2009 IEEE International Advance Computing Conference (IACC 2009)

Patiala, India, 6-7 March 2009

A FEM-Based Forward Solver for Studying the Forward Problem of Electrical

Impedance Tomography (EIT) with A Practical Biological Phantom

Tushar Kanti Bera J. Nagaraju

Department of Instrumentation Department of Instrumentation
Indian Institute of Science-Bangalore Indian Institute of Science-Bangalore
Kamataka, India - 560012 Kamataka, India - 560012
Email: [email protected] Email: solarjnr(isu.iisc.emet.in

Abstract- A Finite Element Method based forward solver is also depends on the Jacobean matrix formation, regularization
developed for solving the forward problem of a 2D-Electrical technique and the converging nature of iterative algorithm
Impedance Tomography. The Method of Weighted Residual [17] used in inverse problem. Before reconstructing the
technique with a Galerkin approach is used for the FEM conductivity from the measured potentials it is advantageous
formulation of EIT forward problem. The algorithm is written in to eliminate the measurement error produced by the phantom
MatLAB7.0 and the forward problem is studied with a practical geometry and EIT-hardware for better image reconstruction.
biological phantom developed. EIT governing equation is Hence it is very important to study and analyze the forward
numerically solved to calculate the surface potentials at the solver performance using a practical EIT-phantom prior to the
phantom boundary for a uniform conductivity. An EIT-phantom
mage reconstruction
is developed with an array of 16 electrodes placed on the ine Element process. In this direction solver
(FEM) [18] based forward
Method forward
a Finite
is
surface of the phantom tank filled with KCl solution. A developed and the is
problem of 2-D EIT studied with
sinusoidal current is injected through the current electrodes and
the differential potentials across the voltage electrodes are
develectrod phantorw aproblem FeMof
a 16 electrode phantom. A detail interpretation of FEM
measured. Measured data is compared with the differential formulation using Method of Weighted Residual (MWR) with
potential calculated for known current and solution conductivity. a Galerkin approach is presented for better understanding of
Comparing measured voltage with the calculated data it is 2D-EIT forward algorithm. Differential potential is measured
attempted to find the sources of errors to improve data quality at the phantom boundary by injecting sinusoidal current.
for better image reconstruction. Differential potential is also calculated using the forward
solver for a known current and solution conductivity.
Keywords-electrical impedance tomography, forward solver, Comparing measured voltage (Vdm) with the calculated data
forward problem, inverse problem, finite element method, EIT (VdS) it is attempted to find the sources of error associated with
phantoms, common modefeedback. the electronic hardware, phantom geometry and the electrode
array structure to improve data quality.
I. INTRODUCTION
Electrical Impedance Tomography (EIT) is an imaging l
modality which is being researched in different fields of
science and engineering due to its several advantages [1].
Being a non-invasive, non-radiating and inexpensive
methodology, EIT is researched in medical sciences [2],
biotechnology [3] and industrial process tomography [4].
Attempts are also being made to apply EIT in landmine
_ A=C
IMAGE RECONSTRUCTION
detection [5], semiconductor wafer characterization and [6],
nondestructive testing of brick walls [7] and rocks [8]. In PHANTOM W rrH
medical diagnosis bedside measurement facility is an EN
important advantage for neonates, pregnant women and the _
patients under intensive care and hence the impedance
imaging finds a good application in different clinical
investigations [9-14]. The first impedance imaging system, the
Impedance Camera [15], was constructed by Henderson and E_
Webster to study the pulmonary edema in 1978. But due to
poor signal to noise ratio [2] and poor spatial resolution [16]
VQtag
Measrement
imaging. The reconstructed image quality in impedance
tomography greatly depends on the performance of forward
solver and the measurement errors which is influenced by the
phantom geometry, electronic behavior of the current injector, Figure-i: An EIT system with electrode array on the domain to be imaged.
data acquisition module and signal conditioners. Image quality

