IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 48, NO. 9, SEPTEMBER 2001 1053
Communications______________________________________________________________________
Computational Modeling of Three-Dimensional Microwave
Tomography of Breast Cancer
Alexander E. Bulyshev*, Serguei Y. Semenov, Alexandre E. Souvorov,
Robert H. Svenson, Alexei G. Nazarov, Yuri E. Sizov, and
George P. Tatsis
Abstract—Microwave tomographic approach is proposed to detect and
image breast cancers. Taking into account the big difference in dielectrical
properties between normal and malignant tissues, we have proposed using
the microwave tomographic method to image a human breast. Because of
the anatomical features of the objects, this case has to be referred to the
tomography with a limited angle of observation. As a result of computer
experiments we have established that multiview cylindrical configurations
are able to provide microwave tomograms of the breast with a small size
tumor inside. Using the gradient method, we have developed a computer
code to create images of the three-dimensional objects in dielectrical prop-
erties on microwave frequencies.
Index Terms—Dielectric model of breast, gradient method, inverse
problem, microwave tomography.
I. INTRODUCTION
Many sources [1]–[3], have demonstrated that malignant tissues are
different from normal tissues in dielectrical properties on microwave
frequencies. Based on these possibilities, the authors of [4] proposed a
method for detecting breast cancer. They have shown that even a small
spot with high values of complex dielectrical permittivities can affect a
back-scattered microwave signal. They also underlined significant ad-
vantages of the microwave diagnostic abilities over other existing tech-
niques. In [5] microwave imaging of another kind of cancer has been
presented. In this paper, we have applied a tomographic approach to
create microwave images which allows us to obtain three-dimensional
(3-D) images of a woman’s breast with high quality.
The 3-D microwave tomography of the breast has significant differ-
ence in comparison with standard approaches: in this case, we cannot
provide all around illumination and observation of the object. More-
over, the object (breast) borders the high-contrast region (chest), and
this circumstance could be a serious obstacle to obtaining high-quality
microwave images of the breast. However, our computational experi-
ments show that we can produce images even under these conditions
using cylindrical geometry of the tomograph described in [6].
The reconstruction algorithm is also an important issue in any at-
tempts to construct a tomographic device. We have figured out that in
our case the gradient method [7] could be an acceptable choice. We de-
scribed our modifications of this method previously [6], and we have
applied it to solve breast cancer imaging problems.
Manuscript receivedJanuary 13, 2000; revised June 2, 2001. Asterisk indi-
cates corresponding author.
*A. E. Bulyshev is with the Laser and Applied Technologies Laboratory,
Carolinas Medical Center, Charlotte, NC 28203 USA (e-mail: ssemenov@car-
olinas.org).
S. Y. Semenov, A. E. Souvorov, R. H. Svenson and G. P. Tatsis are with the
Laser and Applied Technologies Laboratory, Carolinas Medical Center, Char-
lotte, NC 28203 USA.
A. G. Nazarov and Y. E. Sizov are with the Kurchatov Institute of Atomic
Energy, Moscow, Russia.
Publisher Item Identifier S 0018-9294(01)07449-3.
Fig. 1. A vertical slice of the device and the model. 1: Working chamber;
2: a set of transmitters; 3: a set of receivers; 4: a semisphere; 5: a small
inhomogeneous area; 6: a layer of fat; 7: a layer of muscles; 8: a layer of bone.
Using the experimental data about dielectrical properties of living
tissues, we have developed a dielectrical model of the breast. We have
taken into account dielectrical properties of breast tissues, fat, muscles,
bone, and malignant tissues.
In Section II, we describe geometry of the device and models which
are proposed. In Section III, we formulate microwave tomographical
problems from a mathematical point of view. Section IV contains the
results of our computational experiments, which is followed by our
conclusions.
