IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control ,

vol. 59, no. 2,

February

2012

217

Reconstruction Algorithm for Improved

Ultrasound Image Quality
Bruno Madore and F. Can Meral
AbstractA new algorithm is proposed for reconstructing
raw RF data into ultrasound images. Previous delay-and-sum
beamforming reconstruction algorithms are essentially onedimensional, because a sum is performed across all receiving
elements. In contrast, the present approach is two-dimensional,
potentially allowing any time point from any receiving element to contribute to any pixel location. Computer-intensive
matrix inversions are performed once, in advance, to create
a reconstruction matrix that can be reused indefinitely for
a given probe and imaging geometry. Individual images are
generated through a single matrix multiplication with the raw
RF data, without any need for separate envelope detection
or gridding steps. Raw RF data sets were acquired using a
commercially available digital ultrasound engine for three imaging geometries: a 64-element array with a rectangular fieldof-view (FOV), the same probe with a sector-shaped FOV,
and a 128-element array with rectangular FOV. The acquired
data were reconstructed using our proposed method and a delay-and-sum beamforming algorithm for comparison purposes.
Point spread function (PSF) measurements from metal wires
in a water bath showed that the proposed method was able to
reduce the size of the PSF and its spatial integral by about 20
to 38%. Images from a commercially available quality-assurance phantom had greater spatial resolution and contrast when
reconstructed with the proposed approach.

I. Introduction

he reconstruction of raw RF data into ultrasound

images typically includes both hardware-based and
software-based operations [1]. One especially important
step, called delay-and-sum beamforming because it involves applying time delays and summations to the raw
RF signal, can be performed very rapidly on dedicated
hardware. Such hardware implementations of the delayand-sum beamforming algorithm made ultrasound imaging possible at a time when computing power was insufficient for entirely digital reconstructions to be practical.
However, improvements in computer technology and the
introduction of scanners able to provide access to digitized
RF signals [2][6] have now made digital reconstructions
possible. Software-based reconstructions are more flexible
[6][11], and may allow some of the approximations inherited from hardware-based processing to be avoided. A
main message of the present work is that delay-and-sum

Manuscript received February 22, 2011; accepted December 6, 2011.

Financial support from National Institutes of Health (NIH) grants
R21EB009503, R01EB010195, R01CA149342, and P41RR019703 is acknowledged. The content is solely the responsibility of the authors and
does not necessarily represent the official views of the NIH.
The authors are with the Department of Radiology, Brigham and
Womens Hospital, Harvard Medical School, Boston, MA (e-mail:
bruno@bwh.harvard.edu).
Digital Object Identifier 10.1109/TUFFC.2012.2182
08853010/$25.00

beamforming is not a particularly accurate reconstruction

algorithm, and that software-based remedies exist which
are capable of providing significant image quality improvements, especially in terms of spatial resolution and contrast.
Delay-and-sum beamforming is essentially a one-dimensional operation, as a summation is performed along
the receiver-element dimension of the (properly delayed)
RF data. Several improvements upon the basic scheme
have been proposed, whereby different weights are given
to different receiver elements in the summation to control
apodization and aperture. More recently, sophisticated
strategies have been proposed to adjust these weights in
an adaptive manner, based on the raw data themselves, to
best suit the particular object being imaged [12][17]. For
example, in the minimum variance beamforming method
[12][15], weights are sought that minimize the L2-norm of
the beamformed signal (thus making reconstructed images
as dark as possible), while at the same time enforcing a
constraint that signals at focus be properly reconstructed.
The overall effect is to significantly suppress signals from
undesired sources while preserving, for the most part, signals from desired sources, thus enabling several different
possible improvements in terms of image quality [18]. Such
adaptive beamforming approaches will be referred to here
as delay-weigh-and-sum beamforming, to emphasize that
elaborately selected weights are being added to the traditional delay-and-sum beamforming algorithm. It should
be noted that, irrespective of the degree of sophistication
involved in selecting weights, delay-weigh-and-sum beamforming remains essentially one-dimensional in nature. In
contrast, the approach proposed here is two-dimensional,
allowing any time sample from any receiver element to
potentially contribute, in principle at least, toward the
reconstruction of any image pixel. Image quality improvements are obtained as correlations between neighboring
pixels are accounted for and resolved. Reductions in the
size of the point spread function (PSF) area by up to
37% and increases in contrast by up to 29% have been
obtained here, compared with images reconstructed with
delay-and-sum beamforming.
In the proposed approach, the real-time reconstruction
process consists of a single step, a matrix multiplication
involving a very large and very sparse reconstruction matrix, without any need for separate envelope detection or
gridding steps. Throughout the present work, considerable attention is given to the question of computing load,
and of whether computing power is now available and affordable enough to make the proposed approach practical
on clinical scanners. Computing demands for reconstruc-

2012 IEEE

218

IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control ,

tion algorithms can be separated into two very different

categories: Calculations performed only once, in advance,
for a given transducer array and imaging geometry, and
calculations that must be repeated in real-time for every
acquired image. Although long processing times may be
acceptable for the first category (algorithms performed
only once, in advance), fast processing is necessary for
the second category (real-time computations). Although
the present method does involve a heavy computing load,
most of these calculations fit into the first category, i.e.,
for a given probe and FOV geometry, they must be performed only once. The actual real-time reconstruction
involves multiplying the raw data with the calculated reconstruction matrix. In the present implementation, depending on field-of-view (FOV) and raw data set sizes,
this multiplication took anywhere between about 0.04 and
4s per time frame. In the future, further algorithm improvements and/or greater computing power may enable
further reductions in processing time. Even in cases where
frame rates sufficient for real-time imaging could not be
reached, the present method could still prove useful for
reconstructing the images that are saved and recorded as
part of clinical exams.
It may be noted that the proposed approach is not
adaptive, in the sense that the reconstruction matrix depends only on the probe and on the geometry of the imaging problem, and does not depend on the actual object
being imaged. As such, the proposed approach should not
be seen as a competitor to adaptive beamforming, but
rather as a different method, addressing different limitations of the traditional delay-and-sum bemforming algorithm. Ideas from the adaptive beamforming literature
may well prove compatible with the present approach,
thus enabling further improvements in image quality. More
generally, the 2-D approach proposed here might prove a
particularly suitable platform for modeling and correcting
for image artifacts, to obtain improvements beyond the
increases in spatial resolution and contrast reported in the
present work.
II. Theory
A. Regular (Delay-and-Sum Beamforming)
Reconstructions
Ultrasound imaging (USI) reconstructions typically
rely on delay-and-sum beamforming to convert the received RF signal into an image [1]. Other operations such
as time gain compensation (TGC), envelope detection and
gridding may complete the reconstruction process. The
acquired RF signal can be represented in a space, here
called et space, where e is the receiver element number
and t is time. This space is either 2- or 3-dimensional,
for 1-D or 2-D transducer arrays, respectively. A single
et space matrix can be used to reconstruct a single ray,
multiple rays, or even an entire image in single-shot imaging [19], [20]. In the notation used subsequently, all of the

vol. 59, no. 2,

February

2012

points in a 2-D (or 3-D) et space are concatenated into a

single 1-D vector, s. The single-column vector s features
Nt Ne rows, i.e., one row for each point in et space,
where Ne is the number of receiver elements and Nt is the
number of sampled time points. As further explained below, a regular delay-and-sum beamforming reconstruction
can be representedas

o = A{G V {D 0 T0 s}} = A{G V {R 0 s}}, (1)

