Bioelectromagnetics 23:516^521 (2002)

Electrical ImpedanceTomography via

Filtered-Back Projection of Fan Current
Distribution: A Numerical Simulation
B.C. De Simone,1 R. Siciliano,1 A. Pachi0,2 C. Cametti,3 and F. De Luca3*
1
Dipartimento di Chimica, Universita’della Calabria, Rende, Italy
2
Seconda Clinica di Ostetricia e Ginecologia, Universita’ ‘‘La Sapienza’’, Roma, Italy
3
Dipartimento di Fisica and INFM, Universita’ ‘‘La Sapienza’’, Roma, Italy
An approach to electrical impedance tomography (EIT) data acquisition inspired by NMR-filtered
back-projection imaging with fan isochromat distribution is proposed. A current projection is gene-
rated by injecting current at a certain point of the sample and simultaneously collecting the current
itself at different points on the half space opposite the point of current injection. After that, the
injection is shifted to another point and collected as above and so on. A very simple algebra and
software support the numerical simulations. This method is expected to be more sensitive than the
traditional method based on potential measurements. A preliminary low resolution experiment is
presented. Bioelectromagnetics 23:516–521, 2002.
2002 Wiley-Liss, Inc.

Key words: impedance mapping; EIT; reconstruction algorithm; electrical impedance;

filtered back-projection

INTRODUCTION is injected in a point of the plane and is measured

simultaneously on an array of electrodes surrounding
Electrical impedance tomography (EIT) is an
the half plane opposite to the point of injection. Such
imaging technique, which maps some properties of
a data collection constitutes an electrical current pro-
solids and/or liquids through the spatial distribution
jection. Successively, the point of injection is shifted
of the electrical impedance [Webster, 1989]. The main
to its nearest neighbor electrode and the new current
fields for EIT applications are those concerning bio-
projection stands for the half plane opposite to the
logical materials, which, as it is well known, present
new injection point and so on for all the electrodes
structural and functional heterogeneity mostly corre-
surrounding the plane.
lated to electrical phenomena.
From an appropriate number of projections, it is
The usual EIT procedure utilizes a potential
possible to obtain an EIT map by a suitable recon-
approach [Barber et al., 1983]. The electrical spatial
struction algorithm. The assumption made in this
property of a sample (say a plane) is valued by pro-
approach is that each different current path is confined
ducing an electrical current flow between two adjacent
in an area that is practically coincident with the one
points of the plane border and measuring the potential
whose vertices are the injection point and two adjacent
drop between other adjacent couples of points along
pick-up points. This assumption is, of course, limited
the plane perimeter. All the electrodes are placed at
by the real behavior of the current. The electric current
equal spacing around the plane and, in turn, are utilized
paths, in isotropic media, coincide with the electric
in pairs either for current injection or for the mea-
field lines in accordance with the differential Ohm law,
surement of the potential drop between them [Brown
and Seagar, 1985, 1987]. The details that specify the
relation between the modality of current injection and
————— —
that of potential measurements identify the different *Correspondence to: F. De Luca, Dipartimento di Fisica and
EIT methods [Webster, 1989]. In all cases, the algo- INFM, Universita ‘‘La Sapienza’’, Piazza Aldo Moro, 2, 00185
rithms that support the EIT map reconstruction, the Roma, Italy. E-mail: [email protected]
more effective of which seems to be the modified
Newton–Raphson method [Yorkey et al., 1986], Received for review 6 March 2001; Final revision received 7
December 2001
request remarkable time-consuming computation.
In this article, we present a different method DOI 10.1002/bem.10044
to collect data for EIT. A current of known intensity Published online in Wiley InterScience (www.interscience.wiley.com).