978-1-4244-2928-8/09/$25 0OO © 2009 IEEE 1375

 II. MATERIALS AND METHODS Where, 1] is the outward unit normal vector on electrode
surface.
A.A. EiT
FIT
Forward Problem. A relation between the measured
EIT is a non linear inverse problem in which the electrical potentials and the domain conductivity can be obtained [22]
conductivity distribution across a closed domain (Q) of from the equation (1) using FEM as:
interest inside a volume conductor is reconstructed from the
surface potentials measured at the domain boundary (OQ). A
low frequency (in between 10 kHz to 1 MHz) [19] constant [D] = [K][y] (2)
current is injected through an array of 16X2' (n= 0, 1, 2, 4...)
electrodes surrounding the domain to be imaged and the Where [ci] is a vector of conductivity values, [(D] is the vector
potential developed on the boundary electrodes is measured. of voltage measurements and [K] is the transformation matrix
The voltage data collected by the data acquisition system is relating p to (. If K and ( are known, equation (2) can be
processed in PC using an image reconstruction algorithm solved numerically to calculate the nodal potentials of the
consisting of forward and inverse solver (figure-1). domain for the known current injection. It is known as the
Governing Equation. Electrical potential ((p) produced by "forward problem".
a low frequency (<100 k Hz) sinusoidal current applied to a Inverse Problem. From the equation (2) we get,
homogeneous and isotropic medium with low magnetic
permeability (biological tissue) can be represented as [20]: (-3 6
=Kd) ~~~~~(3)
E =-V b(x, y) That means if transformation matrix K and the surface
potentials 1 are known then the conductivity can be mapped.
The point form of Ohm's law gives, This is known as the "inverse problem".

J = TX Y)E(X, Y) Image Reconstruction Algorithm. Potentials to be

developed within the domain for known (initial guess)
Where, J is the current density, a is the conductivity. If there is conductivity (so) is numerically calculated using forward
no current source within the domain, we have, solver and compared with the measured potential data. Using
Modified Newton Raphson (MNR) iterative technique [22],
V J-o [17] (o is modified to achieve a specified error limit in the
difference between calculated and measured voltages using the
Combining the above equations, we get equation:

V.u(x,y)Vsb(x,y>=7 (1) k = 07 + k (4)

Where,
Where, V= a 1 + a ak = the conductivity for the kth iteration
ox x Fy
axo0= initial guessed conductivity
AGs = conductivity update for the kth iteration, defined as,
This nonlinear partial differential equation representing the '
Au [FTF + I] 1l-_
electro-dynamics of EIT is known as the Governing Equation
of EIT [21] and has an infinite number of solutions. Boundary
1F- [V1dS
* J.
V
(5)
conditions required to restrict these solutions can be applied to
specify the value of certain parameters on the surface. These Where, F is the Jacobean matrix [27] which is defined as,
may be either the potential at the surface (Dirichlet
conditions), the current density crossing the boundary
(Neumann conditions), or mixed conditions. [F] - aVds (6)
i) Dirichlet Boundary Condition. a(h
0 0k Where, i =1, 2 ... m, are the measured potentials on the
=
Where, g = 1, 2 ... number of element, h = 1, 2... M [M =

electrodes [m = number of electrodes]. (number of data measured per current projections) x (number
of current projections)] and I is identity matrix, X is the
ii..Neumann Condition..
ii) Neumann Boundary Condition. regularizing parameter [23].

on the source electrode B. Forward Solver Computation

__0 on th sikeecrd Forward Problem (FP) is the basis of EIT and it is
U 0 J 011 1 ~~~~~~~~~~~~~essentially
to be solved to calculate the boundary potential for
n ~~otherwise estimating the conductivity update (/AG) for all iteration in

11376 2009 IEEE Internaftionafl Advalnce Computing Conference (IACC 2009)