II. OBSERVATION SCHEME AND MODEL OF THE OBJECT
The cylindrical multiview illumination scheme has been recently ap-
plied to solve full-body microwave tomographic problems [6]. Trying
to use this scheme to image the breast, we face two significant differ-
ences. First, the angle of observation is strictly limited in vertical (z)
direction (see Fig. 1). Second, the object (breast) borders high-contrast
region (chest). In the present investigation, we are not interested in the
chest imaging. The aim is the breast imaging, but existence of a high
contrast area could distort images of the object. However, the results of
our computational experiments show that even under these conditions
this method can provide high-quality breast images and can visualize
small size cancer zones.
In Fig. 1, one can see that the tomographical chamber is a cylinder
with the transmitter and receiver antennas attached to the cylindrical
surface. In our computational experiments, we have used one, three,
or five rows of transmitters with 32 transmitters in each row and 32
rows of receivers with 32 receiver in each row. Transmitters work one
after another, and receivers work all together simultaneously. This is
the main reason why the numbers of transmitters and receivers are so
different. Both computational time and real working time of the tomo-
graph are proportional to the number of transmitters and do not depend
on the number of receivers.
Using the experimental data of tissues dielectrical properties, we
developed the dielectrical model of a woman’s breast. This model
consists of semi-sphere with radius R
c
and three cylindrical layers
on the top of this semi-sphere. The first layer represents a fat layer,
the second—muscles, the third—bones. We were modeling tumors
as spheres with a small radius placed inside the semi-sphere. The
0018–9294/01$10.00 © 2001 IEEE
Administrator
高亮
Administrator
高亮
Administrator
高亮
快速收敛方法,用于成像的算法
Administrator
线条
Administrator
高亮
Administrator
高亮
Administrator
高亮
Administrator
高亮
Administrator
高亮
Administrator
高亮
Administrator
高亮
Administrator
高亮
Administrator
高亮
Administrator
线条
Administrator
线条
Administrator
高亮
Administrator
高亮
Administrator
线条
1054 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 48, NO. 9, SEPTEMBER 2001
schematic of this model is drawn in Fig. 1. In our computational exper-
iments, we accept the following values of the dielectrical properties.
Fat: "
f
= 10 + i; bone: "
b
= 11 + 2i; muscles: "
m
= 25 + 15i;
breast tissue: "
bt
= 10 + 1:5i; malignant tissue: "
c
= 23 + 15i.
III. MATHEMATICAL MODEL
The mathematical model of the microwave tomographic system
was described in our previous paper [6]. This model is based on
the Helmholtz equation for solving the direct problems and on the
gradient method for inverse problem solving. As we have showed,
the multiview illumination scheme has to be used for good quality
imaging. The significant difference between this case and the situation
discussed in [6] is that we can provide only limited angle for illumi-
nation in z-direction because of anatomical features of the object (see
Fig. 1).
A. Direct Problem
As we mentioned, we use the scalar Helmholtz equation to describe
electromagnetic waves propagation in free space and in the object
r
2
E
s
+ k
2
E
s
= (k
2
0
� k
2
)E
0
(1)
where
E
0
and E
s
z component of the incident and scattered electric field;
k wave number;
k
2
= (2�=�)
2
";
� wavelength of the radiation in vacuum.
As a rule, the scalar approximation has been used in the works devoted
to two-dimensional (2-D) microwave tomographic problems. We used
this approximation as the first reasonable step to solve the 3-D vector
problem. In our computational experiments, we used the point source
field as the incident field
E
0
(r) =
exp(ik
0
r)
4�r
(2)
where r is the distance between the current point and the phase center
of the transmitter antenna, k
0
is the wave number for homogeneous
space.
In order to solve (1), we need to formulate boundary conditions for
E
s
. As a computational region we used a cylinder placed inside of
the tomographical chamber. We used approximated absorbed boundary
conditions [8] on the surface of the cylinder. One can find more details
about how to solve this problem in [6].
B. Inverse Problem
Being a nonlinear problem, the inverse problem for the Helmholtz
equation requires iterative methods to solve it. Three different ap-
proaches [7], [9], [10] have been applied to such kinds of problems.