2002 Wiley-Liss, Inc.

 Electrical ImpedanceTomography 517

so distortion of current paths is the same as distortion realistic goal of EIT should be a map showing a
of electric field. In inhomogeneous media, the electric qualitative contrast related to the impedance distri-
field lines are distorted due to the conductivity gra- bution. In this point of view the EIT method based
dient, so that r ~
~J ¼ 0, the electric current density ~
J on current measurements may constitute a good alter-
flows ‘‘around’’ obstacles made of areas with lower native to potential methods.
conductivity. Changes in the geometry of electric field To assess the feasibility of the method, we per-
and hence, in electric current, due to unknown hetero- form a numerical simulation in which current data are
geneity of the medium under study, will limit resolution calculated by a priori knowledge of the impedance
and accuracy of the impedance map reconstruction just distribution. The results show that current method for
as in traditional EIT. Our basic assumption, in fact, is EIT data collection is very promising and is able to
similar to that which supplies the equipotential line give good contrast impedance maps. The first result,
paths utilized in the usual EIT approach [Barber et al., through preliminary, of a low spatial resolution experi-
1983], although less restrictive. The conditions that ment, obtained with a rudimental EIT current scanner,
limit the knowledge of the electrical potential of every on a two-value impedance phantom, confirms the nu-
pixel in the plane are in fact the same as those that merical simulation and the simple management of this
prevent the knowledge of the current density on it. EIT approach.
Despite the current method having limitations
similar to those of the potential methods, it never-
THEORY
theless possesses some advantages with respect to
using potentials. The first is related to the very simple For the sake of simplicity, we suppose that sam-
algebra and software of the reconstruction algorithm ples are planar and have a circular geometry and that
needed to get the plane impedance. The second ad- N electrodes are distributed uniformly around their
vantage concerns its better sensitivity. The current perimeter (the extension to different geometry or to the
method, in fact, facilitates the spreading of current three dimensional case [Le Hyaric and Pidcock, 2001]
over the plane to be imaged while, in the potential is similar to those developed for X-ray tomography
approach, it remains largely confined around the pair and 3D-NMR back projection [Herman, 1980; Call-
of electrodes for the current injection. Furthermore, aghan, 1991]).
because of the existence of high numbers of zero We suppose a current I0 is injected in one of the
potential pick-up points in our approach, the current electrodes, the source S, and currents are picked up
has many ‘‘options’’ to get the best minimum imped- by N/2 electrodes, arranged on the semicircle dia-
ance paths and then, the more confined one. The metrically opposed to the source S (P line of Fig. 1).
confinement of the current paths could somewhat be
favored in biological structures, when the frequency of
the current is high enough. It is well known that cell
membrane becomes ‘‘transparent,’’ with respect to the
current flowing through it, as the frequency of current
increases [Schwan, 1957]. Even if this current behavior
is effective on a cellular scale, at frequencies high
enough it might support the confinement of the current
paths. In summary, the approach we propose, even if it
maintains in practice the fundamental limitation of the
potential methods, seems more sensitive than tradi-
tional EIT approach and is supported by simple and
easily handled algorithm software.
It could be argued that, in the current method, the
electrode impedance could make the assignment of
the plane impedance quite imprecise and therefore
reduce the sensitivity of the method itself. This is
principally true from a quantitative point of view,
which considers EIT maps established by the absolute
value of the local electrical impedance. As remarked
above, this is in any case unfeasible because only the
boundary data of impedance distribution, either cur-
rents or potential, are accessible by EIT. Therefore, the Fig. 1. Scheme of the ray coordinates.
518 De Simone et al.

The positioning of electrodes on the semicircle avoids where summation extends over the n,m rays which are
current paths of very low impedance with a compres- passing across the (x,y) point and Sn,m is the surface
sion of the map sensitivity. over which extends the integral of Equation 2. In
We call a ‘‘ray’’ the line connecting the source this way, each ray is assigned a constant impedance
S and a pick-up electrode; the current associated with density. The bar in Equation 3 emphasizes the dif-
a ray is that confined in the triangular area delimited ference between the calculated and the true impedance
by the bisector of the angle under which the source S density value. In fact, this calculation ignores both the
is connected to the nearest neighbor’s electrodes of the impedance distribution inside the ray and the fact that
specific pick-up electrode (area ASB in Fig. 1). A Z ðx; yÞ also receives points coming from outside the
‘‘fan’’ is the whole set of rays starting from the same original object because of back projection. Filtered
source. If all the pick-up electrodes are at the same back projection [Bracewell and Riddle, 1967] modi-
potential, then a ray of current is given by fies the single profiles to counter balance the blurring
of the reconstruction proceeding. In this method, the