Inverse Problem (IP). A FEM based EIT-forward solver is statement and hence it is the most versatile approach to
developed in MatLAB7.0 using MWR with Galerkin approach deriving element properties is the Method of Weighted
to calculate the boundary potentials for a circular domain. Due Residuals [26]. The finite element formulation of EIT using
to its ability to model arbitrary geometries and various MWR with Galerkin approach is used to develop the EIT
boundary conditions the FEM is the most common method forward solver though there are several different methods used
currently used for the numerical solution of EIT problems in weighted residuals such as Collocation [18], Least Squares
[24], [25]. The finite element method reduces a continuum [18]. A circular domain having diameter equal to that of the
problem of infinite dimension to a discrete problem of finite phantom is discretized with a finite element mesh having 2048
dimension in which the nodal values of the field variable and triangular elements and 1089 nodes as shown in figure-2. Let
the interpolation functions for the elements completely define us assume the triangle of the figure 3(a) represents an
the behavior of the field variable within the elements and the elemental block whose vertices are denoted by 1(xi, y1), 2(x2,
individual elements collectively define the behavior of the Y2) and 3(x3, y3) having their corresponding shape functions
field over the entire medium [26]. NI, N2 and N3 respectively and nodal potentials (p, qP2 and qP3
Finite Element Method (FEM): The finite element respectively. In a 2-D domain the triangle is a first order
formulation for EIT [26], [27] first discretizes the medium
element
element and the potential
Hncthe varies linearly
nodalpotenilcne within as:all the
exrese
under analysis into a finite number of elements collectively
called a finite element mesh. Within each element the field
variable is approximated by shape functions (interpolation S(x, y) p + qx + ry [i x y][p q r]T (7)
functions) that are defined only within the individual element
in terms of the values of the field variables at specified points
on the element called nodes. Most EIT work uses linear shape
functions in which all nodes lie on the element boundaries
where adjacent elements are connected [26]. Method of
Hence, q0 (Xl,yl) =l p+ qxl + ry, (8)
Weighted Residuals (AIIWR) [18] is the most versatile approach
[26] used to formulate the finite element problems although in 1k (x2 Y2)02 P + qx2 + ry2 (9)
some cases the Direct Approach [26] and Variational
Approach [18] are also used. 03(X3,y3) 03 p + qX3 + rY3 (10)
13
Using (8), (9) and (1O), equation (7) yields,
]'~ ~~~~~~~~~~~~~~~~~O Y01xXiO,
y,1 ,y, xi 011
1_ V l Lf11 AX, Y) 02 x2 Y2 + 02 y2.x+ 1 x2 022.Y
L J X3 Y3 10l53 3 13 03 j

( l \=_ = = S = ~~~~~~~~~~~Where
=2D ofx Area the element(A)=21 x2 \Y2 l
XI

1 x3 y3

:
0(X, Y) =-01
D [{(X2Y3 - Y2X3 )+ (Y2 - y3 )X (X3 - X2 )y}
+
4 6
+ 02 {(X3YY-X1 Y3 ) +(Y3 -Y x+ (x1 -X3 )yf
Figure-2: Current injection to the discretized domain having 2048 elements + 03 {(X Y2 - X2 Y1 ) + (Y1 - Y2 )x + (X2 -X )Y}]
and 1089 nodes

