Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Med Biol Eng Comput (2011) 49:233–240

DOI 10.1007/s11517-011-0733-3

ORIGINAL ARTICLE

Numerical modeling of magnetic induction tomography using


the impedance method
Airton Ramos • Julia G. B. Wolff

Received: 15 September 2010 / Accepted: 2 January 2011 / Published online: 13 January 2011
Ó International Federation for Medical and Biological Engineering 2011

Abstract This article discusses the impedance method in biomedical applications have been proposed for detection
the forward calculation in magnetic induction tomography of brain edema or hemorrhage [6, 17–19] and early diag-
(MIT). Magnetic field and eddy current distributions were nosis of some types of tumors [20]. Due to the low con-
obtained numerically for a sphere in the field of a coil and ductivity of biological tissues, biomedical applications of
were compared with an analytical model. Additionally, MIT require relatively high frequencies, typically in the
numerical and experimental results for phase sensitivity in range from 100 kHz to 10 MHz.
MIT were obtained and compared for a cylindrical object MIT is based on measuring the perturbation produced by
in a planar array of sensors. The results showed that the an object in an applied magnetic field. The measurement of
impedance method provides results that agree very well the disturbance field using different sources and at various
with reality in the frequency range from 100 kHz to positions around the object gives enough information to
20 MHz and for low conductivity objects (10 S/m or less). construct images of internal electromagnetic properties of
This opens the possibility of using this numerical approach the object. MIT offers some advantages over the traditional
in image reconstruction in MIT. low frequency and low resolution technique named elec-
trical impedance tomography (EIT), specifically, there is
Keywords Eddy currents  Tomography  Impedance no need for physical contact with the object, the technique
method  Magnetic field  Numerical modeling is sensitive to the three electromagnetic properties (per-
meability l, permittivity e, and conductivity r) of the
object, and the positions of the coils are fixed and inde-
1 Introduction pendent of the object, a great advantage for obtaining
measurements and image reconstruction [17].
Magnetic induction tomography is a non-invasive tech- Several methods can be used to model the distribution of
nique for visualization of passive electromagnetic proper- eddy current and magnetic field numerically in a MIT
ties inside an object. This is a new technique still under system. Most previous studies in MIT have used the finite
development and presents significant potential for indus- element method (FEM) [2, 6, 18, 19], but modeling by
trial and biomedical applications. It can, for example, be finite difference approaches (FDM) has also been used [7,
used in non-destructive testing of conductive or magnetic 8]. This article presents an alternative approach using the
material, in geophysical inspections, in visualization and impedance method (IM) [9–11, 13] in which the object is
control in pipelines, and verification of metal contaminants modeled as a three-dimensional network of lumped
in food, etc. Relevant information on MIT applications are impedances where fields and currents are calculated based
presented by Griffiths [3] and references therein. Some on the Biot and Savart Law, Faraday’s Law, and Kirch-
hoff’s Voltage Law. Finite difference and Impedance
Method are conceptually simpler than the Finite Element
A. Ramos (&)  J. G. B. Wolff
Method. FEM is a more powerful and versatile numerical
Department of Electrical Engineering, University of Santa
Catarina State, Joinville, Santa Catarina, Brazil technique for handling problems involving complex
e-mail: airton_ramos@joinville.udesc.br geometries and inhomogeneous media, mainly due to the

123
234 Med Biol Eng Comput (2011) 49:233–240

discretization process. FDM and IM use difference equa- (A)