V filtered projection are obtained according to [Laksh-
in;m ¼


ð1Þ
Zn;m
minarayanan, 1975; Herman and Naparstek, 1977]




Zn;m


where DV is the potential difference between the Zn;m ¼ D cosðm
gÞ  gm ð4Þ
source and the pick-up electrodes and Sn;m

Z m
g
where gm ¼ 12 ðsinðm
gÞÞ2 hm and D is the distance of the

Zn;m
¼ jZ ðx; yÞj dxdy ð2Þ fan source S from the origin of the reference frame,
n;m
namely the radius of the sample. The filter function is
given by
where jZ ðx; yÞj is the modulus of the impedance
KM sin2 ð2pKM m
gÞ
density (in this work we limit our attention only to hm ¼ sinð2pKM m
gÞ ð5Þ
the modulus of the impedance). The integral in pm
g p2 ðm
gÞ2
Equation 2 is evaluated over areas like that ASB of
Figure 1 is identified by the angular coordinates where KM ¼ ð2
gÞ1 corresponds to the maximum
mDg, with m ¼ 0, 1,. . ., (N2)/ 2 where NDg ¼ 180 8 angular projection sampling. From these values, we
and nDb, with n ¼0, 1,. . ., N1, where N D b ¼ 360 8 have
(Fig. 1). The angle b ¼ nDb from x axis characterizes ð8
g2 Þ
1
m¼0
the current source point on the sample circumference,
so it is the fan beam or projection angular coordinate, gm ¼ 0 m even : ð6Þ
 2 2 1
which is stepped by Db. The angle mDg characterizes  2p sin m
g m odd
the m-ray path of the b fan beam. The angle g ¼ m2Dg
from the projection axis (the P axis of Fig. 1: it is Finally we can obtain the Z ðx; yÞ distribution by back

rotated by an angle b-90 8 with respect to the x axis) projecting along the fan the N profiles Zn;m , that is
characterizes the pick-up point of the m-ray current X

along the sample circumference. The N/2 rays are Z ðx; yÞ ffi
g Zn;m : ð7Þ
distributed symmetrically around the source current n;m

point (Fig. 1). Accordingly, the N  N/2 rays are indi-

viduated by the couples n,m.
The set of m values Zn  fjZn;m jg gives the RESULTS AND DISCUSSION
nth projection. The reconstruction of image from
projections, taken at various bs, is common to many This approach has been verified by numerical
scientific areas as astronomy [Bracewell, 1956], optics simulation of a known impedance distribution by
[Rowley, 1969; Berry and Gibbs, 1970], and medicine calculating the current profiles or projections and
[Hounsfield, 1973]. By following the back projection back-projecting them after filtering. In Figure 2, the
approach and considering that n,m are discrete vari- relation between the current sources and the pick-up
ables, the reconstruction function is given by electrode array is schematized. We simulated sample


with both axial and nonaxial impedance distributions.
X
Zn;m
The first block of the flow chart software
Z ðx; yÞ ¼ ð3Þ estimates the N/2 current rays, separated by Dg from
Sn;m
n;m each other, for each of the N current projections of
 Electrical ImpedanceTomography 519

Fig. 3. a: Scheme of the axial phantom utilized for numerical

simulation with ZZAC ððx;yÞ
x;yÞ ¼ 50. b: Scheme of the non-axial phantom,
simulated with Z1(x,y)/Z3(x,y) ffi 1/50, Z2(x,y)/Z3(x,y) ffi 10.