J W
Hence, 5(X,Y)> ,
y3i4
(ai +bix+c
~~~~~~~~~~~~~~~~~~i=1 D
(12)

_< _ Thus, the potential within a typical triangular element in a 2D

mesh can be approximated by:

Figure-3: (a) A single triangular element, (b) Practical Biological Phantom y (1 3)

Method of Weighted Residuals (MWR): The weighted

residuals approach begins with the governing equations of the Where, Ni is called the shape function which is defined as
problem and proceeds without relying on a variational

2009 IEEE International Advance Computing Conference (IACC 2009) 1377

 1 1 Now for a scalar function a, and a vector function B, the
NE = D (ai + ciy) = 2A (ai + b;X + ciy) (14) "vector derivative identity" [26] equation (23) yields,
Hence, the shape functions at the three nodes of the element
are expressed as:
1[V (WcV OFEM)dQ = f[OV OFEM VW JQ
Q
(24)

=1 {(xy Invoking Gauss' Theorem on (24) permits the introduction of

N 2A (X2Y3 -y2x3)+(y2 -y3)x+(x3 -x2)y} (15) boundary conditions:

N2 2A4 {(x3y1 -x1y3)+(y3 -y1)x+(xI -x3})y} (16) fWO-V FEM ndz J[uVO!FEM VWIPQ (25)
1
N3 = {(xy2 -x2y1)+(yI -y2)x+(x2 -x)y} (7
(17)
Where, V . n 0-0and F = 7Q represents the boundary. So,
Shape functions are interpolatory on the three vertices of the an
triangle. Each function, Ni is zero at all vertices except one equation (24) can be rewritten as:
where its value is one:
NZ(X/,Y / )=(
(18) JCTVjOFEM VWdQ LWO X (26)

The left side of equation (26) is for the entire mesh. When
Where, i = 1, 2, 3 and j = 1, 2, 3 examined for a single triangular 2D element, k, the left hand
Equation (13) completely defines the potential within the side of the equation (26) is given by:
triangular element as a function of the values of the potential
at the element's three nodes. Let us consider the approximate
solution found by FEM and the exact solution of the equation
JO-kVO *VWdQ (27)
Ek
(1) are given by PFEM and Pe respectively. Hence from the
equation (1) we get, Substituting the definitions of the interpolating versions of d)
and W yield:
V.o7S0e =0 and V.c75~
SE __________________
3 3

Let VA 7VOFEM=R (19) JkVl VWdQ

Ek
Z
/=1
ZIj EkJVN, VN1dQ
]=1
(28)
Where, R is called the "Residual" 3
3i
Hence, V * CV0FEM V cFV0e R (20) Ek i=1 =1 (29)

To establish a numerical procedure to get the solution 41FEM as

the approximate solution of the equation (1), we force R to be
zero using the following equation: Sk - JVN1 VN;dQ
Ek (30)
[WRd Q = O
(21) Thus applying Finite Element Method (FEM), we get a 3 x 3
matrix for each element of the mesh. This 3 x 3 matrix is
known as the local stiffness matrix. In 3D each element
n produces a 4 x 4 matrix. In terms of the entire domain, the left
W(x) = iN (X) side of (26) will be
i=1 (22) K 3 3
where wi are the coefficients that weigh the interpolation 7V4 VWdQ= ,
kY, (31)
functions Ni. Using equation (19) and (21) we have,

fL w(v .cN3F£M)dQ -0° (23)

Where, i,j=1,2,3,...N

l 378 2009 IEEE Internaftionafl Advalnce Computing Conference (IACC 2009)

Similarly, right hand side of the equation 26 reduces to: [I] [ i * * n l, i]T (39)

00 3 3
f W7 d -= 57 wY0 NVNj n^d (32) All the elements in [I] will be zero except the elements
On i=1 j=1 representing the nodes at which the current is injected.

The ability to formulate solutions for individual elements Hence, [I]= [o 0 + z . . 0 -I * * o f]T (40)
before putting them together to represent the entire problem is
an important advantage of the finite element method. For a Using the phantom geometry and coordinates and bathing
single element of the FEM, equation (26) becomes: solution conductivity, [K] matrix is constructed and for a
current value, 1 mA (r.m.s.), [I] matrix is constructed. Using
3 3 33 3 LU decomposition technique in MatLAB equation (37) is
kZ SU
