crosses. In figures 6-8, the conical targets, which are
seen in long axis, include contrasts of -16 dB, -10 dB and -3 ds. In each
case, the speckle shows significantly less contrast. Measurements of
speckle SNR for individual lines in these images show approximately a
factor of 2 increase as in the experiments described above. The smoother
speckle background in each figure enhances the visibility of the targets.
This is particularly true of the low contrast, -3 de target in figure 8
which is almost undetectable in the control image. Likewise, figure 9
showing the -10 dB cone in short axis at a diameter of 1.8 cm, demonstrates
reduced speckle contrast and improved detectability for the Maltese cross
processing.
163
Downloaded from uix.sagepub.com at PENNSYLVANIA STATE UNIV on March 5, 2016
SMITH AND VON RAMM
Pig. 6
Comparison of long axis B-mode images of conical speckle target
irrrmersed in speckle background for conventional processing
(left) versus Maltese cross processing (right). Target contrast
is -16 dB.
IV. DISCUSSION
We have described the design and implementation of the Maltese cross
signal processor which increases speckle SNR while maintaining lateral
Fig. 7
Comparison of long axis B-mode images of conical speckle target
immcrsed in speckle background for conventional processing
(left) versus Maltese cross processing (right). Target contrast
is -10 dB.
164
Downloaded from uix.sagepub.com at PENNSYLVANIA STATE UNIV on March 5, 2016
MALTESE CROSS PROCESSOR
Fig. 8
comparison of long axis B-mode images of conical speckle target
inmersed in speckle background for conventional processing
(left) versus Haltese cross processing (right). Target contrast
is -3 dB.
resolution for mechanically scanned circular transducers. Due to the
multiplicative signal processing of orthogonal arms of each cross,
measurements using the point spread response and the MUM test phantom
showed no loss of lateral resolution for the Maltese cross. For the six
operating Maltese crosses, measurements of speckle SNR showed an
improvement of approximately a factor of two which is equivalent to four
Fig. 9
images of conical speckle target
Comparison of short axis
imnersed in speckle background for conventional processing
(left) versus Maltese cross processing (right). Target contrast
is -10 dB.
165
Downloaded from uix.sagepub.com at PENNSYLVANIA STATE UNIV on March 5, 2016
SMITH AND VON RAMM
independent samples of the speckle noise. In addition, images of high and
low contrast targets show improvement in the detectability of such objects.
With these preliminary evaluations of the mltese cross, we can
discuss more complex figures of merit for the problem of lesion detection
in medical ultrasound. We have previously proposed a theoretical model
[ll] for the detection of low contrast lesions which yields an area-wise
SNR
where C is the lesion contrast, d the lesion diameter, N the number of
independent samples of the speckle pattern via spatial or frequency
compounding, Scx is the speckle size in the lateral direction which is
proportional to the lateral resolution and S c z the speckle size in the
axial direction. Using Eq. (7), for high speed spatial compounding with a
one-dimensional linear array, Trahey et al. [17] have found a maximum of
3.2 independent samples (N = 3 . 2 ) of the speckle for a factor of two loss
of lateral resolution, i.e., Scx increased by a factor of 2. By Eq. ( a ) ,
then, the increase in area-wise SNR for spatial compounding is ( 3 . 2 / 2 ) 1 ' 2 =
1.26.
Patterson et al. [91 proposed the contrast-to-speckle ratio as a
measure of cyst detection in a speckle background, i.e.,
si - so
CSR =
(Ui
2
+
P2
uo2
where, for many independent image planes of a cylindrical cyst, Si is the
mean signal inside the cysts, So is the mean signal outside the cysts, ui2
is the variance of the many measurements inside the cysts and uo2 is the
variance of the measurements outside the cysts.
Kerr et al. [lo] have shown experimentally that for the signal
processing techniques discussed in Equations (1) to (3) above, the best
processor is the "phase insensitive sector addition" of Eq. (1) which
yields experimental results showing a factor of 1.66 increase in CSR.