The first method [9], [11], [12] is an iterative scheme based on the
Born–Rytov approximation and has a limited area of applications.
This method is very effective to image weak scattering objects, but
is unable to image strong scattering objects. The best results were
achieved applying Newton iterative schemes [10], [13]–[15] to solve
2-D problems. The key points of the Newton method are calculation
and inversion of a matrix of a large dimension, and in the 3-D case
these procedures become extremely time consuming.
We successfully employed the gradient method [7] to solve 3-D mi-
crowave tomographic problems [6]. This method was also applied in
the present work. A gradient method operates with functional J ["]
J ["] =
i; j
kU
teor
i; j
� U
expr
i; j
j
2 (3)
where U teor
i; j
is the theoretical prediction of the scattered field mea-
sured by receiver number j when transmitter number i was employed,
and U expr
i; j
is the experimentally measured value of the scattered field.
An inverse problem can be reformulated as a minimization problem
for the functional J ["]. It is very important to know the gradient of the
functional J ["] to solve a minimization problem. In this case, the so-
lution is given in [7], [16]. It could be shown that gradient J 0 can be
represented as
J
0
=
i
U
�
i
V
i
(4)
where U
i
is the electrical field produced by transmitter with number i.
Function V
i
can be found from
r
2
V
i
+k
2
V
i
= (k
2
�k
2
0
)
j
(U
expr
i; j
�U
teor
i; j
)G
0
(k
0
jr
j
�rj)) (5)
where
G
0
Green’s function for the uniform space;
r current point;
r
j
receiver location.
In the 3-D case, the Green function can be represented as
G
0
(k
0
r) =
exp(k
0
r)
4�r
: (6)
Knowing gradient of the functional J , we can construct the minimizing
procedure
"
n+1
= "
n
� J
0
["
n
]s
n (7)
where step sn was chosen according to a simple algorithm [6].
C. Computational Experiments
We used computational experiments as investigational tools. The
standard scheme of the closed circle was employed [17]. First, we
choose the model of the object. Then, we calculate “experimental” data.
Then, we restore the image of the object. Finally, we are able to com-
pare the results of restoration with the original. The main goal of the
experiments was to establish the fact that a small but high contrast in-
homogeneous feature could be made visible under these experimental
conditions. In our computational experiments, we can vary geometric
characteristics of the system, numbers of transmitters and receivers,
working frequency, and parameters of computational schemes.
IV. RESULTS AND DISCUSSION
In our computational experiments, we used the model with the fol-
lowing parameters: a radius of the semi-sphere R
c
= 4.5 cm, heights
of the layers H
1
= H
2
= H
3
= 1 cm, a radius of the small sphere
R
t
= 0.3 cm, dielectrical properties "
bt
= 10 + 1:5i, "
b
= 11 + 2i,
"
m
= 25+ 15i, "
f
= 10+ i, dielectrical properties of the immersion
liquid were chosen "
0
= 10 + i. The radius of the tomograph was 10
cm and the height of the tomograph was 20 cm. In our computational
experiments, we used three rows of transmitters with 16 transmitters in
each, and we used 32 receivers on one vertical line (see Fig. 1).
In our computational experiments, we used the dual mesh approach
[18]. The direct problems were solved applying the cylindrical mesh
with the numbers of cells 64, 64, and 127 in the radial, azimuthal, and
z directions, consequently. The inverse problems were solved on the
cubic mesh with 64 cells in all directions. In our experiments, we used a
DEC ALPHA 8200 server; total computational time to solve the inverse
problem is about 30 min.