where o is the image rendering of the true sonicated object

o; it is a 1-D vector, with all Nx Nz voxels concatenated
into a single column. TGC is performed by multiplying
the raw signal s with the matrix T0, which is a diagonal
square matrix featuring Nt Ne rows and columns. Delayand-sum beamforming is performed by further multiplying with D0, a matrix featuring Nl Nax rows and Nt
Ne columns, where Nl is the number of lines per image and
Nax is the number of points along the axial dimension. The
content and nature of D0 will be described in more detail
later. The operator V{} performs envelope detection,
which may involve non-linear operations and thus could
not be represented in the form of a matrix multiplication.
Gridding is performed through a multiplication with the
matrix G featuring Nx Nz rows and Nl Nax columns.
The operator A{} represents optional image-enhancement
algorithms, and will not be further considered here. The
reconstruction matrix, R0, is given by D0 T0. An example of an RF data set in et space and its associated
reconstructed image o (rendered in 2-D) is shown in Figs.
1(a) and 1(b), respectively. Fig. 1(b) further includes
graphic elements to highlight regions-of-interest (ROIs)
that will be referred to later in the text (white boxes and
line). A main goal of the present work is to improve R0 in
(1) as a means to increase the overall quality of reconstructed images o, such as that in Fig. 1(b).
B. An Improved Solution to the Reconstruction Problem
As assumed in delay-and-sum beamforming reconstructions, the signal reflected by a single point-object should
take on the shape of an arc in the corresponding et space
RF signal. The location of the point-object in space determines the location and curvature of the associated arc
in et space. For a more general object, o, the raw signal
would consist of a linear superposition of et space arcs,
whereby each object point in o is associated with an et
space arc in s. The translation of all object points into a
superposition of et space arcs can be describedas

T0 s = E arc o, (2)

where Earc is an encoding matrix featuring Nt Ne rows

and Nl Nax columns. The matrix Earc is assembled by
pasting side-by-side Nl Nax column vectors that correspond to all of the different et space arcs associated with
the Nl Nax voxels to be reconstructed. The reconstruction process expressed in (1) is actually a solution to the

madore and meral: reconstruction algorithm for improved ultrasound image quality

Fig. 1. (a) Raw RF data in the element-versus-time space, called et

space here, can be reconstructed into the image shown in (b) using a
regular delay-and-sum beamforming reconstruction. Note that the carrier frequency of the raw signal was removed in (a) for display purposes,
because it would be difficult to capture using limited resolution in a
small figure (the raw data were made complex and a magnitude operator
was applied). Regions of interest are defined in (b) for later use (white
boxes and line).

imaging problem from (2): Multiplying both sides of (2)

with Earc+, the Moore-Penrose pseudo-inverse of Earc, one
obtains o Earc+ T0 s. Upon adding the envelope
detection and gridding steps, one can obtain (1) from (2)
given that

D 0 = E arc +. (3)

Alternatively, the operations involved in a digital delayand-sum beamforming reconstruction (i.e., multiplying
the raw signal in et space with an arc, summing over
all locations in et space, and repeating these steps for a
collection of different arcs to reconstruct a collection of
different image points) can be performed by multiplying
the RF signal with EarcH, where the superscript H represents a Hermitian transpose. In other words, D0 in (1) is
givenby

D 0 = E arc H. (4)

Combining (3) and (4) gives a relationship that captures

one of the main assumptions of delay-and-sum beamforming reconstructions:

E arc + = E arc H. (5)

In other words, delay-and-sum beamforming reconstructions assume that assembling all et space arcs together
in a matrix format yields an orthogonal matrix. This assumption is badly flawed, as we will show.

219

Fig. 2. (a) A point in the imaged object is typically assumed to give rise
to an arc in et space. (b) However, reconstructing the arc in (a) with
delay-and-sum beamforming does not yield a point in the image plane,
but rather a spatially broad distribution of signal. (c) At least in principle, when applied to artificial signals such as that in (a), the proposed
approach can reconstruct images that are vastly improved in terms of
spatial resolution compared with delay-and-sum beamforming.

An example is depicted in Fig. 2 in which the et signal

consists of a single arc [Fig. 2(a)], meaning that the object
o should consist of a single point. However, reconstructing
this arc using (1) and (4) does not give a point-like image,
but rather the broad distribution of signals shown in Fig.
2(b). This is because EarcH, and thus D0 through (4), is
only a poor approximation to Earc+.
Instead of using the approximation from (5), the imaging problem as expressed in (2) can be better handled
through a least-squares solution. Eq. (2) is first multiplied
from the left by EarcH to obtain the so-called normal equations [21]: EarcH (T s) = EarcH Earc o. Inverting
the square normal matrix EarcH Earc and multiplying
from the left with (EarcH Earc)1 allows o to be estimated: o = (EarcH Earc)1 EarcH (T s). Upon the
addition of a damped least-squares regularization term
2L, and the insertion of 1 as part of a preconditioning
term EarcH 1, an equation is obtained which exhibits
the well-known form of a least-squares solution [7], [11],
[22],[23]:

D1 = (E arc H 1 E arc + 2L)1 E arc H 1,

(6)
o = G V {D1 T1 s} = G V {R 1 s}.

In the present work, is simply an identity matrix, and

L will be discussed in more detail later. The signal s may
here include both legitimate and noise-related components, and o is a least-squares estimate of the actual object o. The image in Fig. 2(c) was reconstructed using (6).
Compared with Fig. 2(b), Fig. 2(c) presents a much more
compact signal distribution and a greatly improved rendering of the point-object. But even though images reconstructed using the R1 matrix [e.g., Fig. 2(c)] may prove

220

IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control ,

greatly superior to those reconstructed with delay-andsum beamforming and the associated R0 matrix [e.g., Fig.
2(b)] when dealing with artificial et space data such as
those in Fig. 2(a), such improvements are typically not
duplicated when using more realistic data. The reason for
this discrepancy is explored in more detail in the next section.
C. Including the Shape of the Wavepacket in the Solution
Data sets acquired from a single point-like object do
not actually look like a simple arc in et space. In an actual data set, the arc from Fig. 2(a) would be convolved
with a wavepacket along t, whereby the shape of the wavepacket depends mostly on the voltage waveform used at
the transmit stage and on the frequency response of the
piezoelectric elements. A wavepacket has both positive
and negative lobes, whereas the arc in Fig. 2(a) was entirely positive. Even though the delay-and-sum beamforming assumption in (5) is very inaccurate, negative errors
stemming from negative lobes largely cancel positive errors from positive lobes. For this reason, delay-and-sum
beamforming tends to work reasonably well for real-life
signals, even though it may mostly fail for artificial signals
such as those in Fig. 2(a).
The reconstruction process from (6) avoids the approximation made by delay-and-sum beamforming as expressed
in (5), but it remains inadequate because it is based on
Earc, and thus assumes object points to give rise to arcs
in et space. Whereas Earc associates each point in the
object o with an arc in the raw signal s through (2),
an alternate encoding matrix Ewav associates each point
with a wavepacket function instead. Because Ewav features
several nonzero time points per receiver element, the reconstruction process truly becomes two-dimensional in
nature, as whole areas of et space may be used in the
reconstruction of any given pixel location, as opposed to
one-dimensional arc-shaped curves as in delay-and-sum
beamforming. Sample computer code to generate Ewav is
provided in the Appendix.
The solution presented in (6) can be rewritten using
a more accurate model relying on Ewav rather thanEarc:

vol. 59, no. 2,

February

2012

ber of reconstructed voxels Nvox is simply equal to Nx

Nz, whereas for a sector-shaped FOV, it is only about
half of that (because of the near-triangular shape of the
FOV). As shown in Section IV and in the Appendix, a
prior measurement of the wavepacket shape, for a given
combination of voltage waveform and transducer array,
can be used to generate Ewav. Note that unlike Earc, Ewav
is complex.
Eq. (8), below, is the final step of the present derivation. The index 2 from (7) can now be dropped without
ambiguity, and a scaling term (I + 2L) is introduced,
where I is an identity matrix, to compensate for scaling
effects from the 2L regularization term:
D = (E wav H 1 E wav + 2L)1 (I + 2L)
E wav H 1,