tions to model the electromagnetic equations in discrete
space, so the discretization is made with volume elements
in the form of regular hexahedrons. FEM, in turn, is based object
coil doc 2Rc
on the minimization of functionals in a space described by
non-regular discretization elements, which gives it much Ro
more precise description of complex geometries. Despite
the advantages of FEM, the impedance method is an
interesting alternative because it allows developing algo-
rithms very easily in view of their formulation based on (B) ΔLn
Is
equations of electric circuit and gives directly the induced
r - sn
currents on the object without the need for additional post
ΔA
processing calculation.
The aim of this work is to show that this method is sn
suitable for the analysis in MIT with low conductivity x r
objects providing reliable estimates of primary field, eddy
currents, secondary field, and sensitivity values for image
y z
reconstruction. To reach this conclusion, numerical results
for eddy currents and magnetic field were compared with Fig. 1 a Schematic representation of a planar MIT system. b Vectors
the prediction from an analytical model applied to a used in calculating the applied magnetic potential and magnetic field.
spherical object. Furthermore, numerical and experimental The dimensions: object radius (Ro), coil radius (Rc) and object–coil
results for phase sensitivity for a cylindrical object in a distance (doc) are specified in Table 1
planar array of sensors were also compared.
the position vector of DL. The applied magnetic potential
can be calculated by the same method.

2 Method l o Is X
Nc
DLn
AðrÞ ¼ ð2Þ
4p n¼1 jr  sn j
MIT systems consist of several sources and several sensors
The impedance method is then used to obtain eddy
of magnetic field around the object. Coils are usually used
currents. The object is discretized into a three-dimensional
both to generate and measure the field. Figure 1 shows a
regular network of voxels. The voxels are connected to
schematic representation of a planar system. The technique
their neighbors by lumped impedances depending on voxel
involves injecting an alternating current in a coil to gen-
size and electric properties of the media. The resulting
erate a primary magnetic field and measuring the secondary
electric circuit is shown in Fig. 2. In the discrete space, we
field produced by eddy currents in the object using the
can write equations for the induced voltage in each mesh
other coils. The set of all measurements is equal to the
number of sources times the number of sensors and is used
to reconstruct the image of the conductivity distribution in
z Zy (i,j,k+1)
the object using a discretization scheme and regularization
techniques since the inverse problem is always ill-posed
Zz (i,j,k)
[15]. Zx (i,j,k+1)
If the generated field wavelength is much larger than the Ix (i,j,k) Zz (i,j+1,k)
highest distance source-sensor or source-object, a good
estimate for this field can be obtained from the Biot–Savart Zy (i,j,k) y
law. Using a discretization with Nc divisions in the coil and Iy (i,j,k)
Zx (i,j,k)
neglecting the wire cross section area, the primary field Bp
can be calculated according to Eq. 1: Zz (i+1,j,k) Iz (i,j,k) Zx (i,j+1,k)
l Is X
Nc
DLn  ðr  sn Þ
Bp ðrÞ ¼ o ð1Þ x
4p n¼1 j r  sn j 3 Zy (i+1,j,k)
where lo is the magnetic permeability of vacuum, Is is the
Fig. 2 Unit cell of the equivalent circuit after discretization of the
current in the coil, DL is the displacement along the coil, media. Adjacent voxels are connected by impedances. Eddy currents
r is the position vector where the field is calculated and s is are shown as mesh currents in this circuit

123
Med Biol Eng Comput (2011) 49:233–240 235

by the primary field using Eqs. 3–5 based on Faraday’s Zy ði;j;kÞ½Ix ði;j;kÞ  Ix ði;j;k  1Þþ Iz ði  1;j;kÞ  Iz ði;j;kÞ
Law:  
 þ Zz ði;j;kÞ Ix ði;j;kÞ  Ix ði;j  1;kÞþ Iy ði  1;j;kÞ Iy ði;j;kÞ

Vx ði; j; kÞ ¼ ix h Ay ði; j; k þ 1Þ þ Az ði; j; kÞ þ Zy ði;j;k þ 1Þ Ix ði;j;kÞ Ix ði;j;k þ 1Þ þ Iz ði;j;k þ 1Þ
 
Ay ði; j; kÞ  Az ði; j þ 1; kÞ ð3Þ Iz ði  1;j;k þ 1Þ
 
Vy ði; j; kÞ ¼ ix h Ax ði; j; kÞ þ Az ði þ 1; j; kÞ þ Zz ði;j þ 1;kÞ Ix ði;j;kÞ  Ix ði;j þ 1;kÞþ Iy ði;j þ 1;kÞ
Ax ði; j; k þ 1Þ  Az ði; j; kÞ ð4Þ Iy ði  1;j þ 1;kÞ ¼ Vx ði;j;kÞ ð9Þ
  