Three different frequencies have been used in the computational ex-
periments: 2, 3.5, and 5 GHz. The results of restoration on frequency
Administrator
线条
Administrator
线条
Administrator
椭圆
Administrator
高亮
Administrator
线条
Administrator
线条
Administrator
线条
Administrator
线条
Administrator
高亮
Administrator
矩形
Administrator
高亮
Administrator
线条
Administrator
高亮
Administrator
矩形
Administrator
线条
Administrator
矩形
Administrator
线条
Administrator
线条
Administrator
矩形
Administrator
矩形
Administrator
线条
Administrator
线条
Administrator
线条
Administrator
矩形
Administrator
矩形
Administrator
线条
Administrator
线条
Administrator
线条
Administrator
线条
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 48, NO. 9, SEPTEMBER 2001 1055
(a)
(b)
Fig. 2. The vertical slice of the image of the model of the breast. Frequency= 3.5 GHz. (a) Real part. (b) Imaginary part. The circle shows the place of the tumor.
3.5 GHz are represented in Fig. 2. Analyzing both the real and imagi-
nary parts of the complex dielectrical permittivity one can see the shape
of the object, location of the muscles, and location and dielectrical
properties of the “cancer zone.” We have to conclude that this frequency
provides better quality of the image, sensitivity, and space resolution.
The results of the computer modeling at 2 and 5 GHz show poor
quality of images. Using 2 GHz, cannot provide sufficient space res-
olution because of the large wavelength. In the 5-GHz case, images
are much more distorted. This circumstance could be connected with
the significant absorption of microwaves in biological tissues at high
frequencies. Even though our results are preliminary, we can conclude
that the 3-GHz frequency is close to the optimal frequency to image a
woman’s breast.
However, the important questions about the optimal frequency, space
resolution, sensitivity are subject to future investigations. They have to
be solved taking into account detailed schematic of the tomographic
system, and they would be solved completely in combined theoretical
and experimental investigations.
V. CONCLUSION
It has been shown in computational experiments that 3-D microwave
images of the woman’s breast can be obtained on the tomographical
system in low GHz region with quality sufficient for the small size
tumor area detection. The cylindrical working chamber with transmitter
and receiver antennas attached to the cylindrical surface was consid-
ered. The modified gradient method was shown to be capable to solve
3-D inverse problems for Helmholtz equation and provide 3-D images
of the breast in complex dielectrical permittivity. It has been shown that
1056 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 48, NO. 9, SEPTEMBER 2001
the tomographic system are able to detect small high-contrast inhomo-
geneous features.
REFERENCES
[1] C. Gabriel, S. Gabriel, and E. Corthout, “The dielectrical properties of
biological tissues—Part I: Literature survey,” Phys. Med. Biol., vol. 41,
p. 2231, 1996.
[2] J. L. Schepps and K. R. Foster, “The UHF and microwave dielectrical
properties of normal and tumor tissues: Variation in dielectrical prop-
erties with tissues water content,” Phys. Med. Biol., vol. 25, no. 6, pp.
1149–1159, 1980.
[3] W. T. Joines, R. L. Jirtle, M. D. Rafal, and D. J. Schaefer, “Microwave
power absorption differences between normal and malignant tissue,” Int.
J. Radiat. Oncol. Biol. Phys., vol. 6, pp. 681–687, 1980.
[4] S. C. Hagness, A. Taflove, and J. E. Bridges, “Two-dimensional FDTD
analysis of a pulsed microwave confocal system for breast cancer de-
tection: Fixed-focus and antenna-array sensors,” IEEE Trans. Biomed.
Eng., vol. 45, pp. 1470–1479, Dec. 1998.
[5] S. M. Panas, L. T. Reconas, and T. D. Tsiboukis, “Microwave imaging
using the finite element method and sensitivity analysis approach,” IEEE
Trans. Med. Imag., vol. 18, pp. 1108–1114, Nov. 1999.
[6] A. E. Bulyshev, S. Y. Semenov, A. E. Souvorov, R. H. Svenson, A. G.
Nazarov, Y. E. Sizov, and G. P. Tatsis, “Three-dimensional microwave
tomography. Theory and computer experiments in scalar approxima-
tion,” Inverse Prob., vol. 16, pp. 863–875, 2000.