(8)

o = D T s = R s.
Based on (8), an image o can be generated in a single
processing step by multiplying the raw RF signal s with a
reconstruction matrixR.
D. Geometric Analogy
There is a simple geometric analogy that may help provide a more intuitive understanding of the least-squares

solution
in (8). Imagine converting a 2-D vector, s = s 1i

+ s 2j , from
 the usual xy reference system defined by unit
vectors i and j to a different
 reference system
 by

defined


the unit vectors u 1 = (i + j )/ 2 and u 2 = (i j )/ 2
instead. This can be done through projections, using a dot

  
product: s = l (s u l )u l . Projections are appropriate in

this case because the basis vectors u l form an orthonormal
 
set: u l u k = lk. In contrast, projections would not be ap
propriate when converting s to a non-orthonormal
refer


v
ence system,
such
as
that
defined
by
=
(
i
+
j
)/
2
and
1



by v 2 = i . In such case, the coefficients s and s in s =


s v 1 + s v 2 should instead be obtained by solving the following system of linear equations:

(ss ) = 11//
1
2

1
D 2 = (E wav E wav + L) E wav ,
s 1/ 2 1 s 1
s
0
2
2
s
1
2
=


.
=
=
s 1/ 2 0 s 2
1 1 s 2
s1 s 2
o = D 2 T2 s = R 2 s,
(7) (9)
H

where the TGC term T2 may be equated to T1 in (6), T2

= T1. Note that no envelope detection and no gridding
operation are required in (7), unlike in (1) and (6). Because Ewav already contains information about the shape
of the wavepacket, envelope detection is effectively performed when multiplying by D2. Furthermore, because a
separate envelope detection step is not required, there is
no longer a reason to reconstruct image voxels along ray
beams. Accordingly, the Nvox reconstructed voxels may
lie directly on a Cartesian grid, removing the need for a
separate gridding step. For a rectangular FOV, the num-

2 1 s
;
2 0 s

( ) (

)( ) (

Direct substitution confirms that s = 2s 2 and s =


(s1 s2) is the correct
here, because it leads to s
 solution



= s v 1 + s v 2 = s 1i + s 2j . The delay-and-sum beamforming reconstruction algorithm, which attempts to convert
et space signals into object-domain signals, is entirely
analogous to a change of reference system using projections. Every pixel is reconstructed by taking the acquired
signal in et space and projecting it onto the (arc-shaped)
function associated with this particular pixel location.
Such a projection-based reconstruction algorithm neglects
any correlation that may exist between pixels, and a bet-

madore and meral: reconstruction algorithm for improved ultrasound image quality

ter image reconstruction algorithm is obtained when taking these correlations into account, as is done in (8) and
(9).
E. Generalization to Multi-Shot Imaging
As presented in Section II-C, (8) involved reconstructing a single et space data set s from a single transmit
event into an image o. However, (8) can readily be generalized to multi-shot acquisitions, whereby transmit beamforming is employed and only part of the image is reconstructed from each transmit event. In such a case, data
from all Nshot shots are concatenated into the columnvector s, which would now feature Nshot Nt Ne elements. The number of columns in the reconstruction matrix R also increases to Nshot Nt Ne. In the simplest
scenario, in which any given image voxel would be reconstructed based on RF data from a single transmit event
(rather than through a weighted sum of multiple voxel
values reconstructed from multiple transmit events), the
number of nonzero elements in R would remain the same
as in the single-shot imaging case. As shown in Section IV,
the number of nonzero elements in R is the main factor
determining reconstruction time. Although the increased
size of s and R may cause some increase in reconstruction
time, the fact that the number of nonzero elements in the
sparse matrix R would remain unchanged suggests that
the increase in reconstruction time may prove to be modest.
F. On Extending the Proposed Model
The present work offers a framework whereby information anywhere in the et space can be used, in principle
at least, to reconstruct any given image pixel. This more
flexible, two-dimensional approach may lend itself to the
modeling and correction of various effects and artifacts.
Two possible examples are considered.
1) Multiple Reflections: The number of columns in the
encoding matrix Ewav could be greatly increased to include not only the et space response associated with each
point in the reconstructed FOV, but also several extra
versions of these et space responses shifted along the t
axis, to account for the time delays caused by multiple
reflections. Such an increase in the size of Ewav could,
however, lead to a prohibitive increase in memory requirements and computing load.
2) Effect of Proximal Voxels on More Distal Voxels: Ultrasound waves are attenuated on their way to a distal
location, which may give rise to the well-known enhancement and shadowing artifacts, but they are also attenuated on their way back to the transducer, which may affect the et space function associated with distal points.
For example, whole segments of the et space signal might
be missing if one or more proximal hyperintense object(s)
would cast a shadow over parts of the transducer face.

221

The model being solved through (8) is linear, and cannot

account for the exponential functions required to represent attenuation. One would either need to opt for a different type of solution, possibly an iterative solution, or to
make the model linear through a truncated Taylor series,
ex (1 + x). We did pursue the latter approach, and obtained encoding matrices the same size as Ewav, to be used
in solutions of the same form as (8). At least in its current
form, the approach proved to be impractical for two main
reasons: 1) Although no bigger than Ewav in its number of
rows and columns, the new encoding matrix was much less
sparse than Ewav, leading to truly prohibitive reconstruction times and memory requirements; and, perhaps more
importantly, 2) the encoding matrix becomes dependent
on the reconstructed object itself, so that most of the
processing would have to be repeated in real-time for each
time frame, rather than once, in advance.
III. Methods
A. Experimental Setup and Reconstruction
All data were acquired with a Verasonics V-1 system
(Redmond, WA) with 128 independent transmit channels
and 64 independent receive channels. Two different ultrasound phantoms were imaged: A CIRS 054GS phantom
(Norfolk, VA) with a speed of sound of 1540m/s and an
attenuation coefficient of 0.50 0.05dB/cmMHz, and
a homemade phantom consisting of a single metal wire
in a water tank. Two different probes were employed: an
ATL P42 cardiac probe (2.5MHz, 64elements, pitch of
0.32mm; Philips Healthcare, Andover, MA) and an Acuson probe (3.75 MHz, 128 elements, pitch of 0.69 mm; Siemens Healthcare, Mountain View, CA). The ATL probe
was used either in a rectangular-FOV mode (all elements
fired simultaneously) or in a sector-FOV mode (virtual
focus 10.24mm behind the transducer face), whereas the
Acuson probe was used only in a rectangular-FOV mode.
When using the Acuson probe, signal from the 128 elements had to be acquired in two consecutive transmit
events, because the imaging system could acquire only 64
channels at a time. The system acquired 4 time samples
per period, for a 107s temporal resolution with the ATL
probe and 6.7 108s with the Acuson probe. About
2000 time points were acquired following each transmit
event (either 2048 or 2176 with the ATL probe, and 2560
with the Acuson probe). The reconstructed FOV dimensions were 2.05 14.2cm for the ATL probe in a rectangular-FOV mode, 17.7 11.0cm for the ATL probe in
a sector-FOV mode, and 8.77 10.1cm for the Acuson
probe in a rectangular-FOV mode.
Reconstruction software was written in the Matlab programming language (The MathWorks Inc., Natick, MA)
and in the C language. Sample code is provided in the
Appendix for key parts of the processing. The reconstruction problem from (8) was solved using either an explicit
inversion of the term (EwavH 1 Ewav + 2L), or

