IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 55, NO.
6, JUNE 2008
Confocal Microwave Imaging for Breast Cancer
Detection: Delay-Multiply-and-Sum Image
Reconstruction Algorithm
Hooi Been Lim*, Nguyen Thi Tuyet Nhung, Er-Ping Li, Fellow, IEEE, and Nguyen Duc Thang
Abstract—A new image reconstruction algorithm, termed as
delay-multiply-and-sum (DMAS), for breast cancer detection us-
ing an ultra-wideband confocal microwave imaging technique
is proposed. In DMAS algorithm, the backscattered signals re-
ceived from numerical breast phantoms simulated using the finite-
difference time-domain method are time shifted, multiplied in pair,
and the products are summed to form a synthetic focal point. The
effectiveness of the DMAS algorithm is shown by applying it to
backscattered signals received from a variety of numerical breast
phantoms. The reconstructed images illustrate improvement in
identification of embedded malignant tumors over the delay-and-
sum algorithm. Successful detection and localization of tumors as
small as 2 mm in diameter are also demonstrated.
Index Terms—Breast cancer detection, finite-difference time-
domain (FDTD) method, microwave imaging, ultra-wideband
(UWB) radar.
I. INTRODUCTION
REAST cancer is the most common cancer affecting
B women in Singapore. Out of 20 Singaporean women, one
will be diagnosed with breast cancer in their lifetime, with the
highest incidence occurring in women aged 55–59 years. Every
year, about 1100 new cases of breast cancer are diagnosed and
about 270 die from the disease [1]. Early detection of breast can-
cer remains the best protection to patients in terms of success-
ful treatment and long-term survival. Mammography, an X-ray
imaging of compressed breast, is currently the most widely used
technique for breast cancer screening [2]. While mammography
offers clear advantages, it also carries some limitations and po-
tential risks. These include low sensitivity (ability to identify
presence of tumor) and specificity (ability to identify tumor as
Manuscript received May 2, 2007; revised November 7, 2007. Asterisk indi-
cates corresponding author.
*H. B. Lim is with the Advanced Electronics and Electromagnetics Group,
Institute of High Performance Computing, Agency for Science, Technology
and Research (A∗ STAR), Singapore 117528, Singapore (e-mail: limhb@ihpc.
a-star.edu.sg).
N. T. T. Nhung was with the Advanced Electronics and Electromagnet-
ics Group, Institute of High Performance Computing, Agency for Science,
Technology and Research (A∗ STAR), Singapore 117528, Singapore. She is
now with the Infowave Pte. Ltd., Singapore 575733, Singapore (e-mail:
nhung71167@yahoo.com).
E.-P. Li is with the Advanced Electronics and Electromagnetics Group,
Institute of High Performance Computing, Agency for Science, Technology
and Research (A∗ STAR), Singapore 117528, Singapore (e-mail: eplee@ihpc.
a-star.edu.sg).
N. D. Thang is with the School of Electrical and Electronic Engineer-
ing, Nanyang Technological University, Singapore 639798, Singapore (e-mail:
a0283655b@ntu.edu.sg).
Digital Object Identifier 10.1109/TBME.2008.919716
0018-9294/$25.00 © 2008 IEEE
Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on October 12,2023 at 13:35:57 UTC from IEEE Xplore. Restrictions apply.
1697
benign or malignant), uncomfortable and painful breast com-
pression, and exposure to low-dose ionizing radiation that poses
possible threat in increase cancer risk. Other available screening
techniques such as ultrasound and MRI are either less effective
or are too costly. The shortfalls of these techniques have moti-
vated the search for better alternatives [3].
During the past several decades, many microwave imaging
techniques, including passive, hybrid, and active approaches,
have been explored for breast cancer detection. The passive
approaches detect malignant tumors based on their increased
temperature compared to normal breast tissues [4], whereas the
hybrid approaches use microwave energy to illuminate the breast
and ultrasound transducers to detect pressure waves generated
by the expansion of heated tumors [5]. The active approaches
rely on the significant electrical properties contrast between ma-
lignant and normal breast tissues at microwave frequencies [6],
and two different types of technique have achieved promising
results: microwave tomography [7] and ultra-wideband (UWB)
microwave imaging [8]–[14]. In the microwave tomography
technique, the electrical properties profile of breast is recov-
ered from measurement by using near-field tomographic image
reconstruction algorithms. On the other hand, the UWB mi-
crowave imaging technique seeks to identify the presence and
location of significant backscatters such as malignant tumors.
The main challenge of UWB microwave imaging technique is
to devise an image reconstruction algorithm that provides high
resolution and good suppression of strong artifacts and noise.
Many different image reconstruction algorithms have been pro-
posed for UWB microwave imaging technique [8]–[14]. A con-
focal microwave imaging (CMI) technique that employs sim-
ple delay-and-sum (DAS) beamforming algorithm was first
proposed [8]–[10]. However, this algorithm does not account
for dispersive propagation effects and offers limited capabil-
ity for discriminating against artifacts and noise. An alternative
CMI technique, termed as microwave imaging via space–time
(MIST) beamforming [11], uses filters that compensate for dis-
persion and other limitations of the earlier DAS algorithm. Other
promising image reconstruction algorithms that have been con-
sidered are generalized likelihood ratio test (GLRT) [12], time
reversal (TR) [13], and multistatic adaptive microwave imaging
(MAMI) [14].
In this paper, we propose a new image reconstruction al-
gorithm, namely delay-multiply-and-sum (DMAS). In DMAS
algorithm, the backscattered signals at all antennas are first
time shifted as in the DAS algorithm, then an additional sig-
nal processing step that involves pairing multiplications is
 1698 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 55, NO. 6, JUNE 2008