Vz ði; j; kÞ ¼ ix h Ay ði; j; kÞ þ Ax ði; j þ 1; kÞ Zx ði;j;kÞ Iy ði;j;kÞ  Iy ði;j;k  1Þ þ Iz ði;j  1;kÞ Iz ði;j;kÞ
 
Ay ði þ 1; j; kÞ  Ax ði; j; kÞ ð5Þ þ Zz ði;j;kÞ Iy ði;j;kÞ  Iy ði  1;j;kÞþ Ix ði;j  1;kÞ Ix ði;j;kÞ

þ Zx ði;j;k þ 1Þ Iy ði;j;kÞ Iy ði;j;k þ 1Þ þ Iz ði;j;k þ 1Þ
where h is the edge of the voxel used in the discretization 
(grid parameter) and Ax, Ay, and Az are the magnetic Iz ði;j  1;k þ 1Þ

potential components calculated by Eq. 2 in the x, y, and þ Zz ði þ 1;j;kÞ Iy ði;j;kÞ  Iy ði þ 1;j;kÞþ Ix ði þ 1;j;kÞ
z direction, respectively. That is shown in Fig. 3. Ix ði þ 1;j  1;kÞ ¼ Vy ði;j;kÞ ð10Þ
Eddy currents can be calculated from traditional mesh  
Zx ði;j;kÞ Iz ði;j;kÞ Iz ði;j  1;kÞþ Iy ði;j;kÞ  Iy ði;j;k  1Þ
analysis. Each branch of the equivalent circuit is shared by
four meshes in two perpendicular planes. Hence, there are þ Zy ði;j;kÞ½Iz ði;j;kÞ Iz ði  1;j;kÞ þ Ix ði;j;k  1Þ  Ix ði;j;kÞ

four mesh currents that must be algebraically added for þ Zx ði;j þ 1;kÞ Iz ði;j;kÞ  Iz ði;j þ 1;kÞþ Iy ði;j þ 1;k  1Þ

each branch. For an easy calculation of the impedances, it Iy ði;j þ 1;kÞ
is considered that the electric field is uniform and the 
þ Zy ði þ 1;j;kÞ Iz ði;j;kÞ  Iz ði þ 1;j;kÞþ Ix ði þ 1;j;kÞ
medium is homogeneous inside the voxel. Thus, the
Ix ði þ 1;j;k  1Þ ¼ Vz ði;j;kÞ ð11Þ
impedances are given by Eqs. 6–8:
Zx ði; j; kÞ ¼ f½rx ði; j; kÞ þ jxeo ex ði; j; kÞhg1 ð6Þ Therefore, if the discretization grid consists of N voxels,
  1 the number of equation to solve is 3 N. To specify the
Zy ði; j; kÞ ¼ ry ði; j; kÞ þ jxeo ey ði; j; kÞ h ð7Þ boundary conditions, all current terms outside the grid are
made equal to zero. After obtaining the distribution of
Zz ði; j; kÞ ¼ f½rz ði; j; kÞ þ jxeo ez ði; j; kÞhg1 ð8Þ
mesh currents, the secondary field is calculated through the
where r is the conductivity, e is the dielectric constant, and Biot–Savart law. Equation 12 for the secondary field uses
eo is the vacuum permittivity. Since the properties of the the branch currents in each voxel (Ibx, Iby, Ibz) which are
medium are described by functions of position (i, j, k) and calculated from the mesh currents. That is shown in Fig. 3.
direction (x, y, z) in the discretization grid, this model  
lo h X
N
Ibx ux þ Iby uy þ Ibz uz n ðr  rn Þ
allows the representation of anisotropic and heterogeneous Bs ðrÞ ¼ ð12Þ
media. Application of Kirchhoff’s Voltage Law to the
4p n¼1 jr  rn j3
meshes involving the voxel (i, j, k) results in Eqs. 9, 10, The symbols ux, uy, and uz refer to the unit vectors in
and 11 for the mesh currents: directions x, y, and z, respectively, and rn is the position
vector of the voxel.
A program for current and field calculation using the
methods described previously was built in the C?? lan-
Ay(i,j,k+1) guage. Table 1 shows the values used in the simulation for
Ax(i,j,k+1) a spherical object with conductivity and dielectric constant
Ax(i,j+1,k+1)
Az(i,j,k) Az(i,j+1,k) typical of biological material [1]. The coil consists of only
one turn. We used the frequencies 100 kHz, 1 MHz, and
10 MHz. The dimensions of the object and the coil are
Ibz(i,j,k) Az(i+1,j+1,k)
Az(i+1,j,k) realistic when compared to prototypes of tomographs
Iby(i,j,k) Ay(i,j,k) reported in the literature [14, 16]. Solutions to the mesh
Ax(i,j,k) equations were obtained by the successive-over-relaxation
Ibx(i,j,k) Ax(i,j+1,k) Gauss–Seidel method using the acceleration factor of 1.84.
Simulations were performed on a Pentium 4 computer with
Ay(i+1,j,k)
2 GB of RAM and Windows XPTM operating system.
Fig. 3 Branch currents in the voxel (i, j, k) and magnetic potential in Convergence for each simulation in this large system with
each branch of the unit cell 331,776 equations required 25 s with stopping criterion of