222

IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control ,

a least-squares (LSQR) numerical solver. Although the

LSQR solution was vastly faster than performing an explicit inverse, and proved very useful throughout the developmental stages of the project, it would be impractical
in an actual imaging context as the LSQR solution requires the actual RF data, and thus cannot be performed
in advance. In contrast, performing the explicit inverse
may take a long time, but it is done once, in advance,
and the result can be reused indefinitely for subsequent
data sets, potentially allowing practical frame rates to be
achieved. Reconstruction times quoted in SectionIV were
obtained using either an IBM workstation (Armonk, NY)
model x3850 M2 with 4 quad-core 2.4-GHz processors and
128GB of memory, or a Dell workstation (Round Rock,
TX) Precision T7500 with 2 quad-core 2.4-GHz processors
and 48 GB of memory. The Dell system was newer and
overall significantly faster, allowing shorter reconstruction
times to be achieved, whereas the IBM system proved
useful for early development and for reconstruction cases
with greater memory requirements.
B. Optimization of Reconstruction Parameters
1) Time Gain Compensation: The data set shown in
Fig. 1 was reconstructed several times using (1) and (6)
while adjusting the TGC matrices T0 and T1 from one
reconstruction to the next. The different TGC matrices
were computed based on different attenuation values, in
search of matrices which were able to generate fairly homogeneous image results. The effects of reconstructing images one column or a few columns at a time rather than
all columns at once were also investigated, as a way of
gaining insights into the proposed algorithm.
2) Regularization: The data set shown in Fig. 1 was
reconstructed using several different settings for the regularization term 2L in (8), by varying the value of the
scalar 2 and using L = I, an identity matrix. Although
a single time frame was shown in Fig. 1, the full data set
actually featured Nfr = 50 time frames. A standard deviation along the time-frame axis was calculated, and ROIs
at various depths were considered [as shown in Fig. 1(b)].
As a general rule, the regularization term should be kept
as small as possible to avoid blurring, but large enough
to avoid noise amplification if the system becomes ill conditioned. A depth-dependent regularization term 2L is
sought, with L I, whereby an appropriate amount of
regularization is provided at all depths.
3) Maintaining Sparsity: The proposed approach involves manipulating very large matrices featuring as many
as hundreds of thousands of columns and rows. The approach may nevertheless prove computationally practical
because these matrices, although large, tend to be very
sparse. Only the nonzero elements of a sparse matrix need
to be stored and manipulated, and one needs to make sure
that all matrices remain sparse at all times throughout
the reconstruction process. If large amounts of nearly-zero

vol. 59, no. 2,

February

2012

Fig. 3. To reduce computing requirements, processing is performed

over several overlapping patches rather than for the whole field-of-view
(FOV) at once. Results from all patches can be combined into a single
reconstruction matrix R. Examples in the xz plane are shown for all
three FOV geometries used in the present work. (a) ATL 64-element; (b)
ATL 64-element (sector-shaped FOV); (c) Acuson, 128-element.

(but nevertheless nonzero) elements were generated at any

given processing step, processing time and memory requirements could easily grow far beyond manageable levels. Three main strategies were used to ensure sparsity.
First, as shown in Section IV, the wavepackets used as
prior knowledge when constructing Ewav were truncated
in time to keep only Nwpts nonzero points, to help keep
Ewav sparse. Second, instead of solving for all of D in one
pass, the areas of D where nonzero values are expected
were covered using a series of Npatch overlapping regions,
each one only a small fraction of D in size. In the image
plane, these patches can be thought of as groups of voxels
that are located roughly the same distance away from the
virtual transmit focus. For rectangular-FOV geometries,
different patches simply correspond to different z locations
[Fig. 3(a)] and additional reductions in processing requirements can be achieved by further sub-dividing the x-axis
as well [Fig. 3(c)]; for sector-shaped FOV geometries, the
patches correspond to arc-shaped regions in the xz plane
[Fig. 3(b)]. Alternatively, these patches can be understood
as square sub-regions along the diagonal of the square
matrix (EwavH Ewav + 2L)1, which are mapped onto
the non-square D and R matrices through multiplication
with EwavH in (8). Third, once all patches are assembled
into a D or R matrix, a threshold is applied to the result
so that only the largest Nnz values may remain nonzero.
Preliminary thresholding operations may also be applied
to individual patches. Smaller settings for Nnz lead to
sparser R matrices and shorter reconstruction times, but
potentially less accurate image results. The need for fast
reconstructions must be weighed against the need for image accuracy.
After selecting a reasonable setting of Npatch = 20 for
the data set in Fig. 1 (see Fig. 3a), images were generated
using several different values for Nnz while noting the effect on reconstruction speed and accuracy. The so-called
artifact energy was used as a measure of image accuracy:

E N nz =

voxels

o N nz o ref

)/(

voxels

o ref

), (10)

madore and meral: reconstruction algorithm for improved ultrasound image quality

where o N nz and o ref were obtained with and without

thresholding, respectively. The number of nonzero elements in R should be made as small as possible to achieve
shorter reconstruction times, but kept large enough to
avoid significant penalties in terms of image quality and
artifact energy.
C. Comparing Reconstruction Methods
1) PSF: The metal-wire phantom was imaged using the
ATL P42 cardiac probe both in a rectangular-FOV and a
sector-FOV mode, and the Acuson probe in a rectangularFOV mode. The acquired data sets were reconstructed
using both delay-and-sum beamforming [(1) and (4)] and
the proposed approach [(8)]. Earc in (4) consisted of about
Nvox Ne nonzero elements, in other words, one nonzero
element per receive channel for each imaged pixel. Any
interpolation performed on the raw data would be builtin directly into Earc, and would lead to an increase in the
number of nonzero elements. Interpolating the raw data
might bring improvements in terms of secondary lobe suppression [24], but would also degrade the sparsity of Earc
by increasing the number of nonzero elements.
Because the water-wire transition had a small spatial
extent, the resulting images were interpreted as a PSF.
The full-width at half-maximum (FWHM) of the signal
distribution was measured along the x and z axes, giving
FWHMx and FWHMz. The size of the PSF was interpreted here as the size of its central lobe, as approximated
by ( FWHMx FWHMz/4). A second measurement
was performed which involved the whole PSF distribution,
rather than only its central lobe: After normalizing the
peak signal at the wires location to 1.0 and multiplying
with the voxel area, the absolute value of the PSF signal
was summed over an ROI about 3cm wide and centered
at the wire. The result can be understood as the minimum area, in square millimeters, that would be required
to store all PSF signal without exceeding the original peak
value anywhere. This measure corresponds to the L1-norm
of the PSF, and along with the size of the central lobe it
was used here to compare PSF results obtained from different reconstruction methods.
2) Phantom Imaging: The CIRS phantom was imaged
using the same probes as for the metal-wire phantom described previously, and the data sets were reconstructed
using both delay-and-sum beamforming (1) and the proposed approach (8). Resulting images were displayed sideby-side for comparison. Small hyperechoic objects allowed
differences in spatial resolution to be appreciated, whereas
a larger hyperechoic object allowed differences in contrast
to be measured.
IV. Results
A. Optimization of Reconstruction Parameters
1) Equivalence of Implementations: Fig. 4 shows 1-D
images obtained from the same data set as in Fig. 1, for a