ij Uk Z Z 0 N n solved and nodal potentials are calculated. The nodes under
i=1 j=1 i=1 j=1 (33) the electrodes are extracted and corresponding potentials
called electrode potentials are calculated converted to the
Calculating Local Stiffness matrix: The elemental stiffness differential potentials. The forward solver developed can
matrix is defined as, calculate the potential data for circular domain of different
diameters, different mesh refinements, different current
Se = JVN VN JdF (34) injection and 16X21 (n= 0, 1, 2, 4 ...) number of electrodes.
e
C. EIT-Phantom
Equation (35) is recognizable as equation (29) and can be An EIT phantom [28] is developed for studying EIT
written in matrix quadratic form as: forward problem. A cylindrical shallow crystallizing dish is
used for phantom tank and an array of 16 equally spaced
3 3 square stainless steel electrodes is pasted on the inner wall of
V Vq=Y ZZ
i=1
JVNi VNjdFqj
F
(35) the glass tank filled up with a 0.05% (w/v) KCl solution
(figure-3b). Electrodes were connected with the multiplexer
board through the low resistive flexible multi-strand copper
Now from equations 15, 16 and (17), we have, wires of equal lengths for getting an identical impedance path
for each electrode. A Common Mode Electrode (CME) [28] is
_ placed at the phantom centre and connected to the ground point
2(a + x + c,y)
bI of the EIT hardwires to reduce the common mode error [29] of
the electronic circuits. The solution conductivity is essentially
N2 =-(a2 + b2x + c2Y) to be known to solve the forward problem of EIT. A liquid
2A conductivity measurement setup consisting of a glass chamber
N =-I(a3
2A
+ b3x + c3y) (20 mm x 20 mm x 50 mm) and two square (20 mm x 20 mm)
SS electrodes is developed. Electrical impedance of the KCl
solution is measured with a test signal of ImA using a LCR
Hence the elemental stiffness matrix K(e) is given by: meter (Model-QuadTech 7600, QuadTech Inc.USA) and the
conductivity is calculated as 0.21 S/m.
b I 2 + C2
1
cl2I b 2 +c 2 b3 + c 3
1 D. Instrumentation
e bb±c2c1 b2 ±c2 b2b3 c2c3 (36) A simple electrode switching multiplexers is developed
e b3b +c3cC b3b2+c3c2 b32+c32 using single pole single throw (SPST) slide actuated DIP
switches [30]. A 50 kHz sinusoidal function generator (VCO)
After assembling all the elemental stiffness matrices the is developed using MAX038 IC (Maxim Inc.) [31]. A variable
frequency constant current injector is developed using the
Globalqutifnesis matrix
of equations is got.
Kifomedanthfoloingsy
VCO and a modified Howland constant current source using
AD811 IC (Analog Devices Inc.) [32]. ImA, 50 kHz signal
generated by the current injector is applied to the phantom
[K]nxn [(]nxl [Inxl(37) =
through 16 current electrode pairs using the Neighboring
Method [33]. The voltage signal developed on the voltage
Where, [@]nxl is the matrix of potentials at all the nodes and is electrodes is passed through an instrumentation amplifier and
given by, narrow band pass filter. The differential potential (Vdm) from
the filter output is measured by a Multimeter (Model -
[r.] * ]T (38) Keithley 2002, Keithley Instruments, Inc., USA) as well as
-i Y2
.4
*I'-i V'n Digital Storage Oscilloscope (Model - TDS3O14B, Tektronix
[I]nX1 is the matrix of currents at all the nodes and is given by, Ic)

2009 IEEE International Advance Computing Conference (IACC 2009) 1379

 III. RESULTS AND DISCUSSION mV) at the voltage electrode pair opposite to the current
The differential potential is calculated with the phantom electrodes (figure-4). Vdm data is further corrected by
data provided by Ider et al [34] using our forward solver. The subtracting the offset potential from it and it gives almost