123
236 Med Biol Eng Comput (2011) 49:233–240

Table 1 Simulation parameters the Hz-16 (Rohde & Schwarz) amplifier. The amplified
Parameter Spherical object Cilindrical object
signal was read on a DPO4104 oscilloscope (Tektronix).
The voltage applied to the coil was read simultaneously on
Coil radius Rc (m) 0.01 0.011 another channel and the phase difference between these
Object radius Ro (m) 0.10 0.045 signals was measured using a specific oscilloscope func-
Object-coil distance doc (m) 0.13 0.075 tion. The phase shift at each position was obtained by
Grid divisions (each axis) 48 48 measuring the phase difference between these two signals
Grid parameter (m) 5 9 10-3 2.2 9 10-3 with and without the object and recording the difference
Frequency (MHz) 0.1, 1, 10 1–20 between these values. An aqueous solution of NaNO3 with
Current (A) 1 0.1 a concentration of 10% in volume and conductivity 10 S/m
Conductivity (S/m) 1 10 measured with CD-4303 (Lutron) meter was used as the
Dielectric constant 1,000 80 object in a cylindrical glass container with a diameter of
Relative permeability 1 1 0.09 m and a height of 0.2 m. The simulation conditions
using the impedance method for this analysis are shown in
Table 1.
maximum variation between successive iterations lower
than 10-13 A.
The numerical results for the spherical object were 3 Results
compared with an analytical model derived from the
analysis performed by Pham and Peyton [12]. These Figure 5 shows the current density distribution at three
equations are presented in Appendix. The analytical model different distances from the center of the sphere at a fre-
was implemented in a program developed in the MatLabTM quency of 1 MHz for polar angle ranging from zero to p
environment using 200 terms in the series expansions of radians. The discrete points correspond to the values cal-
functions for Eqs. A.1–A.5. culated numerically and analytically at the vertices of the
For measuring the phase shift of the magnetic field, we voxels. Therefore, the distance from each point to the
built the assembly shown in Fig. 4. A coil made of five center is not exactly the nominal values, but the nearest
turns with 0.022 m diameter AWG 20 wire was used to value in the grid. The numerical value of the current den-
generate the primary field. The measurements were done in sity for these graphs was obtained from the branch currents
15 positions arranged uniformly on a circle of 0.15 m in Ibx and Iby (see Fig. 3) using the coordinate transformation
diameter. The 16 devices (1 coil and 15 probe positions) given by Eq. 13:
were spaced with 0.39 rad (22.5°) between them. The coil  
was fed by a model TG2000 (Thurlby Thandar Instru- j/ ði; j; kÞ ¼ h2 Iby ði; j; kÞ cos /  Ibx ði; j; kÞsin/ ð13Þ
ments) generator with adjustable frequency from 1 to where / is the azimuthal angle in the grid in relation to the
20 MHz and amplitude of 20 V. The magnetic field was x direction. Numerical and analytical results agree well in
detected using the RSH 400-1 magnetic probe connected to most of the positions calculated for the three nominal

50
oscilloscope
generator
0.09 m num.
40
0.06 m num.