223

1-D FOV that passes through the line of beads from Fig.
1(b). Of particular interest are the 3 results plotted with
the same black line in Fig. 4(a). These results are indistinguishable in the sense that differences between them were
much smaller than the thickness of the black line in Fig.
4(a). One was obtained using delay-and-sum beamforming and R0 from (1), the two others using R1 and (6),
including only one ray (i.e., one image column) at a time
into the encoding matrix. Results from (6) diverged from
delay-and-sum beamforming results only when many or
all image rays were included at once in the same encoding matrix [gray and dashed lines in Fig. 4(a)]. The main
point is that differences between our approach and delayand-sum beamforming reported here do not appear to
come from one being better implemented than the other,
but rather from our method resolving the correlation between adjacent voxels and rays, as it was designed todo.
2) Time Gain Compensation: Fig. 4(b) shows that with
delay-and-sum beamforming and R0 in (1), a TGC term
based on a 0.30dB/cmMHz attenuation seemed appropriate, as it would keep the amplitude of the various beads
in Fig. 1(b) roughly constant with depth. On the other
hand, when using R1, a correction based on a higher attenuation of 0.50dB/cmMHz proved more appropriate.
Documentation on the CIRS phantom lists the true, physical attenuation as 0.50 0.05dB/cmMHz the same
value used here with our proposed reconstruction method. It would appear that with the proposed reconstruction, TGC might become a more quantitative operation
based on true signal attenuation. However, as shown in
Fig. 4(b) (gray arrow), signals at shallow depths tend to
be overcompensated when employing a value of 0.50dB/
cmMHz. To prevent the near-field region from appearing
too bright in the images presented here, further ad hoc
TGC was applied over the shallower one-third of the FOV.
Furthermore, an ad hoc value of 0.35dB/cmMHz (rather
than 0.50dB/cmMHz) had to be used when reconstructing data from the higher-frequency Acuson array, so that
homogeneous-looking images could be obtained. Overall,
although the TGC operation does appear to become more
quantitative in nature with the proposed approach, ad hoc
adjustments could not be entirely avoided.
3) Regularization: The 50-frame data set from Fig. 1
was reconstructed several times using different values for
2, the regularization parameter. For each reconstruction,
the standard deviation along the time-frame direction was
computed and then spatially averaged over 5 ROIs located
at different depths [shown in Fig. 1(b) as white rectangular boxes]. Fig. 5 gives the mean standard deviation associated with each of these ROIs, as a function of the regularization parameter 2. For each curve in Fig. 5, an
indicates the amount of regularization that appears to be
roughly the smallest 2 values that can be used, while still
avoiding significant noise increases. Defining a normalized
depth r = ( x 2 + (z + d vf )2 d vf ) /w probe, where dvf is
the distance to the virtual focus behind the transducer

224

IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control ,

vol. 59, no. 2,

February

2012

Fig. 4. A single column from a phantom image, highlighted in Fig. 1(b), is plotted here for different reconstruction algorithms and settings. (a)
When reconstructing one column at a time, our modified reconstruction from (6) gives results that are essentially identical to the delay-and-sum
beamforming reconstruction from (1) (black solid line). As more columns are included in the reconstruction, our method diverges from delay-andsum beamforming (gray solid and black dashed lines). (b) With all columns included in the reconstruction, the TGC must be changed from 0.30 to
about 0.50dB/cmMHz to restore the magnitude at greater depths. The nominal attenuation value for this phantom is 0.50 0.05dB/cmMHz, in
good agreement with the TGC compensation required with our method. However, signal becomes overcompensated at shallow depths (gray arrow).
The plots use a linear scale, normalized to the maximum signal from the curve in(a).

and wprobe is the width of the transducer probe in the x

direction; the location of the marks in Fig. 5 correspond
to 2 = r/20. Because having no regularization at r = 0
might be problematic, a minimum value of 0.1 was used
for 2, so that 2 = max(r/20, 0.1). In practice, the regularization parameter in (8) was equated to the constant
part of this expression, 2 = 1/20, and the diagonal of the
Nvox by Nvox matrix L was equated to the variable part, so
that 2 diag(L) = max(rj /20,0.1), where j ranges from
1 to Nvox.
More generally, this expression cannot be expected
to hold for all FOV and probe geometries. For example,
when using the ATL probe in a sector-FOV mode rather
than the rectangular-FOV mode employed in Fig. 5, a
much larger number of voxels are reconstructed from essentially the same number of raw-data points, suggesting
that conditioning might be degraded and that a higher
level of regularization might prove appropriate. For both
the sector-FOV results and the Acuson-probe results presented here, regularization was scaled up by a factor of 4
compared with the previously given expression, leading to
2 diag(L) = max(rj /5,0.1).
4) Maintaining Sparsity: The data set from Fig. 1(a)
was reconstructed using the proposed method, and the
magnitude of the Nvox by (Ne Nt) matrix D is shown in
Fig. 6(a). Because D is very sparse, one can greatly decrease computing requirements by calculating only the regions with expected nonzero signals, using a series of overlapping patches. The calculated regions, where elements
can assume nonzero values, are shown in Fig. 6(b). R is
calculated from D, and the plots in Fig. 6(c) show the effect that thresholding R had on reconstruction time and
accuracy. The horizontal axis in Fig. 6(c) is expressed in

Fig. 5. A 50-frame data set was reconstructed several times, with different settings for the regularization parameter 2. The standard deviation
across all 50 frames was taken as a measure of noise, and averaged over
the 5 ROIs shown in Fig. 1(b). With d = z/wprobe, the ROIs were located
at a depth of d = 1.0, 2.5, 3.5, 4.5, and 5.5. For each ROI, the standard
deviation is plotted as a function of the regularization parameter 2, and
an indicates the 2 = d/20 setting selected here.

terms of Nnz0 = 7131136, the number of nonzero elements

in R0, as obtained when performing a regular delay-andsum beamforming reconstruction on the same data (1).
As seen in the upper plot in Fig. 6(c), reconstruction time
scales linearly with the number of nonzero elements in R
with a slope equivalent to 3.10 107 nonzero elements
per second, for the IBM workstation described in Section
III. Based on the lower plot, a setting of Nnz = 40 Nnz0
was selected, which is roughly the lowest value that can
be used while essentially avoiding any penalty in terms of
artifact energy. Compared with the non-thresholded case,
an Nnz = 40 Nnz0 setting allowed a three-fold increase

madore and meral: reconstruction algorithm for improved ultrasound image quality

225

in the reconstruction speed, at essentially no cost in image

quality.
B. Comparing Reconstruction Methods

Fig. 6. (a) The D matrix (8) tends to be very sparse. (b) The areas where
nonzero signal is expected are covered using many overlapping smaller
patches, greatly reducing the computing requirements compared with
solving for the entire D matrix all at once. (c) The R matrix in (8) and/
or the D matrix can be thresholded, so that only the Nnz largest values are allowed to remain nonzero. The smaller Nnz becomes, the faster
the reconstruction can proceed; about 32.3ms were needed for every
106 nonzero elements, using the IBM workstation described in the text.
However, thresholding that is too aggressive leads to increased artifact
content. A compromise was reached in this case for Nnz = 40 Nnz0 =
2.852 108 elements.