[7] H. Harada, D. Wall, T. Takenaka, and M. Tanaka, “Conjugate gradient
method applied to inverse scattering problem,” IEEE Trans. Antennas
Propagat., vol. 43, pp. 784–792, Aug. 1995.
[8] G. Mur, “Absorbing boundary conditions for the finite-difference ap-
proximation of the time-domain electromagnetic-field equations,” IEEE
Trans. Electromag. Compat., vol. EMC–23, pp. 377–382, 1981.
[9] A. E. Souvorov, A. E. Bulyshev, S. Y. Semenov, R. H. Svenson, A. G.
Nazarov, Y. E. Sizov, and G. P. Tatsis, “Microwave tomography: A two-
dimensional Newton iterative scheme,” IEEE Trans. Microwave Theory
Tech., vol. 46, pp. 1654–1659, Nov. 1998.
[10] N. Joachimowicz, C. Pichot, and J. Hugonin, “Inverse scattering: An
iterative numerical method for electromagnetic imaging,” IEEE Trans.
Antennas Propagat., vol. 39, pp. 1742–1752, Dec. 1991.
[11] A. Burov, A. Gorunov, A. Sascovez, and T. Tihonova, “Inverse scat-
tering problems in acoustic” (in Russian), Acoust. J., vol. 32, no. 4, pp.
433–449, 1986.
[12] W. C. Chew and Y. M. Wang, “An iterative solution of two-dimensional
electromagnetic inverse scattering problem,” Int. J. Imag. Syst. Technol.,
vol. 1, no. 1, pp. 100–108, 1989.
[13] A. E. Souvorov, A. E. Bulyshev, S. Y. Semenov, R. H. Svenson, and
G. P. Tatsis, “Two-dimensional computer analysis of a microwave flat
antenna array for breast cancer tomography,” IEEE Trans. Microwave
Theory Tech., vol. 48, pp. 1413–1415, Aug 2000.
[14] W. C. Chew and Y. M. Wang, “Reconstruction of two-dimensional per-
mittivity distribution using the distorted Born iterative method,” IEEE
Trans. Med. Imag., vol. 9, pp. 218–225, Apr. 1990.
[15] P. M. Meaney, K. D. Paulsen, A. Hartov, and R. K. Crane, “Microwave
imaging for tissue assessment: Initial evaluation in multitarget tissue-
equivalent phantoms,” IEEE Trans. Biomed. Eng., vol. 43, pp. 878–890,
Sept. 1996.
[16] R. Kleinman and P. van der Berg, “A modified gradient method for two-
dimensional problems in tomography,” J. Comput. Appl. Math., vol. 42,
p. 17, 1992.
[17] A. N. Tikhonov and V. Y. Arsenin, Solutions of Ill-Posed Prob-
lems. Washington, D.C.: Winston, 1977.
[18] K. D. Paulsen, P. M. Meaney, M. J. Moskowitz, and J. Sullivan, “A dual
mesh scheme for finite element based reconstruction algorithms,” IEEE
Trans. Med. Imag., vol. 14, pp. 504–514, June 1995.
Variation in the Dominant Period During Ventricular
Fibrillation
Michael Small*, Dejin Yu, and Robert G. Harrison
Abstract—Time-varying periodicities are commonly observed in biolog-
ical time series. In this paper, we discuss three different algorithms to detect
and quantify change in periodicity. Each technique uses a sliding window
to estimate periodic components in short subseries of a longer recording.
The three techniques we utilize are based on: 1) standard Fourier spectral
estimation; 2) an information theoretic adaption of linear (autoregressive)
modeling; and 3) geometric properties of the embedded time series. We
compare the results obtained from each of these methods using artificial
data and experimental data from swine ventricular fibrillation (VF).
Spectral estimates have previously been applied to VF time series to show
a time-dependent trend in the dominant frequency. We confirm this result
by showing that the dominant period o