The second simulation uses the nonaxial phan-

tom arranged as shown in Figures 3b. The inner cir-
cular homogeneous media have diameter d1 and d2 and
impedance Z1(x,y) and Z2(x,y), respectively, while the
Fig. 2. The schematic relation between the current sources greater circular homogeneous medium has impe-
and the pick-up electrodes. The currents Ib;g  In;m of a given
projection are simultaneously measured. The injection current
dance Z3(x,y) and diameter d3. For this phantom, the
I0 is maintained constant among the different projections. The following ratios have been fixed between dimension
other projections are obtained by rotating the current source S to and impedance density: d3/d1 ffi 6, d3/d2 ffi 3.5, Z1(x,y)/
next electrodes and by rotating the set of pick-up electrodes of the Z3(x,y) ffi 1/50, Z2(x,y)/Z3(x,y) ffi 10.
same angle. The generic n-projection is directly evaluated by
Equation 2 through the paths described by Equation 8.
Because of the simple geometry of phantoms, the
angular coordinates b ¼ nDb (see Fig. 1). For the sake integral of Equation 2 is replaced by a summation over
of simplicity, in this simulation each ray is constituted the rays or chord segments with constant impedance
by a straight line and the integrals of Equation 2 are density. The segment impedance is evaluated by sup-
evaluated along lines joining the source S and the pick- posing that it is simply proportional to the chord
up electrodes. segment lengths with an appropriate proportionality
This is evaluated on the basis of Equation 1. The coefficient, which, for example, is 50 or 1 for the
first step evaluates the current paths between the sample with an axial impedance distribution, depen-
generic current source point Sn and the generic pick- dending upon whether segments are outside or inside
up point Pn,m, which provides the current of the m-rays the inner circle (Fig. 3a). In these simulations, no
of the n projection. The current source point co- interpolation between adjacent rays is used, so that any
ordinates are given by Sn(R cos(nDb), R sin(nDb)), artifact due to the reconstruction algorithm is fully
where R is the sample radius. The pick-up current point observed.
coordinates of the n,m ray are therefore given by After having obtained the whole n,m set of
Pn,m(R cos(b m2g), Rsin(b  m2g)). The b fan impedance rays or the N projections, the impedance
beam equation or n-projection is therefore given by image has been reconstructed by means of Equations 4
and 7. The results are shown in Figure 4 for different
ðy  R sin n
bÞ ½sinðb  m2gÞ  sin n
b N’s. The figures show the good spatial definition of
¼ :
ðx  R cos n
bÞ ½cosðb  m2gÞ þ cos n
b the impedance map obtained with the back-projection
ð8Þ approach, even if many reconstruction ghosts are still
observable, especially for low N’s. This is, to some
The scheme of a phantom with symmetrical extend, typical of a crude back projection recon-
impedance distribution is shown in Figure 3a. It is struction approach, because it modifies the density of
composed by a circular distribution of a homogeneous information over the image plane and is complicated
medium of impedance ZC(x,y) surrounded by an by the pixel sizes distribution due to the fan geometry.
annular distribution of a homogeneous medium with In Figure 5, simulations of the phantom of
impedance ZA(x,y). The radius ratio is RA/RC ffi 3.5 Figure 3b are reported for different N’s. In this sample,
while ZA(x,y)/ZC(x,y) ffi 50. a full heterogeneous sample is considered, because
520 De Simone et al.

Fig. 4. Simulated EITof the phantom reported in Figure 3a. From Fig. 5. Simulated EIT of the phantom reported in Figure 3b.
A^D the number of projections is N ¼16, N ¼ 32, N ¼ 64, and From A^D the number of projections is N ¼16, N ¼ 32, N ¼ 64,
N ¼128. and N ¼128.