1) PSF: The wavepacket shapes used to calculate Ewav

are shown in Fig. 7(a) for all probe and FOV geometries
used here. As shown with black curves in Fig. 7(a), the
wavepackets were cropped to only Nwpts nonzero points
to help maintain sparsity in Ewav. In Fig. 7(a) and in all
reconstructions, a setting of Nwpts = 50 points was used.
Images of the wire phantom are shown in Figs. 7(b)7(d),
both for a delay-and-sum beamforming reconstruction (1)
and for our proposed method (8), along with profiles of
the PSFs along the x- and z-directions. All images shown
here are windowed such that black means zero or less,
white means signal equal to the window width w or more,
and shades of gray are linearly distributed between them.
Area and L1-norm measurements of the PSF are provided
in Table I; Table II lists reconstruction times and matrix
sizes. Note that delay-and-sum beamforming results were
reconstructed with very high nominal spatial resolution (
/8, Table II), to help ensure a fair comparison.
As seen in Table I, the size of the PSF was reduced by
up to 37% (ATL probe with rectangular FOV), and the
L1-norm of the PSF was reduced by up to 38% (Acuson
probe with rectangular FOV). Compared with the ATL
probe results, using the wider 128-element Acuson array and reducing the depth of the metal-wire location to
only about 4cm led to very compact PSF distributions
(0.32mm2, from Table I). In this case, our method had
very little room for improvement in terms of PSF size
(1% improvement, from Table I), but a 38% reduction of
the L1-norm of the PSF was achieved. Reduction of the
L1-norm means that less signal may leak away from highintensity features, and that higher contrast might be obtained, as verified subsequently using the CIRS phantom.
Bold numbers in Table I refer to an optimized reconstruction tailored to the metal-wire phantom, whereby a
1.5 1.5cm square region centered at the object-point
location was reconstructed using a single patch (Npatch
= 1) and high spatial resolution ( /4, where is
the wavelength = c/f). Such optimized reconstructions
were performed for comparison purposes on both data

TABLE I. Measurements of PSF Size and L1-Norm for Delay-And-Sum Beamforming and for the Proposed Approach for
Different Probes and FOV Settings.
Point-object
x-z location
(cm)
ATL P42
rectangular FOV
ATL P42
sector FOV
Acuson
rectangular FOV

Delay-and-sum Proposed method

central lobe
central lobe
(mm2)
(mm2)

(0.0, 9.2)

1.55

(0.0, 9.2)

1.51

(0.3, 3.7)

0.324

Bold indicates values for optimal processing.

0.973
0.972
1.26
1.23
0.322

Improvement
(%)
37.3
37.4
16.6
18.5
0.64

Delay-and-sum
L1-norm
(mm2)
9.20
11.3
6.13

Proposed
method
L1-norm
(mm2)

Improvement
(%)

6.64
6.53
8.28
8.11
3.81

27.8
29.0
26.7
28.3
37.8

226

IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control ,

vol. 59, no. 2,

February

2012

TABLE II. Matrix Sizes and Reconstruction Times, For Our Proposed Approach and for Delay-And-Sum Beamforming, for
Different Probes and FOV Settings.
Raw data size
ATL P42,
rectangular FOV
ATL P42,
sector FOV
Acuson,
rectangular FOV

64 2176
same
64 2048
same
128 2560
same

Image size

Npatch

Voxel size
(with = c/f)

Nnz

Reconstruction time,
stage 1

Reconstruction time,
stage 2

64 924
64 1850
286 716
286 1434
213 985
213 1971

10
1
15
1
3 40
1

pitch /4
pitch /8
/4
/8
/4
/8

3.03e8
7.40e6
1.05e9
2.58e7
3 8.58e8
4.75e7

13.6 h/0 s
31.62 s/0 s
66.7 h/0 s
4.17 min/0 s
3 52.2 hb/0 s
16.0 minb/0 s

0.039 0.004 s/frc

0.044 s/fr
1.70 0.06 s/frc
0.18 s/fra
3 (1.39 0.06) s/frc
0.30 s/fr

Bold indicates values for delay-and-sum beamforming.

not include gridding time.
bPerformed on the 128 GB IBM system.
cProcessed using a C program with 8 threads.
aDoes

Fig. 7. Imaging results from a metal-wire phantom are interpreted here in terms of a point-spread-function (PSF). (a) Prior knowledge about the
shape of the wavepacket is used as part of the reconstruction. (b)(d) Single-shot images reconstructed with delay-and-sum beamforming [R0 in (1)]
and with the proposed approach [R in (8)] are shown side-by-side. (b) ATL probe, rectangular field of view (FOV); (c) ATL probe, sector-shaped
FOV; (d) Acuson probe, rectangular FOV. All images are windowed such that black is zero or less, white is equal to the window width w or greater,
and all possible shades of gray are linearly distributed in-between. The ROIs indicated by white ellipses/circles were used for the calculations of the
L1-norms listed in Table I [3cm in diameter, 2cm minor diameter for the ellipse in (a)]. Gray boxes show the area surrounding the point-object using a window width w that is 1/4 that used for the corresponding main images, to better show background signals. Profiles across the location of the
point-object are also shown, along both the z- and x-directions, for delay-and-sum beamforming (gray curves) and for the proposed method (black
curves). All plots use a linear scale normalized to the maximum response.

madore and meral: reconstruction algorithm for improved ultrasound image quality

227

sets obtained with the ATL probe, with rectangular- and

sector-shaped FOV. Note that the optimum reconstruction brought very little further improvement in terms of
PSF size or L1-norm (see Table I, bold versus non-bold
numbers).
2) Phantom Imaging: Because the R0 and R matrices
had already been calculated for the PSF results shown in
Fig. 7, no further processing was required and the precomputed matrices were simply reused for reconstructing
the images in Fig. 8. The fact that no new processing was
needed is emphasized in Table II by entries of 0s in the
Recon time, stage 1 column.
A schematic of the imaged phantom is provided in Fig.
8(a), and single-shot images are shown in Figs. 8(b)8(d)
for both the delay-and-sum beamforming (R0 matrix) and
the present reconstruction method (R matrix), for the
3 imaging geometries considered here. The side-by-side
comparison appears to confirm that the present approach
[using (8)], succeeds in increasing spatial resolution compared with a delay-and-sum beamforming reconstruction
[using (1)], at least in results obtained with the 64-element
ATL probe. Using the wider 128-element Acuson probe,
the improvement in spatial resolution appears to be more
subtle. On the other hand, the reduction in the L1-norm
of the PSF (Table I) does appear to have detectable effects
in the images shown in Fig. 8(d). Using the ROIs defined
in Fig. 8(d), and with SC the mean signal over the inner
circular ROI and SR the mean signal over the ring-shaped
ROI that surrounds it, contrast for the hyperechoic circular region [arrow in Fig. 8(d)] was defined as (SC SR)/
(SC + SR). The inner circular ROI had an 8mm diameter,
equal to the known size of the phantoms hyperechoic target, and the surrounding ring-shaped ROI had an outer
diameter of 12mm. Because less of the signal was allowed
to bleed away from the hyperechoic region when using our
proposed reconstruction approach, contrast, as previously
defined, was increased by 29.2% compared with the delayand-sum beamforming results, from a value of 0.248 to a
value of 0.320.
V. Discussion
An image reconstruction method was presented that
offers advantages over the traditional delay-and-sum
beamforming approach. Without any increase in risk or
exposure to the patient, and without any penalty in terms
of ultrasound penetration, spatial resolution and contrast
could be increased through a more accurate reconstruction of the acquired data. The proposed reconstruction
process involved a single matrix multiplication without
any need for separate envelope detection or gridding steps,
and improvements by up to 38% in the area and the L1norm of the PSF were obtained for three different FOV
and probe configurations. The acquired data enabled a
quantitative characterization of the PSF at only a single
location within the imaged FOV, a limitation of the re-