Fig. 1. 2-D FDTD homogeneous breast phantom containing a 5 mm diameter

malignant tumor centered along the y-axis at a depth of 3 cm below a 1-mm-
thick skin layer. The seven black dots on the surface of the skin layer represent
antenna locations.

introduced before they are summed to create a synthetic focal

point. We will demonstrate the improved performance of DMAS
over DAS by simulating backscattered signals received from nu-
merical breast phantoms using the finite-difference time-domain
(FDTD) method [15].
The remainder of this paper is organized as follows. In
Section II, we describe the 2-D and 3-D breast phantoms and
system configurations simulated using the FDTD method. Sec-
tion III presents the DMAS algorithm for image reconstruction,
Fig. 2. 3-D FDTD breast phantom with a malignant tumor embedded in the
and Section IV provides numerical examples to illustrate im- otherwise homogeneous breast tissue interior, at a depth of 3 cm below a 2-mm-
proved malignant tumor detection capabilities of DMAS. Fi- thick skin layer. A five-elements antenna array, represented by five triangles, is
nally, concluding remarks are summarized in Section V. centered upon the breast phantom at 1 cm above the skin layer. (a) 3-D view.
(b) Side cross-sectional view.

II. FDTD BREAST MODELS FOR DATA ACQUISITION

order differentiated Gaussian pulse, with a maximum spectrum
Data for the UWB microwave imaging techniques are ac-
content near 7 GHz.
quired from numerical breast phantoms using the FDTD
A monostatic approach is employed for data acquisition. In
method. The natural spatial heterogeneity of normal breast tis-
this approach, the same antenna element is used to transmit the
sue is not accounted for in the breast phantoms. An antenna
UWB pulse and to receive the backscattered signal from the
array is placed either directly on or at a short distance from the
breast phantom. This process is repeated sequentially for each
surface of the breast phantoms. Each element in the antenna ar-
element in the antenna array, resulting in a total of M received
ray sequentially transmits a UWB pulse into the breast phantom
backscattered signals. The grid size used in the 2-D FDTD model
and the backscattered signals are recorded for illustration of the
is ∆x = ∆y = 0.4 mm and the time step is ∆t = 0.4 ps. The 2-
feasibility of our proposed DMAS algorithm in detecting small
D FDTD problem space is terminated with second-order Mur’s
malignant tumors.
absorbing boundary condition ( [15], Ch. 6).
A. 2-D Numerical Breast Phantom
B. 3-D Numerical Breast Phantom
The 2-D numerical breast phantom we used in our FDTD
The 3-D numerical breast phantom we used in our FDTD
model is shown in Fig. 1. The tissues are assigned frequency-
model is shown in Fig. 2. The tissues are either assigned nondis-
independent (nondispersive) dielectric parameters similar to
persive dielectric parameters using the same values as those in
those selected by Fear et al. [9]. The 2-D breast phantom in-
the 2-D breast phantom, or the dispersive nature of tissues is
cludes a 1-mm-thick skin layer (εr = 36.0, σ = 4.0 S/m) and
incorporated into the FDTD model using the time-domain aux-
a malignant tumor (εr = 50.0, σ = 7.0 S/m) of 5 mm in di-
iliary differential equation ( [15], Ch. 9) for a two-pole Debye
ameter embedded within normal breast tissue (εr = 9.0, σ =
model as
0.4 S/m) at a depth of 3 cm below the skin layer, centered along
the y-axis. The breast phantom is immersed in a liquid with the σ
2
εsp − ε∞
same dielectric properties as that of normal breast tissue. The εr − j = ε∞ + (1)
ωε0 p=1
1 + jωτp
black dots on the surface of the skin layer, as shown in Fig. 1,
represent the elements of an antenna array. The antenna array where εr is the relative permittivity, ε∞ is the permittivity in
consists of M = 7 elements equally spaced at 1 cm intervals, the terahertz frequency range, εs is the permittivity at static
with each element modeled as infinitely long line source. The frequency, ε0 is the free-space permittivity (=8.854 pF/m), σ
UWB pulse used as the antenna excitation is a 350 ps second- is the conductivity (in siemens per second), ω is the angular

Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on October 12,2023 at 13:35:57 UTC from IEEE Xplore. Restrictions apply.
BEEN LIM et al.: CMI FOR BREAST CANCER DETECTION: DELAY-MULTIPLY-AND-SUM IMAGE RECONSTRUCTION ALGORITHM
Fig. 3. FDTD modeled (a) relative permittivity and (b) conductivity as a
function of frequency from 0.1 to 15.5 GHz for malignant tumor (thick solid
curve) and normal breast tissue (thin solid curve).
frequency (in radians per second), τ is the relaxation time
√ (in seconds), p is the number of Debye poles, and
constant
j = −1. The Debye parameters for normal breast tissue (ε∞
= 2.68, εs1 = 5.01, εs2 = 3.85, τ1 = 15.84 ps, τ2 = 0.10 ns),
malignant tumor (ε∞ = 11.05, εs1 = 51.67, εs2 = 43.35, τ1 =
8.56 ps, τ2 = 0.23 ns), and skin layer (ε∞ = 4.62, εs1 = 37.10,
εs2 = 41.22, τ1 = 7.51 ps, τ2 = 0.31 ns) have been chosen to
approximate the Cole–Cole parametric dispersion model [16]
in the frequency range of interest from 3.1 to 10.6 GHz using a
two-step numerical fitting procedure [17]. Fig. 3 shows the rela-
tive permittivity and conductivity of breast tissues as a function
of frequency from 0.1 to 15.5 GHz. Generally, the two-pole De-
bye model deviates less than 10% from the Cole–Cole model at
frequencies above 1 GHz, and the deviation increases drastically
up to 35% for relative permittivity and 90% for conductivity at
frequencies below 1 GHz.
As shown in Fig. 2, the 3-D breast phantom is modeled as
a rectangular block consisting of homogeneous normal breast
tissue and a 2-mm-thick skin layer. The breast phantom is not
realistically shaped, but it is a reasonable approximation for
feasibility studies and comparisons of UWB microwave image
reconstruction algorithms. The breast phantom is modeled with
dimensions 10 cm (height) × 10 cm (width) × 5 cm (depth).
A spherical tumor of diameter 2 mm or 1 cm is embedded
within the breast tissue at a depth of 3 cm, and centered on
the yz plane or shifted by 2.5 cm from the center along the y-
axis. The entire breast phantom is immersed in a lossless liquid
with the same permittivity as that of normal breast tissue at
6.85 GHz (εr = 9.0, σ = 0.0 S/m for the nondispersive tissue
model, and ε∞ = εs1 = εs2 = 4.32, τ1 = τ2 = 0.00 s for the
dispersive tissue model).
An antenna array consisting of M = 5 elements is used,
with each element approximated as a point source. The antenna
elements are placed at 1 cm away from the breast skin and are
excited with a modulated Gaussian pulse of the form
  2 
(t − t0 )
V (t) = sin [2πf (t − t0 )] exp − (2)
τ free breast phantom is applied. The tumor response xij [n] is
where f = 6.85 GHz, τ = 80.2 ps, and t0 = 4τ . This pulse is
centered about 6.85 GHz and has a full-width at half-maximum
(FWHM) bandwidth of 6.6 GHz.
In our 3-D FDTD simulations, the data are acquired using a
multistatic approach. Each element in the antenna array takes
Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on October 12,2023 at 13:35:57 UTC from IEEE Xplore. Restrictions apply.
1699
turn to transmit and the backscattered signals are received at
all elements. A total of M × M backscattered signals will be
recorded after all the elements have taken their turn to trans-
mit. A grid size of ∆x = ∆y = ∆z = 1 mm is used to dis-
cretize the 3-D FDTD problem space and the time step used
is ∆t = 1.28 ps (∼780 GHz sampling frequency). Perfectly
matched layer (PML) absorbing boundary condition (geometric
progression profile with common ratio 2.15, eight layers) [18]
is placed at a minimum distance of ten grids from the antennas
or breast phantom. The computational time for each antenna
transmission, allowing enough time for the backscattered sig-
nals to be received at the antenna array, is approximately 2 h on
a Pentium4 3 GHz CPU.
III. IMAGE RECONSTRUCTION PROCEDURE
The recorded backscattered signals include early-time and
late-time contents. The early-time content is dominated by the
incident UWB pulse, reflections from the skin, and residual
antenna reverberations whereas the late-time content contains
tumor response and clutter. The goals of signal processing are
to reduce the early-time content, which is of much greater am-
plitude than the tumor response, to suppress the clutter in the
late-time content with minimum distortion level to the tumor
response, and to enhance the tumor response so that reliable
tumor detection in the reconstructed image can be achieved.
A. Artifact Removal
The backscattered signal transmitted by the ith antenna ele-
ment and received at the jth antenna element in discrete form is
denoted as bij [n]. In this paper, different artifact removal pro-
cedures are applied to the backscattered signals received from
the 2-D and 3-D breast phantoms. For the 2-D case, a reference
waveform R[n], which closely approximates the early-time con-
tent but with negligible late-time content is created by averaging
all the M recorded backscattered signals [8], [9]. The reference
waveform is then used for calibration by subtracting it from each
of the M recorded backscattered signals, resulting in M cali-
brated backscattered signals xii [n], which essentially contain
only the late-time content:
1