Using the data of Patterson et al. 191, we have shown previously [181 that
the CSR parameter is approximately equivalent to our areawise SNR. Thus,
166
Downloaded from uix.sagepub.com at PENNSYLVANIA STATE UNIV on March 5, 2016
MALTESE CROSS PROCESSOR
we can also use the data of Kerr et a1 in Eq. ( 8 ) . e.g. for 8
approximately independent sectors (N=8), there is approximately a factor of
three loss of lateral resolution, i.e., SCx increased by a factor of three.
The increase in the theoretical areawise SNR of Fq. 8 is then (8/3)1'2=
1.63, which is in agreement with the experimental results of Kerr et al.
In the case of the Maltese cross, the six operating ccosses are
partially correlated and yield the equivalent of four independent samples
of the speckle pattern (N = 4 ) . As a result of the multiplicative
processing of the orthogonal arms, there is evidently no loss of lateral
resolution as measured by the point response function. Thus, our model for
low contrast lesion detection (Eq. 8) predicts a factor of two improvement
in areawise SNR or CSR. Of course, this preliminary estimate needs to be
confirmed by lesion detection measurements.
There are two other interesting features of the Maltese cross
processor. First, unl i k ~s p a t i a l compounding systems using siihavrtnres in
a phased linear array, there seems to be little compounding improvement for
specular targets for the Maltese cross. This is seen in the image of the
MUM test phantom (Fig. 4 ) . Our preliminary Maltese cross images of cone
targets, which show a specular boundary due to trapped air, also show only
a slight improvement in the specular echoes.
The second difference from linear array systems is in the area of data
acquisition rate. High speed spatial compounding using N apertures require
N parallel processors to maintain real time image acquisition. We have
previously developed such a system [6] for spatial compounding in a phased
linear array which produced four independent receive lines for a single
transmit burst, but required four independent delay line prisms and
associated circuitry for each transducer channel. The Maltese cross
processsor, for a single focused piston transducer, requires only the
parallel multipliers and detectors for each cross to maintain adequate data
acquisition rate to achieve real time imaging.
Of course, there are still significant areas for improvement in the
Maltese ccoss processor. Our most significant problem during fabrication
was electrical and acoustic crosstalk between array elements which caused
high correlation coefficients between the six crosses. We eliminated much
of the electrical crosstalk by joining the array sectors to their 180"
opposite numbers on miniature circuit boards mounted on the transducer with
careful attention to ground planes. However, we believe a significant
component of the high correlation coefficients between neighboring crosses,
shown in figure 5, is still due to crosstalk.
167
Downloaded from uix.sagepub.com at PENNSYLVANIA STATE UNIV on March 5, 2016
SMITH AND VON RAMM
The available signal dynamic range of the Maltese cross needs to be
increased through the use of a better high-pass filter to suppress the
feedthrough in the multiplicative operation. Alternatively, digital
multiplication of the rf signals would eliminate the feedthrough problem.
When the feedthrough problem is solved, the side lobe levels in the
multiplicative processor need to be studied out to -FO& to examine
possible off axis energy which would degrade the image contrast.
Alternatively the side lobe levels can be investigated through theoretical
analysis and computer simulation.
The Maltese cross processor should be implemented on a real time
system to enable clinical trials. Such a next step could be achieved by
modifying a simple mechanical sector scanner with a fixed focus piston
transducer to produce as many crosses as desired. For example, if we use
conventional fabrication techniques for transducer arrays, it is possible
to segment the 19 nun circular crystal into as many as 128 sectors which
would produce 32 Maltese crosses. Based on the results of our
crosscorrelation measurements in figure 5, we would expect the equivalent
of approximately 20 independent samples of the speckle noise which would
increase the speckle SNR by a factor of 4.5, presumably with no loss in
lateral resolution. Alternatively, a more complex approach would be to
modify a phased array beam former to develop a mechanically steered Maltese
cross/annular array. For example, assuming a typical 64 channel system,
one could produce a four element annular array and eight Maltese crosses.
However this system would require 64 short delay lines for focusing and 16
multipliers and detectors. In either case, the Maltese cross shows the
promise of speckle noise reduction with minimal compromise of spatial
resolution for improved image quality.
ACKNOWLEDGEMENTS
We are grateful to David L. Daley of the Center for Devices and
Radiological Health for his assistance in circuit development.