Fig. 8. (a) Imaging results were obtained from the phantom depicted
here. (b)(d) Single-shot images reconstructed with delay-and-sum
beamforming [R0 in (1)] and with the proposed approach [R in (8)] are
shown side-by-side. (b) ATL probe, rectangular field of view (FOV); (c)
ATL probe, sector-shaped FOV; (d) Acuson probe, rectangular FOV.
A magnification of the region surrounding the axial-lateral resolution
targets is shown in (c) (the window width, w, was increased by 250% to
better show the individual objects). Overall, spatial resolution appears
to be improved in the images reconstructed with the proposed method
[i.e., with R in (8)]. Contrast was improved with the proposed method
in (d), as tested using the circular ROI covering the hyperechoic region
indicated with a white arrow and the ring-shaped region that surrounds
it. See the text for more detail.

sults presented here. A series of measurements involving

different relative positions between the imaging probe and
the imaged metal wire would be required if spatial maps
of PSF improvements were to be obtained, rather than
a single spatial location. Although more qualitative in
nature, results from a CIRS imaging phantom suggested
that improvements in PSF may be occurring throughout
the imaged FOV.
It is worth noting that the amounts of spatial resolution and contrast improvements reported here do not

228

IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control ,

necessarily represent a theoretical limit for the proposed

algorithm, but merely what could be achieved with the
present implementation. In principle at least, in a noiseless
case in which the encoding matrix is perfectly known, the
PSF could be reduced to little more than a delta function
[e.g., see Fig. 2(c)]. In more realistic situations, limitations
on the achievable spatial resolution result from inaccuracies in the encoding matrix, the need to use regularization, and limits on both memory usage and reconstruction time. It is entirely possible that with a more careful
design for Ewav and for the regularization term 2L, or
with greater computing resources, greater improvements
in spatial resolution and contrast might have been realized. On the other hand, in especially challenging in vivo
situations where processes such as aberration may affect
the accuracy of Ewav, lower levels of improvement might
be obtained instead. The possibility of including objectdependent effects such as aberration into Ewav, although
interesting, is currently considered impractical because of
the long processing time required to convert Ewav into a
reconstruction matrix R.
Prior information about the transmitted wavepacket
was obtained here from a single transducer element, during a one-time reference scan, using a phantom consisting
of a metal wire in a water tank. Interestingly, when using
the proposed reconstruction scheme, the TGC part of the
algorithm became more exact and less arbitrary in nature,
as the nominal 0.5dB/cmMHz attenuation coefficient of
the imaged phantom could be used directly to calculate
attenuation corrections. Scaling difficulties did however
remain, especially in the near field, and ad hoc corrections
could not be entirely avoided.
A main drawback of the proposed approach is its computing load. Although the real-time part of the processing
consists of a single multiplication operation between a matrix R and the raw data s, the R matrix tends to be very
large and the multiplication is computationally demanding. The use of graphics processing unit (GPU) hardware,
which enables extremely fast processing in some applications, may not be appropriate here. In current systems
at least, the graphics memory is still fairly limited and
physically separate from the main memory, meaning that
much time might be wasted transferring information to
and from the graphics card. Although GPU processing
may prove particularly beneficial in situations which require a large amount of processing to be performed on a
relatively small amount of data, it is not nearly as well
suited to the present case in which fairly simple processing (a matrix multiplication) is performed on a huge
amount of data (mainly, the matrix R). For this reason,
CPU hardware is used here instead, and a multi-threaded
reconstruction program was written in the C language.
Reconstruction times in the range of about 0.04 to 4s
per image were obtained here, using 8 processing threads
on an 8-processor system. Using more threads on a system featuring more cores is an obvious way of reducing
processing time. Further improvements in our programming and future improvements in computer technology

vol. 59, no. 2,

February

2012

may also help. If necessary, sacrifices could be made in

terms of voxel size, spatial coverage, or artifact content, to
further reduce the number of nonzero elements in R and
thus reduce processing time. It should be noted that even
in cases where frame rates required for real-time imaging
could not be achieved, the present method could still be
used to reconstruct images saved and recorded as part of
clinical ultrasound exams.
In contrast to the real-time operation R s, the processing speed for the initial one-time evaluation of R is
considered, for the most part, to be of secondary importance. In the present implementation, processing times
ranged from about 7h to much more than 100h, depending on probe and FOV geometry. Although reducing this time through algorithm improvements or parallel
processing would be desirable, it is not considered to be
an essential step toward making the method fully practical. Because these lengthy calculations can be re-used for
all subsequent images acquired with a given transducer,
excitation voltage waveform, and FOV setting, long initial
processing times do not prevent achievement of high frame
rates. In practice, several R matrices corresponding to
different transducers and a range of FOV settings can be
pre-computed, stored, and loaded when needed.
VI. Conclusion
An image reconstruction method was introduced that
enabled valuable improvements in image quality, and
computing times compatible with real-time imaging were
obtained for the simplest case considered here (0.039s
per frame). The method proved capable of reducing the
area and L1-norm of PSFs by up to about 38%, allowing improvements in spatial resolution and contrast at no
penalty in terms of patient risk, exposure, or ultrasound
penetration.
VII. Appendix
A. Generating the Ewav Matrix:
The generation of the Ewav matrix in (8) can be considered to be of central importance to the proposed approach.
Prior knowledge about the shape of the wavepacket [see
Fig. 7(a)], stored in a row-vector wvpckt featuring Nt
elements, is transformed here to the temporal frequency
domain and duplicated Ne times into the Ne Nt array
wvpckt_f:
wvpckt_f = repmat(fft(wvpckt,[],2), [Ne 1]);
For each voxel ivox to be reconstructed, a corresponding arc-shaped et space wavepacket function is calculated
through modifications to wvpckt_f. First, a travel_time
vector with Ne entries is obtained:
d_travel = d_to_object + d_from_object;
t_travel = d_travel/sound_speed;

madore and meral: reconstruction algorithm for improved ultrasound image quality

With the time point t_ref to be considered as the origin,

a phase ramp is placed on wvpckt_f that corresponds to
the appropriate element-dependent time shift:
t_travel = t_travel - t_ref;
ph_inc = -(2*pi/Nt) * (t_travel/dt + 1);
ph_factor = ph_inc * (0:Nt/21);
arc = zeros(Ne, Nt);
arc(:,1:Nt/2) = wvpckt_f(:,1:Nt/2).*exp(1i*ph_
factor);
arc = ifft(arc,[],2);
To maintain sparsity in Ewav, only a relatively small number of time points (50 here) can be kept for each wavepacket in arc [see Fig. 7(a)]. A sparser version of arc
is thus obtained, called arc_sparse, stored into a 1-D
column-vector featuring Nt Ne rows, and normalized so
that its L2-norm is equal to1:
E_1vox(:) = arc_sparse(:);
scaling = sqrt(sum(abs(E_1vox(:)).^2,1));
E_1vox(:) = E_1vox(:) ./ (scaling+epsilon);
Finally, the calculated result for voxel ivox can be stored
at its proper place within Ewav:
Ewav(:,ivox) = E_1vox(:,1);
The process is repeated for all ivox values, to obtain a
complete Ewav matrix.
B. Matrix Inversion and Image Reconstruction
For each patch within D (see Fig. 6b), and with E representing the corresponding region within Ewav, the inversion in (8) can be performed through
Ep = E;
EpE_inv = inverse(Ep*E + lambda_L);
EpE_inv = double(EpE_inv);
where inverse() is part of a freely-downloadable software
package developed by Tim Davis (http://www.mathworks.com/matlabcentral/fileexchange/24119). Alternatively, the readily available Matlab inv() function may
be used instead, although it is generally considered to be
less accurate:
EpE_inv = inv(Ep*E + lambda_L);
The reconstruction times provided in Table II were obtained using Matlabs inv() function. As shown in (8),
the (EwavH Ewav + 2L)1 term gets multiplied by (I +
2L) and by EwavH:
EpE_inv = EpE_inv * (speye(Nvox_patch,Nvox_
patch)+lambda_L);
D = EpE_inv * Ep;
Optional thresholding may be performed on D. The current patch, which involves all voxels listed into the array
i_vox, can then be stored at its proper place within the
matrixR:
R(i_vox,:) = R(i_vox,:) + W*D*T;