M
xii [n] = bii [n] − R [n] = bii [n] − bi  i  [n],
M 
i =1
where i = 1, 2, . . . , M. (3)
An example showing the successful removal of early-time
content from the recorded backscattered signal at the center
antenna using the reference waveform is given in Fig. 4.
For the 3-D case, a simple and straightforward artifact re-
moval procedure that requires a priori information of a tumor-
extracted by subtracting the M × M backscattered signals re-
ceived from the tumor-bearing breast phantom by their counter-
parts received from a tumor-free breast phantom
xij [n] = bij [n]tum or−b earing − bij [n]tum or−free . (4)
 1700
Fig. 4. (a) Backscattered signal recorded at the center antenna placed on the surface of the 2-D breast phantom shown in Fig. 1. (b) Enlargement of the late-time
content enclosed within the dotted box in (a). (c) Remaining signal after calibration using the reference waveform.
Fig. 5. Block diagram illustrating the DMAS procedure used in the image
reconstruction at location r0 in the breast phantom.
The artifact removal using a priori information is not a fea-
sible approach in real scenario. It is applied here as an easy
way to obtain the tumor response for performance analysis of
the DMAS algorithm. Removal of skin reflections and simi-
lar artifacts are beyond the scope of this paper and there are
publications that deal with this in a more realistic approach.
B. DMAS Algorithm
After artifact removal, the processed backscattered signals are
synthetically focused at each focal point in the breast by time
aligning the signals. The block diagram depicted in Fig. 5 shows
the DMAS procedure used after artifact removal to reconstruct
the intensity value of a focal point r0 in the breast phantom. First,
the round-trip path lengths from each transmitting antenna to a
focal point in the breast and back to the receiving antennas are
computed and converted into time delays Nij . All traversed
media in the round-trip are taken into account, and the signal
propagation velocity in each medium is calculated by assuming
Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on October 12,2023 at 13:35:57 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 55, NO. 6, JUNE 2008
the relative permittivity at the center frequency of the excitation
pulse (dispersive tissues: εr = 4.32 for normal breast tissue
and immersed liquid, εr = 34.22 for skin layer). The processed
backscattered signals are time shifted according to the computed
time delays and pairing multiplications are carried out before
the time-shifted signals are summed and squared. The intensity
value of the focal point I(r0 ) is then obtained by integrating the
contributions over a time window Wa
 Wa
I(r0 ) = S [n]dt (5)
0
where S[n] is the output signal after pairing multiplications,
summing and squaring, Wa = a∆t (in seconds), and a is an
integer representing the number of FDTD time steps. The pro-
cedure is repeated for each focal point in the breast, and finally,
the envelope of the intensities is displayed as an image.
The proposed DMAS algorithm is similar to the DAS algo-
rithm but with the additional pairing multiplication procedure
introduced after time shifting (the products of the pairing multi-
plication are presumed to be uncorrelated and further justifica-
tion on this will be addressed in future work). The improvement
in tumor detection capability of DMAS is based on a similar
principle as multistatic approach, in which additional clutter re-
jection is achieved by increased sample size [10]. For example,
considering a five-elements (M = 5) antenna array, the clutter
and noise rejection in DMAS is achieved by summation of 10
(M C2 ) and 300 (M ×M C2 ) signals for monostatic and multistatic
approaches, respectively, whereas in DAS, only 5 (M ) and 25
(M × M ) signals are summed.
C. Time Window Design
A time window can be used to preserve signal energy while
discriminating against unwanted clutter and noise. A good
choice of time window length ensures that the reconstructed
image intensity value is calculated using only samples of xij [n]
containing tumor backscattered energy, and hence, maximizes
the signal-to-clutter ratio (SCR). The backscattered signal col-
lected from frequency-dependent tissues is a distorted version of
the excitation pulse. The dispersive effect broadens the duration
of the backscattered signal and the broadening effect is directly
proportional to the tumor size [11]. These effects complicate the
selection of the time window.
BEEN LIM et al.: CMI FOR BREAST CANCER DETECTION: DELAY-MULTIPLY-AND-SUM IMAGE RECONSTRUCTION ALGORITHM 1701