0.03 m num.
Jφ ( mA/m2 )

30 analytical

Ro
20
amplifier
object
doc probe
10

coil
2Rc 0
0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
polar angle (rad)

Fig. 4 Assembly for phase shift measurement. The dimensions Ro, Fig. 5 Current density on three different radii inside the sphere in
Rc, and doc are specified in Table 1 1 MHz

123
Med Biol Eng Comput (2011) 49:233–240 237

Table 2 Mean square errors for current density and magnetic field sensor 0.12 m from the object center. The numerical results
Parameter Radius (m) Error (%)
were obtained from the rectangular components of the
fields calculated with Eqs. 1 and 12 and transformed to the
100 kHz 1 MHz 10 MHz radial and polar components according to Eq. 14 and 15
Current density j/ 0.03 2.57 2.57 2.47 below:
0.06 2.28 2.28 2.25 Br ¼ By sinh  Bz cos h ð14Þ
0.09 2.32 2.32 2.30
Bh ¼ By cos h þ Bz sinh ð15Þ
Primary field Bpr 0.12 0.10 0.11 0.11
Primary field Bph 0.12 2.49 2.39 2.44 There is a good agreement between numerical and
Secondary field Bsr 0.12 3.18 3.17 2.58 analytical results for fields outside the sphere. However,
Secondary field Bsh 0.12 8.67 8.67 8.77 the mean square error in calculating the secondary field is
considerably higher than for the primary field. Table 2
shows the values of the average error at the three
distances. The maximum error in the worst case was about
frequencies used in this study. There is not a clear
10%, but the average error in the calculated set of points in
tendency for the error as a function of the frequency but
each curve was around 2.5%. Table 2 shows the values of
in all three cases the calculation of the secondary field has a
average errors in calculating the current density in the
higher error.
simulations with three different frequencies: 100 kHz,
Figure 7 shows the experimental and numerical results
1 MHz, and 10 MHz. The differences are small between
for the phase shift of the magnetic field due to the presence
different positions in the sphere and different frequencies.
of the conductive object in the 15 positions of the mea-
Figure 6 shows the distributions of primary and sec-
suring system and for three different frequencies: 5 MHz,
ondary field outside the sphere on the plane x = 0 at a
10 MHz, and 20 MHz. There is good agreement between
frequency of 1 MHz. This is the plane for placement of
the numerical and experimental results. The mean square
sources and sensors around the object. The radial distance
deviation of the numerical results compared to the mea-
used in these graphs corresponds to the positioning of the
surements is 2.4% for 5 MHz, 5.4% for 10 MHz, and 7.8%
for 20 MHz. The phase shift is maximum (in absolute
(A) value) in locations where the primary field is minimal.
10 4

2
10 2
Bp (nT)

radial
-2
10 0 -4 5 MHz

6
polar 0 50 100 150 200 250 300 350
-2
10
phase shift (degrees)

0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 5


position (rad)
0
(B)
-5
0.3
-10 10 MHz

-15
0.2 0 50 100 150 200 250 300 350
Bs (nT)

10

0.1 0
-10

0 polar -20 20 MHz


radial -30
0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 0 50 100 150 200 250 300 350
position (rad) angle (degrees)

Fig. 6 Primary and secondary field in the radial distance of 0.12 m Fig. 7 Phase shift of the magnetic field in the radial distance of
around the sphere in the plane x = 0 and frequency 1 MHz. Filled 0.075 m around the object in the plane x = 0 in three frequencies.
circle numerical. Solid line analytical Filled circle numerically calculated, open circle measured