229

where T is TGC and W is a diagonal matrix with a Fermi

filter along its diagonal, to smoothly merge contiguous
overlapping patches. The matrix R is thresholded, and
the image corresponding to time frame ifr can be reconstructed with
s = zeros(Nt*Ne,1);
s(:) = data(:,:,ifr);
O_vec = R*s;
The Nvox by 3 array voxs is a record of the x, z, and
matrix location for every image voxel being reconstructed.
The 1-D vector O_vec gets converted into a ready-fordisplay 2-D image format through
O = zeros(Nz, Nx);
O(voxs(:,3)) = O_vec;
Acknowledgments
The authors thank Dr. G. T. Clement for allowing us
to use the Verasonics ultrasound system from his lab, as
well as Dr. R. McKie and Dr. R. Kikinis from the Surgical
Planning Lab (SPL) for providing us access to one of their
high-performance IBM workstations.
References
[1] K. E. Thomenius, Evolution of ultrasound beamformers, in Proc.
IEEE Ultrasonics Symp., San Antonio, TX, 1996, pp. 16151622.
[2] R. E. Daigle, Ultrasonic diagnostic imaging system with personal
computer architecture, U.S. Patent 5795297, Aug. 18, 1998.
[3] S. Sikdar, R. Managuli, L. Gong, V. Shamdasani, T. Mitake, T.
Hayashi, and Y. Kim, A single mediaprocessor-based programmable ultrasound system, IEEE Trans. Inf. Technol. Biomed., vol. 7,
no. 1, pp. 6470, 2003.
[4] F. K. Schneider, A. Agarwal, Y. M. Yoo, T. Fukuoka, and Y. Kim,
A fully programmable computing architecture for medical ultrasound machines, IEEE Trans. Inf. Technol. Biomed., vol. 14, no. 2,
pp. 538540, 2010.
[5] T. Wilson, J. Zagzebski, T. Varghese, Q. Chen, and M. Rao, The
Ultrasonix 500RP: A commercial ultrasound research interface,
IEEE Trans. Ultrason. Ferroelectr. Freq. Control, vol. 53, no. 10, pp.
17721782, 2006.
[6] M. Ashfaq, S. S. Brunke, J. J. Dahl, H. Ermert, C. Hansen, and M.
F. Insana, An ultrasound research interface for a clinical system,
IEEE Trans. Ultrason. Ferroelectr. Freq. Control, vol. 53, no. 10, pp.
17591771, 2006.
[7] J. Shen and E. S. Ebbini, A new coded-excitation ultrasound imaging systemPart I: Basic principles, IEEE Trans. Ultrason. Ferroelectr. Freq. Control, vol. 43, no. 1, pp. 131140, 1996.
[8] F. Lingvall and T. Olofsson, On time-domain model-based ultrasonic array imaging, IEEE Trans. Ultrason. Ferroelectr. Freq. Control, vol. 54, no. 8, pp. 16231633, 2007.
[9] M. A. Ellis, F. Viola, and W. F. Walker, Super-resolution image
reconstruction using diffuse source models, Ultrasound Med. Biol.,
vol. 36, no. 6, pp. 967977, 2010.
[10] D. Napolitano, C. H. Chou, G. McLaughlin, T. L. Ji, L. Mo, D.
DeBusschere, and R. Steins, Sound speed correction in ultrasound
imaging, Ultrasonics, vol. 44, suppl. 1, pp. e43e46, 2006.
[11] B. Madore, P. J. White, K. Thomenius, and G. T. Clement, Accelerated focused ultrasound imaging, IEEE Trans. Ultrason. Ferroelectr. Freq. Control, vol. 56, no. 12, pp. 26122623, 2009.
[12] J. Capon, High resolution frequency-wavenumber spectrum analysis, Proc. IEEE, vol. 57, no. 8, pp. 14081418, 1969.
[13] I. K. Holfort, F. Gran, and J. A. Jensen, High resolution ultrasound imaging using adaptive beamforming with reduced number of
active elements, Phys. Procedia, vol. 3, no. 1, pp. 659665, 2010.
[14] I. K. Holfort, F. Gran, and J. A. Jensen, Broadband minimum variance beamforming for ultrasound imaging, IEEE Trans. Ultrason.
Ferroelectr. Freq. Control, vol. 56, no. 2, pp. 314325, 2009.

230

IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control ,

[15] J. F. Synnevg, A. Austeng, and S. Holm, Adaptive beamforming

applied to medical ultrasound imaging, IEEE Trans. Ultrason. Ferroelectr. Freq. Control, vol. 54, no. 8, pp. 16061613, 2007.
[16] B. Mohammadzadeh Asl and A. Mahloojifar, Eigenspace-based
minimum variance beamforming applied to medical ultrasound imaging, IEEE Trans. Ultrason. Ferroelectr. Freq. Control, vol. 57, no.
11, pp. 23812390, 2010.
[17] C. I. Nilsen and I. Hafizovic, Beamspace adaptive beamforming
for ultrasound imaging, IEEE Trans. Ultrason. Ferroelectr. Freq.
Control, vol. 56, no. 10, pp. 21872197, 2009.
[18] J. F. Synnevg, A. Austeng, and S. Holm, Benefits of minimumvariance beamforming in medical ultrasound imaging, IEEE Trans.
Ultrason. Ferroelectr. Freq. Control, vol. 56, no. 9, pp. 18681879,
2009.
[19] D. P. Shattuck, M. D. Weinshenker, S. W. Smith, and O. T. von
Ramm, Explososcan: A parallel processing technique for high speed
ultrasound imaging with linear phased arrays, J. Acoust. Soc. Am.,
vol. 75, no. 4, pp. 12731282, 1984.
[20] K. F. stner, (2008, Aug.) High information rate volumetric ultrasound imagingAcuson SC2000 volume imaging ultrasound

vol. 59, no. 2,

February

2012

system. [Online]. Available: http://www.medical.siemens.com/siemens/en_US/gg_us_FBAs/files/misc_downloads/Whitepaper_Ustuner.pdf

[21] E. W. Weisstein, (2011, Jan.) Normal equation, in MathWorld
A Wolfram Web Resource. [Online]. Available: http://mathworld.
wolfram.com/NormalEquation.html
[22] P. C. Hansen, Direct regularization methods, in Rank-Deficient
and Discrete Ill-Posed Problems. Philadelphia, PA: SIAM Press,
1998, pp. 100.
[23] W. S. Hoge, D. H. Brooks, B. Madore, and W. E. Kyriakos, A tour
of accelerated parallel MR imaging from a linear systems perspective, Concepts in Magn. Reson. A, vol. 27A, no. 1, pp. 1737, 2005.
[24] C. R. Hazard and G. R. Lockwood, Theoretical assessment of a
synthetic aperture beamformer for real-time 3-D imaging, IEEE
Trans. Ultrason. Ferroelectr. Freq. Control, vol. 46, no. 4, pp. 972
980, 1999.
Authors photographs and biographies were unavailable at time of publication.