Fig. 6. Modulated Gaussian pulse (thick solid line) used as the UWB excitation
signal in the 3-D FDTD breast phantom simulations.

Fig. 8. Comparison of xy cross-section images reconstructed. (a) Monos-

tatic DAS. (b) Monostatic DMAS. (c) Multistatic DAS. (d) Multistatic DMAS.
A 1 cm diameter malignant tumor is embedded at a depth of 3 cm inside a
nondispersive 3-D breast phantom shown in Fig. 2.

Fig. 7. Comparison of images reconstructed. (a) Monostatic DAS. (b) Mono-

static DMAS. A 5 mm diameter malignant tumor is centered at a depth of 3 cm
below a 1-mm-thick skin layer as shown in Fig. 1.

Fig. 6 shows the modulated Gaussian pulse used in the 3-

D FDTD breast phantom simulations. Since we are interested
in detecting early-stage breast cancer in which the tumor is
still very small in size, the length of the time window is se-
lected to be short. We have tested different time window lengths
(W100 , W200 , W300 , and W400 ) to examine the robustness of
our proposed DMAS algorithm against time window length.
Fig. 9. Comparison of xy cross-section images reconstructed. (a) Monos-
tatic DAS. (b) Monostatic DMAS. (c) Multistatic DAS. (d) Multistatic DMAS.
IV. NUMERICAL EXAMPLES A 1 cm diameter malignant tumor is embedded at a depth of 3 cm inside a
dispersive 3-D breast phantom shown in Fig. 2.
In this section, we demonstrate the effectiveness of the pro-
posed DMAS image reconstruction algorithm by applying it to
backscattered signals acquired from various FDTD simulated shows a larger extent of high-intensity region around the tumor
breast phantoms. First, we consider a simple 2-D nondispersive and this might obscure the existence of multiple tumors that
breast phantom. Then, we extend to 3-D and include the dis- are located near this region. Furthermore, the clutter at regions
persive nature of tissues into the FDTD model. For comparison out of the tumor is substantially reduced by using monostatic
purposes, the monostatic and multistatic DAS image reconstruc- DMAS.
tion algorithms [8]–[10] are applied to the same backscattered
signals. The effect of time window length on enhancement of B. 3-D Numerical Breast Phantom
clutter reduction and tumor identification is also investigated.
The time widow used in all the image reconstructions is W400
All the images shown are normalized to the maximum intensity
(i.e., 400∆t wide) unless specified. Figs. 8 and 9 display the
value of the reconstructed plane, where maximum is represented
reconstructed xy cross-section images for the 3-D breast phan-
by white and minimum by black.
tom shown in Fig. 2 for tissues assigned as nondispersive and
dispersive, respectively. Comparisons are made between mono-
A. 2-D Numerical Breast Phantom
static DAS, multistatic DAS, monostatic DMAS, and multistatic
Fig. 7 displays the reconstructed images for the 2-D breast DMAS. The embedded 1 cm diameter tumor is accurately lo-
phantom shown in Fig. 1 using monostatic DAS and DMAS. The calized in both nondispersive and dispersive tissues by all four
maximum intensity is located within the 5 mm diameter tumor at algorithms. However, the images reconstructed by using mono-
3 cm below the skin layer by both algorithms. Monostatic DAS static DAS [see Figs. 8(a) and 9(a)] are filled with strong level

Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on October 12,2023 at 13:35:57 UTC from IEEE Xplore. Restrictions apply.
 1702 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 55, NO. 6, JUNE 2008

Fig. 10. Images reconstructed using monostatic DMAS. (a) yz cross-section. Fig. 11. Images reconstructed using multistatic DMAS. (a) yz cross-section.
(b) zx cross-section. (c) xy cross-section. A 2 mm diameter malignant tumor is (b) zx cross-section. (c) xy cross-section. A 2 mm diameter malignant tumor is
embedded at a depth of 3 cm inside a dispersive 3-D breast phantom shown in embedded at a depth of 3 cm inside a dispersive 3-D breast phantom shown in
Fig. 2. Fig. 2.