the different impedance values are distributed in an

asymmetric way. These images show the same ghost
effect found in Figure 4 and, the same comments may
be made; in addition, it should be emphasized that the
results are independent of the symmetry properties of
sample.
In Figure 6, the first preliminary experimental
result obtained with a coarse EIT scanner based on this
method is reported. The injected current was about
1 mA and N ¼ 32. The sample, similar to that report-
ed in Figure 3a (RA ffi 35 cm, RC ffi 11 cm, and ZA/
ZB ffi 106), was composed of physiological water with
a cylinder of solid polymer immersed in the center of
it (the C zone in Fig. 3a). The frequency of the injec-
ted current was 1 kHz, and the electrodes were the
commercial ones utilized in electrocardiography (1 cm
diameter).
Because of the low angular resolution and the
absence of a filter able to interpolate the signal be-
tween adjacent rays, the results of Figure 6 show the
various current paths assigned by the reconstruction
algorithm. The average experimental ratio ZC/ZA ffi 103
shown on the image matrix results to be much lower
than the phantom impedance, but high enough to make Fig. 6. First experimental result obtained with current injection.
invisible the ZA area (Fig. 3a) on figure. The reasons of The injected current was about 1 mA with N ¼ 32. The sample is
such a loss of contrast is probably due also to the low that sketched in Figure 3a.
 Electrical ImpedanceTomography 521

sensitivity of our current measurement and to the poor Brown BH, Seagar AD. 1987. The Sheffield data collection system.
spatial resolution, as well as to the electrode contact Clin Phys Physiol Meas A8:91–97.
impedance, which in our experimental arrangement is Callaghan PT. 1991. Principles of nuclear magnetic resonance
microscopy. Oxford: Clarendon Press. p 492.
quite high. Herman GT. 1980. Image reconstruction from projections.
The goal of this article was mainly to show that a New York: Academic Press. p 487.
method of current injection for EIT is possible and that Herman GT, Naparstek A. 1977. Fast image reconstruction
it works with good reproduction of the heterogeneity based on a Radon inversion formula appropriate for
of the sample impedance. Although the method needs rapidly collected data. SIAM J Appl Math 33:511–
533.
several improvements, especially for concerning the Hounsfield GN. 1973. Computerized transverse axial scanning
reconstruction algorithm, it seems quite promising; (tomography): part 1. Description of system. Br J Radiol
and the first preliminary experimental finding appears 46:1016–1022.
to confirm its potential. Lakshminarayanan AV. 1975. Reconstruction from divergent ray
data. Department of Computer Science, State University of
New York at Buffalo. Thesis Tech Ref 92.
REFERENCES Le Hyaric A, Pidcock MK. 2001. An image reconstruction
algorithm for three-dimensional electrical impedence tomo-
Barber DC, Brown BH, Freeston IL. 1983. Imaging spatial graphy. IEEE Trans Biomed Eng 48:230–235.
distributions of resistivity using applied potential tomogra- Rowley PD. 1969. Quantitative interpretation of three dimensional
phy. Electr Lett 19:933–935. weakly refractive phase objects using holographic inter-
Berry MV, Gibbs DF. 1970. The interpretation of optical ferometry. J Opt Soc Am 59:1496–1498.
projection. Proc R Soc Lond Ser A314:143–152. Schwan HP. 1957. Electrical properties of tissue and cell suspen-
Bracewell RN. 1956. Strip integration in radio astronomy. Aust J sions. Adv Biol Med Phys 5:147–209.
Phys 9:198–217. Webster JG, editor. 1989. Electrical impedance tomography.
Bracewell RN, Riddle C. 1967. Inversion of fan-beam scans in Madison, Wisconsin: The Adam Hilger Series on Biome-
radio astronomy. Astrophys J 150:427–434. dical Engineering. p 156. (See also the Supplement devoted
Brown BH, Seagar AD. 1985. Applied potential tomography: to EIT: 1996 Physiol Meas 17 Suppl 4A: 1–234.)
data collection problems. Proceeding IEEE International Yorkey TJ, Webster JG, Tompkins WJ. 1986. An optimal
Conference on Electric Magnetic Fields in Medicine and impedance tomographic reconstruction algorithm. Proc Int
Biology p 79–82. Conf IEEE Eng Med Biol Soc 8:339–342.