123
238 Med Biol Eng Comput (2011) 49:233–240

4 Discussion accumulates the errors in a sequence of three processes:


calculating the primary magnetic potential, calculating the
The good agreement observed in Fig. 5 and Table 2 eddy currents, and calculating the secondary field. Fur-
between numerical and analytical results shows that the thermore, the secondary field is very weak, less than 1 nT,
primary magnetic potential calculation through numerical which contributes to the important influence that the errors
integration of the Biot–Savart equation and the current resulting from impedance modeling, space discretization,
calculation through the impedance method are sufficiently and numerical processing have on this calculation.
precise approaches in the conditions of frequency and The good agreement between numerical and measured
conductivity used in the simulations. This conclusion is values of phase shift in Fig. 7 shows that the numerical
also corroborated by the results shown in Fig. 6 for the approach can provide reliable values of sensitivity in MIT
field outside the sphere, but they also indicate that the for biomedical application. It appears that the phase shift
calculation of the secondary field generated by eddy cur- increases approximately linearly with frequency. This can
rents in the object can be done through the Biot–Savart easily be justified for objects of low conductivity, for which
calculation at least up to the highest frequency used in the the phase shift is small and proportional to the secondary
simulations. field as shown in Eq. 16 below. In turn, the secondary field
However, the methods used in this work are frequency is proportional to the intensity of induced currents in the
limited, since they are based on low frequency approxi- object, and, therefore, proportional to the frequency of the
mations of the electromagnetic theory to calculate both the applied field.
magnetic potential and the induced currents. The use of The phase shift of the external field is directly related to
Biot–Savart assumes that the portion of the radiated fields the sensitivity for conductivity variation in the pixels of the
is of little importance in this analysis. This is reasonable if image. The sensitivity calculation is straightforward when
the analysis is done within the near field of the radiator using the Biot–Savart and impedance method approach.
(coil or object), i.e. the distances do not exceed approxi- The current density distribution is calculated for a given
mately k/2p where k is the wavelength of the field. conductivity distribution assumed a priori in the object.
Assuming that k/2p is five times the maximum source- Then, Eqs. 1 and 12 can be used to calculate the primary
sensor distance inside the system, equal to 0.25 m (sphere and secondary field. The phase shift in any measurement
simulation), this criterion determines that the frequency is position can be predicted taking into account that the
limited to about 38 MHz. secondary field adds a small component 90o out of phase to
Furthermore, the electric model of the object does not the primary field. Thus, the phase shift can be obtained
include inductive effects; hence, it does not allow the through Eq. 16:
description of the effects arising from self-induction of the  Np
eddy currents. Therefore, the skin effect cannot be Bsm Bsm X
D/m ¼ a tan ffi ¼ Smn rn ð16Þ
observed and the results for the induced current, in prin- Bpm Bpm n¼1
ciple, are reliable only if the dimensions of the object are
where the indices m and n identify the measuring position
much smaller than the penetration depth of the field.
and the pixel position, respectively, and Np is the number
The penetration depth on a flat surface of a semi-infinite
of pixels in the image. Using the expression of Bs from
volume depends on the frequency and conductivity
(12), the sensitivity can be written in the following form:
according to the equation: d = (pf lor)-1/2. For the con-  
ductivity of 1 S/m and 1 MHz, this value is 0.5 m. And for lo h X Nv
Ibx ux þ Iby uy þ Ibz uz i ðrm  ri Þ
Smn ¼  ur
10 MHz, it is about 0.16 m. These results suggest that an 4pBpm rn i¼1 j rm  ri j 3
object with a diameter of 0.2 m, such as the sphere used in
ð17Þ
this work, cannot be modeled correctly at 10 MHz. How-
ever, if the object is finite and not flat, the skin effect is not where Nv is the number of voxels of the grid corre-
the main factor that determines the distribution of current. sponding to each pixel in the image plane and ur is the
In this case, the current density depends more on the object radial unit vector in the measurement position. The image
shape. As shown in the results presented in Fig. 6 and reconstruction in MIT is based on obtaining the solution
Table 1, in the comparison between analytical and of the system of equations such as (16), from the
numerical results for 100 kHz, 1 MHz, and 10 MHz, the knowledge of the sensitivity matrix of the system and
similarities observed indicate that the influence of skin measurements of phase shift of the magnetic field in the
effect is rather small. presence of the object for all possible combinations of
It is observed in Table 1 that the average error in cal- source and sensor positions have been previously per-
culating the secondary field is higher compared to the formed. Thus, the distribution of conductivity inside the
primary one. This is due to the fact that this calculation object can be obtained.