of clutter that can be easily misinterpreted as tumors. Multi-

static DAS [see Figs. 8(c) and 9(c)] provides a much better
clutter reduction as compared to monostatic DAS, but the per-
formance is still poorer than monostatic DMAS [see Figs. 8(b)
and 9(b)]. Multistatic DMAS performs the best among all, and
is able to provide almost clutter-free images [see Figs. 8(d) and
9(d)]. Broadening effect (larger tumor size) is observed in Fig. 9,
which is an expected effect caused by dispersive tissues.
Figs. 10 and 11 display the images reconstructed on the yz,
zx, and xy cross-sections by using monostatic and multistatic
DMAS, respectively. The breast phantom used is the same as
that in Fig. 9, except that the tumor diameter is now reduced
to 2 mm. There is significantly much more clutter in the im-
ages reconstructed by using monostatic DMAS (see Fig. 10) as
compared to multistatic DMAS (see Fig. 11). But the tumor still
stands out from neighboring clutter in all the images shown in
Fig. 10. The images reconstructed by using multistatic DMAS
are relatively much cleaner and free from clutter. Comparing
Fig. 9(b) with Fig. 10(c) and Fig. 9(d) with Fig. 11(c), it is ob-
served that the area occupied by high-intensity values is very
similar for the 1 cm and 2 mm diameter tumors, except that the
clutter in the images containing the 2 mm diameter tumor is
slightly higher. Fig. 12. Comparison of xy cross-section images reconstructed using DMAS
Next, the enhancement of clutter reduction by using different with different time window lengths. (a) Monostatic W 3 0 0 . (b) Multistatic W 3 0 0 .
(c) Monostatic W 2 0 0 . (d) Multistatic W 2 0 0 . (e) Monostatic W 1 0 0 . (f) Multi-
time window lengths is investigated. Fig. 12 displays the im- static W 1 0 0 . A 2 mm diameter malignant tumor is embedded at a depth of 3 cm
ages reconstructed by using monostatic and multistatic DMAS inside a dispersive 3-D breast phantom shown in Fig. 2.
with W100 , W200 , and W300 . By using a shorter time window
length, the broadening effect is reduced, but without any no- used. Monostatic and multistatic DMAS with W100 and W400
ticeable changes in the extent of the maximum intensity (white) are then applied to backscattered signals received from a tu-
region. However, the clutter at regions out of the tumor can be mor that is shifted by 2.5 cm from the center along the y-
slightly higher or lower, depending on the time window length axis, with the embedded depth remaining at 3 cm. The image

Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on October 12,2023 at 13:35:57 UTC from IEEE Xplore. Restrictions apply.
BEEN LIM et al.: CMI FOR BREAST CANCER DETECTION: DELAY-MULTIPLY-AND-SUM IMAGE RECONSTRUCTION ALGORITHM
Fig. 13. Comparison of xy cross-section images reconstructed using DMAS
with different time window lengths. (a) Monostatic W 4 0 0 . (b) Multistatic W 4 0 0 .
(c) Monostatic W 1 0 0 . (d) Multistatic W 1 0 0 . A 2 mm diameter malignant tumor
is embedded at a depth of 3 cm and shifted by 2.5 cm from the center along the
y-axis inside a dispersive 3-D breast phantom shown in Fig. 2.
TABLE I
SCR AND SMR
reconstructed by using monostatic DMAS with W400 reveals
the presence of a tumor but with a very poor localization ca-
pability [see Fig. 13(a)]. Significant improvement in detection
and localization of the shifted 2 mm diameter tumor is achieved
by using a shorter time window length of W100 . The improved
performance by correct selection of time window length can
be observed by comparing Fig. 13(a) with (c) and Fig. 13(b)
with (d).
To quantify further the performance of DMAS, quantitative
assessments are carried out in terms of SCR and signal-to-mean
ratio (SMR). SCR compares the maximum tumor response with
the maximum clutter response in the same image whereas SMR
compares the maximum tumor response with the mean response
of the same image. The SCR and SMR as well as the location
of maximum response for Figs. 8–13 are shown in Table I.
Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on October 12,2023 at 13:35:57 UTC from IEEE Xplore. Restrictions apply.
1703
The SCR and SMR values for DMAS are higher than DAS for
all cases, which further confirm the improved performance of
DMAS.
V. CONCLUSION
The improvement in clutter rejection offered by our proposed
DMAS image reconstruction algorithm over DAS image re-
construction algorithm has been examined by considering 2-D
and 3-D breast phantoms. DMAS has shown its feasibility and
robustness in detecting tumor as small as 2 mm in diameter, em-
bedded at a depth of 3 cm inside breast phantoms assigned with
tissue natural dispersive properties. We have also shown that
multistatic DMAS is able to provide better imaging results than
its monostatic counterparts when the synthetic aperture formed
by the monostatic approach is similar to the real aperture array
used by the multistatic approach. Further enhancement of tumor
response and reduction of clutter provided by correct selection
of time window length have also been presented.
In a future paper, antenna array consisting of more elements
can be studied to explore the optimum number of antenna ele-
ments required for accurate and rapid tumor detection. Breast
phantoms containing multiple tumors can also be studied to
examine the sensitivity and specificity of DMAS. The effective-
ness and robustness of DMAS can further be demonstrated by
using numerical breast phantoms derived from anatomically re-
alistic MRI-derived FDTD breast models, where natural spatial
heterogeneity of breast tissues and realistic dielectric contrast
between normal and malignant tissues [19] are considered.
REFERENCES
[1] Breast Cancer Foundation. (2007). [Online]. Available: http://bcf.org.sg.
[2] M. Patlak, S. J. Nass, I. C. Henderson, and J. C. Lashof, Eds. Mammogra-
phy and Beyond: Developing Techniques for the Early Detection of Breast
Cancer. Washington, DC: Inst. of Med., National Academy Press, 2000.
[3] M. Säbel and H. Aichinger, “Recent developments in breast imaging,”
Phys. Med. Biol., vol. 41, no. 3, pp. 315–368, Mar. 1996.
[4] B. Bocquet, J. C. Van De Velde, A. Mamouni, Y. Leroy, G. Giaux, J. De-
lannoy, and D. Delvalee, “Microwave radiometric imaging at 3 GHz for
the exploration of breast tumors,” IEEE Trans. Microw. Theory Tech.,
vol. 38, no. 6, pp. 791–793, Jun. 1990.
[5] L. H. V. Wang, X. M. Zhao, H. T. Sun, and G. Ku, “Microwave induced
acoustic imaging of biological tissues,” Rev. Sci. Instrum., vol. 70, no. 9,
pp. 3744–3748, Sep. 1999.
[6] L. Sha, E. R. Ward, and B. Stroy, “A review of dielectric properties of
normal and malignant breast tissue,” in Proc. IEEE Southeast Conf., Apr.
5–7, 2002, pp. 457–462.
[7] P. M. Meaney, M. W. Fanning, D. Li, S. P. Poplack, and K. D. Paulsen,
“A clinical prototype for active microwave imaging of the breast,” IEEE
Trans. Microw. Theory Tech., vol. 48, no. 11, pp. 1841–1853, Nov. 2000.
[8] X. Li and S. C. Hagness, “A confocal microwave imaging algorithm
for breast cancer detection,” IEEE Microw. Wireless Components Lett.,
vol. 11, no. 3, pp. 130–132, Mar. 2001.
[9] E. C. Fear, X. Li, S. C. Hagness, and M. A. Stuchly, “Confocal microwave
imaging for breast cancer detection: Localization of tumors in three di-
mensions,” IEEE Trans. Biomed. Eng., vol. 49, no. 8, pp. 812–822, Aug.
2002.
[10] R. Nilavalan, A. Gbedemah, I. J. Craddock, X. Li, and S. C. Hagness, “Nu-
merical investigation of breast tumor detection using multi-static radar,”
Electron. Lett., vol. 39, no. 25, pp. 1787–1788, Dec. 2003.
[11] E. J. Bond, X. Li, S. C. Hagness, and B. D. Van Veen, “Microwave imaging
via space-time beamforming for early detection of breast cancer,” IEEE
Trans. Antennas Propag., vol. 51, no. 8, pp. 1690–1705, Aug. 2003.