deviation between the potential data generated through our identical data at all the voltage electrode pairs and the
forward solver and that provided by Ider et al [34] is only difference reduced up to 0.05 mV (figure-6). But the voltage
0.08% to 0.80% (figure-4). Vd, is also calculated for different difference at the voltage electrode pairs nearer to the current
number of elements and the corresponding Vd, data is electrodes becomes 10.79 mV due to the high current density.
compared with the Vdm. Result shows that the optimum Measured voltage data after correction becomes almost similar
number of elements for which the algorithm converges to the calculated one which is desired for a better 2D
satisfactorily in a 2.4 GHz PIV-PC with, 1.5 GB RAM is 2048 reconstruction. Use of a floating load VCCS and complete
(figure-5). differential voltmeter may contribute to the more data
correction. Vdm is also measured with different solution height
v data from Iler etal
above or below the electrode array and optimum solution
levels are set.
frotm FS; d:eveopedIl
,50 50 5Q
F Withut CME

X ~~~~~
~~~~~~~~~~~~~d .....': ...... E40 V 40
~~~~ ~~~~~~~~~~~~~~~~~~~~~
um gI

X 30 30
30 30

20 20
~20 m 20

10- S W? _10 X'

= 11 i -.~~~~~~~~~~
=_ |'S 1
0~~~~~~~~~~~~~~~~~~~1
0~~, I I0 I I, 0 /, R S

VI V2 V3 V4 VS VS V7 VB V9 VIOVI VI2Vl 3
Scanning Across Voltage flectrode Nirs 0- I _
Vl V2 V3 V4 VS V6 V7 VS V 10VII2 Vl VI3
Scanning Across Voltage Electrode Pairs

f1r the Mesh with 32 ElemenSlts l

for the Mesh witi 128 Eltments
404 V~for- tho Moth with 512 E1emontu 40
for th| Me|shwith 204S Etemets 1
Figure-5:_Vds data for meMesh
the with 8192 Etmeflements
30

20 -20

10 10

ID0
vi V2~V3 V4 V5 V V7 V8 VS V10VlIIVl2Vl3
Scanning A-cross VoltWg Electrode Pairs
Figure-5: Vd, data for the meshes with different number of elements
Measured differential voltage is compared with the
calculated data. It is noticed that the offset potential developed
due to the electronic circuit and half-cell potential of the
electrode material deviates the Vdm curve from the Vd, curve
with a difference of around 5.32 - 10.86 mV. The offset
potential is reduced by a Common Mode Electrode (CME) put
at the center of the phantom tank and more analogous data is
achieved. The differential potential is reduced maximum (23.2

11380 2009 IEEE InternaftionaflAdvalnce Computing Conference (IACC 2009)

[5] Philip Church, et al. Electrical Impedance Tomographic Imaging of [28] Holder D. S. et al Some practical biological phantoms for calibrating
Buried Landmines, IEEE Transactions on Geoscience and Remote multifrequency electrical impedance tomography, 1996 Physiol. Meas.
Sensing, Vol. 44, No. 9, September 2006 2407 17 A167-A177
[6] F Djamdji et al., Electrical impedance tomography applied to [29] Rosell; Javier et al, Common-mode feedback in electrical impedance
semiconductor wafer characterization, Meas. Sci. Technol. 7 (1996) tomography, Clin. Phys. Physiol. Meas., 1992, Vol. 13, Suppl. 4 11-14.
391-395. Printed in the UK [30] Data Sheet, DIP Switches, EDG/EDS type, Excel Cell Electronic Co.
[7] Jerzy HOLA, New Method of Investigation of Rising Damp in Brick Ltd., No.20, 25th Rd., Taichung Industrial Park, Taichung, Taiwan.
Walls by means of Impedance Tomography, 17th World Conference on [31] Data Sheet, MAX038-High-Frequency Waveform Generator, Maxim
Nondestructive Testing, 25-28 Oct 2008, Shanghai, China, Integrated Products, Inc. CA 94086.
[8] J eong-Seok Yang et al, Nondestructive Imaging of Rock Sample Using [32] Data Sheet, AD811 - High Performance Video Op Amp, Analog
Electr ical Resistance Tomography: Theor etical Appr oach, Journal of Devices, Inc.,
the Earth system, Vol. 41, No. 4 (2004) pp. 265-270
[33] Grootveld; C. J., Measuring and Modeling of Concentrated Settling