123
Med Biol Eng Comput (2011) 49:233–240 239


Different studies already published on numerical mod- rc X 1
2n þ 1 Inþ0:5 ðarÞ a nþ1

eling for MIT used different approaches for field or sen- j/ ðr; hÞ ¼ pffiffiffiffiffi
2aa ar n¼1 nðn þ 1ÞIn0:5 ðarÞ ro ðA:1Þ
sitivity calculation. Igney et al. [5] carried out this
modeling using finite difference-based software and the  P1n ðcos ho ÞP1n ðcos hÞ
sensitivities were computed by setting the conductivity in where In represents the modified Bessel functions of the
each perturbation voxel in turn to 1 S/m and keeping all of first kind and Pn are the associated Legendre functions of
the others as free space. This procedure requires a large the first kind. The constants that appear in this equation are:
computational effort since for each source position and for the sphere radius a, the polar coordinates of the coil (ro, ho)
each perturbation voxel, the complete calculation in all in relation to the sphere center, the propagation constant
sensor positions should be performed twice to determine pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
a ¼ ð1 þ iÞ xlr=2, and the constant c ¼ ixlIs ro sinho .
the variation of the field. Hollaus et al. [4] used the finite
The primary field (Bpr and Bph) and secondary field
element method for calculating the fields and the sensitivity
(Bsr and Bsh) in radial and polar directions outside the
was obtained by applying the reciprocity theorem to cal-
sphere are given by Eqs. A.2–A.5 below:
culate the mutual impedance variation between source and
sensor. Thus, the assembly of the sensitivity matrix ic X
1 n1 nþ1
r ro d  
Bpr ¼ P1n ðcos ho Þ sinhP1n ðcos hÞ
requires only two steps of field calculation for each source- 2x n¼1 nðn þ 1Þ dðcos hÞ
sensor pair, significantly reducing the amount of compu-
ðA:2Þ
tation required. Using the impedance method, only one
calculation of the induced currents is required for each ic X
1
r n1 ronþ1 1
Bph ¼ Pn ðcos ho ÞP1n ðcos hÞ ðA:3Þ
source position. Thereafter, the calculation of sensitivities 2x n¼1 n
is done directly from the Eq. 17. Therefore, there is a very
significant reduction of the computational effort. This is ic X
1
2n þ 1 Inþ0:5 ðaaÞ 1
Bsr ¼ 
even more important when one considers that the image 2x n¼1 nðn þ 1Þ aaIn0:5 ðaaÞ 2n þ 1
reconstruction requires that an iterative process be per- ða=ro Þnþ1 ða=r Þn 1 d  
formed. Starting from an initial estimate of the object Pn ðcos ho Þ sinhP1n ðcos hÞ
r2 dðcos hÞ
conductivity, in every step of the process, the sensitivity
matrix is calculated and from the inverse calculation, a new ðA:4Þ
estimate of the conductivity is achieved. This process may ic X
1
2n þ 1 Inþ0:5 ðaaÞ 1
require several iterations, and, therefore, the calculation of Bsh ¼ 
2x n¼1 ðn þ 1Þ aaIn0:5 ðaaÞ 2n þ 1
sensitivity must be optimized for maximum processing ðA:5Þ
speed. ða=ro Þn ða=r Þnþ1 1
Pn ðcos ho ÞP1n ðcos hÞ
This study showed that the impedance method and r ro
numerical integration of the Biot–Savart equation provide
reliable results on the modeling of the magnetic field and
eddy currents in magnetic induction tomography for bio- References
medical application. Based on these methods, we propose a
sensitivity calculation that accelerates the process of image 1. Gabriel C, Gabriel S, Corthout E (1996) The dielectric properties
reconstruction in MIT. of biological tissues: I literature survey. Phys Med Biol 41:
2231–2249
2. Gencer NG, Nejat TM (1999) Electrical conductivity imaging via
contactless measurements. IEEE Trans Med Imaging 18(7):
617–627
Appendix 3. Griffths H (2001) Magnetic induction tomography. Meas Sci
Technol 12:1126–1131
4. Hollaus K, Magele C, Merwa R, Scharfetter H (2004) Fast cal-
Using the quasi-static approximations for Maxwell’s culation of the sensitivity matrix in magnetic induction tomog-
equations, Pham and Peyton [12] obtained the distribution raphy by tetrahedral edge finite elements and the reciprocity
of electric field inside and outside a spherical object near a theorem. Physiol Meas 25:159–168
5. Igney CH, Watson S, Williams RJ, Griffths H, Dossel O (2005)
circular coil concentric with the object. This field has only Design and performance of a planar array MIT system with
an azimuthal component. From this result, we calculate the normal sensor alignment. Physiol Meas 26:263–278
current density inside the sphere using the relationship 6. Merwa R, Hollaus K, Biró O, Scharfetter H (2004) Detection
j = rEand the magnetic induction outside the sphere using brain oedema using magnetic induction tomography: a feasibility
study of the likely sensitivity and detectability. Physiol Meas
Faraday’s Law B = (i/x)r 9 E, where r is the conduc- 25:347–354
tivity of the sphere. The current density is given by 7. Ming-Xin Q, Li-Cheng J, Hai-Bing W, Hua LV (2005) Simula-
Eq. A.1: tion study of detecting conductivity of brain tissues with