[12] S. K. Davis, H. Tandradinata, S. C. Hagness, and B. D. Van Veen, “Ul-
trawideband microwave breast cancer detection: A detection-theoretic
 1704
approach using the generalized likelihood ratio test,” IEEE Trans. Biomed.
Eng., vol. 52, no. 7, pp. 1237–1250, Jul. 2005.
[13] P. Kosmas and C. M. Rappaport, “FDTD-based time reversal for mi-
crowave breast cancer detection—Localization in three dimensions,”
IEEE Trans. Microw. Theory Tech., vol. 54, no. 4, pp. 1921–1927, Apr.
2006.
[14] Y. Xie, B. Guo, L. Xu, J. Li, and P. Stoica, “Multistatic adaptive microwave
imaging for early breast cancer detection,” IEEE Trans. Biomed. Eng.,
vol. 53, no. 8, pp. 1647–1657, Aug. 2006.
[15] A. Taflove and S. C. Hagness, Computational Electrodynamics: The
Finite-Difference Time-Domain Method, 3rd ed. Boston, MA: Artech
House, 2005.
[16] S. Gabriel, R. W. Lau, and C. Gabriel, “The dielectric properties of biolog-
ical tissues: III. Parametric models for the dielectric spectrum of tissues,”
Phys. Med. Biol., vol. 41, no. 11, pp. 2271–2293, Nov. 1996.
[17] M. A. Eleiwa and A. Z. Elsherbeni, “Accurate FDTD simulation of biolog-
ical tissues for bio-electromagnetic applications,” in Proc. IEEE Southeast
Conf., 30 Mar.–1 Apr. 2001, pp. 174–178.
[18] J. P. Berenger, “Three-dimensional perfectly matched layer for the ab-
sorption of electromagnetic waves,” J. Comput. Phys., vol. 127, no. 2,
pp. 363–379, Sep. 1996.
[19] M. Lazebnik, L. McCartney, D. Popovic, C. B. Watkins, M. J. Lindstrom,
J. Harter, S. Sewall, A. Magliocco, J. H. Booske, M. Okoniewski, and
S. C. Hagness, “A large-scale study of the ultrawideband microwave di-
electric properties of normal breast tissue obtained from reduction surg-
eries,” Phys. Med. Biol., vol. 52, no. 10, pp. 2637–2656, May 2007.
Hooi Been Lim received the B.Eng. (with first class
honors) degree and the Ph.D. degree both in electronic
(communications) engineering from the University
of Sheffield, Sheffield, U.K., in 1999 and 2003,
respectively.
Since January 2004, she has been with the In-
stitute of High Performance Computing, Agency
for Science, Technology and Research (A∗ STAR),
Singapore, where she is currently a Senior Research
Engineer. Her current research interests include an-
tenna and microwave engineering, biomedical engi-
neering, and computational electromagnetics.
Nguyen Thi Tuyet Nhung received the B.Eng. de-
gree in electrical and electronic engineering from
Nanyang Technological University, Singapore, in
2006.
She was with the Institute of High Performance
Computing, Agency for Science, Technology and Re-
search (A∗ STAR), Singapore, as a Research Officer.
She is currently a Product Development Engineer in
Telematics, Infowave Pte. Ltd., Singapore. Her cur-
rent research interests include ultra-wideband imag-
ing, antenna design, and digital signal processing.
Authorized licensed use limited to: ULAKBIM UASL ISTANBUL TEKNIK UNIV. Downloaded on October 12,2023 at 13:35:57 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 55, NO. 6, JUNE 2008
Er-Ping Li (M’90–SM’01–F’08) received the M.Sc.
degree from Xi’an Jiaotong University, Xi’an, China,
in 1986, and the Ph.D. degree from Sheffield Hallam
University, Sheffield, U.K., in 1992, all in electrical
engineering.
From 1989 to 1992, he was a Research Asso-
ciate/Fellow in the School of Electronic and Informa-
tion Technology, Sheffield Hallam University. From
1993 to 1999, he was a Senior Research Fellow, a
Principal Research Engineer, and the Technical Di-
rector at the Singapore Research Institute and Indus-
try. Since 2000, he has been with the Institute of High Performance Computing,
Agency for Science, Technology and Research (A∗ STAR), Singapore, where he
is currently the Senior R&D Manager of Electronic Systems and Electromag-
netics. He is the author or coauthor of over 140 papers at referred international
journals and conferences, and five book chapters. He holds and filed number
of patents at U.S. patent office. His current research interests include fast and
efficient computational electromagnetics, bioelectromagnetics, and nanotech-
nology.
Dr. Li is the recipient of the 2006 IEEE Technical Achievement Award,
the Symposium Chair Award, and the 2007 Singapore Illuminating Engineer-
ing Society (IES) Prestigious Engineering Achievement Award. He is also the
recipient of the prestigious Changjiang (Yangtze) Chair Professorship Award
from the Ministry of Education, China, in 2007. He served as the Symposium
Chair for the 2006 Electromagnetic Compatibility (EMC)-Zurich in Singapore,
and the Chairman of the IEEE EMC Singapore Chapter for 2005–2006. He
is elected as an IEEE Distinguished Lecturer for 2007–2008, and is currently
an Associate Editor for the IEEE MICROWAVE AND WIRELESS COMPONENTS
LETTERS and the IEEE TRANSACTIONS ON EMC. He is elected as a Fellow of
Electromagnetics Academy.
Nguyen Duc Thang received the B.Eng. degree in
electrical and electronic engineering from Nanyang
Technological University, Singapore, in 2006, where
he is currently working toward the Ph.D. degree in
information engineering from the School of Electri-
cal and Electronic Engineering.
His current research interests include algorithms,
statistical learning, information retrieval, data min-
ing, and database.