Disclaimer
The mention of commercial products, their source, or their usein 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
[l]
Burckhardt, C.B., Speckle in ultrasound B-mode scans, IEEE Trans.
Sonics Ultrason. E 5 , 1-6 (1978).
168
Downloaded from uix.sagepub.com at PENNSYLVANIA STATE UNIV on March 5, 2016
MALTESE CROSS PROCESSOR
[2]
Carpenter, D.A., Dadd, M.J., and Kossoff, G., A multimode real time
scanner, Ultrasound Med. and Biol. -6, 279-284 (1980).
[3]
Berson, M., Roncin, A., and Porcelot, L., Compound scanning with an
electrically steered beam, Ultrasonic Imaging 1, 303-308 (1981).
[41
Ligtvoet, C.M., and Eversdijk, C.H., Real Time Compound Scanning, in
Excerpta Medica, No. 553, pp. 51-55, International Congress Series,
Amsterdam, (19811 .
[5]
Shattuck, D.P. and von Ramm, O.T., Compound scanning with a phased
array, Ultrasonic Imaging -4, 93-107 (1982).
[61
Shattuck, D.P., Weinshenker, M.D., Smith, S.W., and von Ramm, O.T.,
Explososcan: a parallel processing technique for high speed
ultrasound imaging with linear phased arrays, J. Acoust. SOC.
Amer.
- - 75, 1272-1282, (1984).
[7]
Trahey, G.E., Smith, S.W. and von Ramm, O.T., Speckle pattern
correlation with lateral aperture translation: experimental results
and implications for spatial compounding, IEEE Trans. Ultrasonics,
Ferroelec. Freq. Control UFFC-33, 257-264, (1986).
[8]
Matzuk, T. and Skolnick, M.L., Real Time Compound Scanner Using Four
Servo Controlled Transducers, in Proc. Ann. Meeting, AIUM p. 184
(abstract only) (AIUM, Bethesda, MD 1979).
[9]
Patterson, M.S. and Foster, F.S., Improvement and quantitative
assessment of B-mode images produced by an annular array/cone
hybrid, Ultrasonic Imaging 2, 195-213, (1983).
[lo]
Kerr, A.T., Patterson, M.S., Foster, F.S., and Hunt, J.W., Speckle
reduction in pulse echo imaging using phase insensitive and phase
sensitive signal processing techniques, Ultrasonic Imaging g, 11-28
( 19861.
[ll]
Smith, S.W., Wagner, R.F., Sandrik, J.M. and Lopez, H., Low Contrast
Detectability and Contrastmtail Analysis in Medical Ultrasound,
IEEE Trans. Sonics Ultrasonics SU-30, 164-173, (1983).
[12]
McKechnie, T.S., Speckle Reduction, in Laser Speckle and Related
Phenomena, pp. 123-170, J.C. Dainty, ed. , (Springer-Verlag,New
York, 1975).
169
Downloaded from uix.sagepub.com at PENNSYLVANIA STATE UNIV on March 5, 2016
[13]
American Institute of Ultrasound in Medicine, Standard 100 nun Test
Object Including Recommended Procedures for Its Use" Reflections
1, 74-91, (1975).
[14]
Madsen. E.L., Zagzebski, J.A., Insana, M.F., Burke, T.M. and Frank,
G.R., Ultrasonically tissue mimicking liver including the
frequency dependence of backscatter, Med. Phys. 2 703-710, (1982).
[15]
Smith, S.W., Insana, M.F., and Lopez, H., New contrast-detail
phantoms for improved precision in lesion detection measurements
(submitted to ultrasound in Medicine and Biology).
[16]
Parzen, E, Modern Probability Theory and Its Applications. (John
Wiley, New York 1960).
[17]
Trahey, G.E., Smith, S.W. and von Ramm, O.T., Speckle reduction in
medical ultrasound via spatial compounding, in Medical Imaging,
,vol. 626, pp. 290-300, (SPIE, Bellingham,
W, 1986).
[18]
Smith, S.W. and Wagner, R . F . , Ultrasound speckle size and lesion
signal to noise ratio: verification of theory, Ultrasonic Imaging 5
174-180, (1984).
170
Downloaded from uix.sagepub.com at PENNSYLVANIA STATE UNIV on March 5, 2016