123
240 Med Biol Eng Comput (2011) 49:233–240

magnetic induction based on FDTD. In: Proceedings of IEEE planar gradiometers: data collection and calibration. Physiol
engineering in medicine and biology, 27th annual conference, Meas 27:271–280
Shangai, China, pp 4369–4672 15. Soleimani M (2005) Image and shape reconstruction methods for
8. Morris A, Griffths H, Cough W (2001) A numerical model for electrical impedance and magnetic induction tomography. PhD
magnetic induction tomography measurements in biological tis- Thesis, The University of Manchester, Manchester, United
sues. Physiol Meas 22:113–119 Kingdom
9. Nadeem M, Thorleif T, Gandhi OP (2003) Computation of 16. Vauhkonen M, Hamsch M, Igney CH (2008) A measurement
electric and magnetic stimulation in human head using the 3-D system and image reconstruction in magnetic induction tomog-
impedance method. IEEE Trans Biomed Eng 50:900–907 raphy. Physiol Meas 29:445–454
10. Orcutt N, Gandhi OP (1988) A 3-D impedance method to cal- 17. Watson S, Willians RJ, Cough W, Griffths H (2008) A magnetic
culate power deposition in biological bodies subjected to time induction tomography for samples with conductivities below 10
varying magnetic fields. IEEE Trans Biomed Eng 35:577–583 S/m. Meas Sci Technol 19:1–11
11. Orcutt N, Gandhi OP (1990) Use of the impedance method to 18. Zolgharni M, Ledger PD, Armitage DW, Holder DS, Griffths H
calculate 3-D power deposition patterns for hyperthermia with (2009) Imaging cerebral haemorrhage with magnetic induction
capacitive plate electrodes. IEEE Trans Biomed Eng 37:6–43 tomography: numerical modeling. Physiol Meas 30:187–200
12. Pham MH, Peyton AJ (2008) A model for the forward problem in 19. Zolgharni M, Ledger PD, Griffths H (2009) Forward modeling
magnetic induction tomography using boundary integral equa- of magnetic induction tomography: a sensitivity study for detect-
tions. IEEE Trans Magn 44(10):2262–2267 ing haemorrhagic cerebral stroke. Med Biol Eng Comput 47:
13. Ramos A, Raizer A, Marques JLB (2003) A new computational 1301–1313
approach for electrical analysis of biological tissues. Bioelec- 20. Zou Y, Guo Z (2003) A review of electrical impedance tech-
trochemistry 59:73–84 niques for breast cancer detection. Med Eng Phys 25:79–90
14. Rossel-Ferrer J, Merwa R, Brunner P, Scharfetter H (2006) A
multifrequency magnetic induction tomography system using

123

You might also like