[9] Bagshaw, Andrew P. et al; Electrical impedance tomography of human Suspensions Using Electrical Impedance Tomography, PhD Thesis,
brain function using reconstruction algorithms based on the finite Delft University of Technology, The Netherlands, 1996.
element method, Neurolmage 20 (2003) 752-764
[34] Ider et al. Electrical Impedance Tomography of Translationally Uniform
[10] Li; Ying et al, A Novel Combination Method of Electrical Impedance Cylindrical Objects with General Cross-Sectional Boundaries, IEEE
Tomography Inverse Problem for Brain Imaging, IEEE Transactions on Transactlons On Medical Imaging, Vol. 9. No. I March 1990 49
Magnetics, Vol. 41, No. 5, May 2005
[11] F J McArdlet, et al, An assessment of dynamic images by applied
potential tomography for monitoring pulmonary perfusion, Clin. Phys.
Physiol. Meas., 1988, Vol. 9, Suppl. A, 87-91.
[12] Hope; Tyna A. and Iles; Sian E, Technology review: The use of
electrical impedance scanning in the detection of breast cancer, , Breast
Cancer Res 2004, 6:69-74 (C 2004 BioMed Central Ltd.
[13] Hoetink; A. E. et al, Imaging of Thoracic Blood Volume Changes
During the Heart Cycle With Electrical Impedance Using a Linear Spot-
Electrode Array, IEEE Tran. on Med. Imaging, Vol. 21, No. 6, '02, 653
[14] Zapata-Ferrer; Angel R. et al, Electrical Impedance Tomography: An
Electronic Design, With Adaptive Voltage Measurements and A
Phantom Circuit For Research In The Epilepsy Field, Proceedings - 19th
International Conference - IEEE/EMBS Oct. 30 - Nov. 2, 1997, USA
[15] Henderson, Ross P; Webster, John G. An Impedance Camera for
Spatially Specific Measurements of the Thorax, IEEE Transactions On
Biomedical Engineering, Vol. Bme-25, No. 3, May 1978
[16] W. D. Hou et al, "Increasing image resolution in electrical impedance
tomography," Electronics Letters, vol. 38, pp. 701-702, 2002.
[17] Yorkey T. J. "Comparing reconstruction methods for electrical
impedance tomography" PhD thesis, University of. Wisconsin at
Madison, Madison, WI 53706, 1986
[18] J. N. Reddy, An Introduction to the Finite Element Method, 3rd Ed., 2nd
Reprint, TATA McGraw-Hill Publishing Company Ltd, 2006.
[19] Margaret Cheneyy, David Isaacsony, Jonathan C. Newelly, Electrical
Impedance Tomography, SIAM Review, 1999 Vol. 41, No. 1, pp. 85
[20] Chen-Ning Huang, Study of Rotational Electrical Impedance
Tomography, PhD thesis, NCU, Republic of China, 2008.
[21] Holder David S.(Editor), "Electrical Impedance Tomography: Methods,
History and Applications (Series in Medical Physics and Biomedical
Engineering)" Institute of Physics Publishing Ltd., 2005.
[22] Y. Ma, S. Miedema, V. Matousek, W. Vlasblom, Tomography as a
Measurement Method for Density and Velocity Distributions, Western
Dredging Association Twenty-third technical conference & Thirty-fifth
texas A&M Dredging Seminar, Chicago, Illinois, June, 2003, 25-36 blz
[23] Jaakko; Malmivuo, Plonsey; Robert, Bioelectromagnetism: Principles
and Applications of Bioelectric and Biomagnetic FieldsNew York,
Oxford Oxford University Press, 1995
[24] Murai T, Kagawa Y, Electrical impedance computed tomography based
on finite element model, IEEE Transactions on Biomedical Engineering,
1985, 32, 177-184.
[25] Pilkington TC, et al A comparison of finite element and integral
equation formulations for the calculation of electrocardiographic
potentials, IEEE Trans Biomed Eng, 1985, 32, 166-173.
[26] Bradley Michael Graham, Enhancements in Electrical Impedance
Tomography (EIT) Image Reconstruction for 3D Lung Imaging, PhD
thesis, University of Ottawa, April 2007.
[27] Silvester PP, Ferrari RL, Finite Elements for Electrical Engineers,
Cambridge University Press, 3rd edition, 1996.

2009 IEEE Inxternational Advanxce Computing Conference (IACC 2009) 138:1