IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 57, NO.

5, MAY 2010 1227

Stable Reconstruction of Piecewise Continuous Plane

Stratified Biological Tissues via Electrical
Impedance Tomography
Madlena Dolgin* and Pinchas D. Einziger

Abstract—Image reconstruction in electrical impedance tomog- biofilms thickness monitoring [5], and excitation with wireless
raphy is, generally, an ill-posed nonlinear inverse problem. Reg- (induced-current) electrodes [6], [7].
ularization methods are widely used to ensure a stable solution. Solution of a full inverse problem, i.e., reconstruction of the
Herein, we present a case study, which uses a novel electrical
impedance tomography method for reconstruction of layered bio- conductivity distribution inside the object, is usually solved via
logical tissues with piecewise continuous plane-stratified profiles. mean least square (MLS) technique [2]. For this purpose, the
The algorithm implements the recently proposed reconstruction forward electromagnetic problem should be solved in 3-D con-
scheme for piecewise constant conductivity profiles, utilizing Leg- figuration for a prescribed conductivity distribution. The MLS
endre expansion in conjunction with improved Prony method. It procedures are based on comparison of the data obtained from
is shown that the proposed algorithm is capable of successfully re-
constructing piecewise continuous conductivity profiles with mod- the forward solution with the potential measured on the bound-
erate slop. This reconstruction procedure, which calculates both aries of the body. There are some pitfalls in these procedures
the locations and the conductivities, repetitively provides inhomo- that scientists try to deal with [3]. First, the analytical solution
geneous depth discretization, i.e., the depths grid is not equispaced. for arbitrary 3-D configurations is hard to achieve and there is a
Incorporation of this specific inhomogeneous grid in the widely need to use numerical solvers repetitively to solve the forward
used mean least square reconstruction procedure results in a sta-
ble and accurate reconstruction, whereas, the commonly selected problem. Second, the unknowns in a general EIT problem are the
equispaced depth grid leads to unstable reconstruction. This obser- layers conductivities and locations. Since the problem is highly
vation establishes the main result of our investigation, highlighting nonlinear in both the conductivity values and their locations, it is
the impact of physical phenomenon (the image series expansion) more desirable to assume a specified spatial grid, i.e., to make the
on electrical impedance tomography, leading to a physically moti- locations known and search only for the conductivities vector. To
vated stabilization of the inverse problem, i.e., an inhomogeneous
depth discretization renders an inherent regularization of the mean this end, the reduced inverse schemes utilize standard numeri-
least square algorithm. The effectiveness and the significance of in- cal solvers (e.g., finite element and boundary element methods)
homogeneous discretization in electrical impedance tomography in conjunction with commonly used MLS procedure [8], [9].
reconstruction procedure is further demonstrated and verified via Unfortunately, this reduced procedure is highly sensitive to the
numerical simulations. selection of both the location grid and the initial conductivity
Index Terms—Biological tissues, depth discretization, EIT, values [10]. In this paper, we focus on the selection of the op-
inverse procedure, plane stratified, reconstruction algorithm, timal depth grid as well as the initial conductivities for plane
regularization, stability. stratified media, utilizing our recently proposed reconstruction
scheme [11], [12]. This scheme focuses on an EIT method uti-
I. INTRODUCTION lizing a quasistatic image series expansion scheme for layered
media [13], [14] in conjunction with the Wentzel–Kramers–
N ELECTRICAL impedance tomography (EIT), the tissue
I conductivity of the interior of a biological object is recon-
structed via quasi-static measurements of electric currents and
Brillouin (WKB) approximation [15]. The WKB image series
expansion is characterized by an outstanding feature, namely,
each image term corresponds analytically and explicitly to a
voltages captured on its boundaries. This relatively new biomed-
unique layer, leading to a one to one mapping between each
ical imaging modality has recently become of increased scien-
image term and the local impedance of the associated layer.
tific and public interest [1]–[3] and has been successfully ap-
This locality feature leads to an effective linear reconstruction
plied in applications such as breast and skin cancer detection [4],
of the electrical impedance profile via Legendre expansion (LE)
Manuscript received August 12, 2009; revised October 18, 2009. First pub- in conjunction with Prony’s method (PM), thus, allowing an ef-
lished February 5, 2010; current version published April 21, 2010. Asterisk ficient reconstruction of locally stratified media (relative to the
indicates corresponding author. Legendre window width) in applications such as breast and skin
*M. Dolgin is with the Engineering School, Kinneret College on the Sea
of Galilee, Tsemach, D.N. Emek HaYarden 15132, Israel (e-mail: madlena@ cancer detection [4] as well as biofilms thickness monitoring [5].
tx.technion.ac.il). It is shown that the choice of quasi-geometrical inhomogeneous
P. D. Einziger is with the Department of Electrical Engineering, grid leads to accurate and stable inverse scheme, whereas the
Technion, Israel Institute of Technology, Haifa 32000, Israel (e-mail:
[email protected]). selection of the widely used equispaced grid results in poor re-
Color versions of one or more of the figures in this paper are available online construction. This crucially important observation is due to our
at http://ieeexplore.ieee.org. physically motivated (the image series expansion) EIT inverse
Digital Object Identifier 10.1109/TBME.2009.2038168
procedure, leading, via PM, to inhomogeneous depth grid, and

0018-9294/$26.00 © 2010 IEEE

1228 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 57, NO. 5, MAY 2010

TABLE I
TWO INDEPENDENT SOLUTIONS OF (6), ri (z, ξ) AND ti (z, ξ), CORRESPONDING
TO SPECIFIED iTH LAYER, FOR TWO DIFFERENT CONDUCTIVITY PROFILES

plane z = z1 = 0; and 3) an air-body contact (namely σ1 = 0),

the surface solution potential V (ρ) for piecewise continuous
Fig. 1. Layered medium configuration and reconstruction setup.
media (3) and (4) can be expressed via inverse Bessel transform
[12]

thus, a subsequent stable reconstruction. In contrast, the con- V (ρ) = Φ(r)|z =z  =0,σ 1 =0
veniently selected conventional equispaced grid necessitates, in
∞
1
general, mathematical compensation, i.e., regularization tech- = ξ[φ(z, ξ)|z =z  =0,σ 1 =0 ]J0 (ξρ)dξ (5)
niques (e.g., [16], [17]) or extensive a priori knowledge on the 2π 0
reconstructed conductivity distribution [18].
where φ(z, ξ), obtained upon substitution of (2)–(5) in (1), is a
solution of the second-order ordinary differential equation
II. PROBLEM DEFINITION
The electromagentic fields, induced by a point-source, are I
φ (z, ξ) + φ (z, ξ)[ln(σ(z)] − ξ 2 φ(z, ξ) = − δ(z − z  ) .
governed by Maxwell’s equations. At low frequencies and small σ(z)
field strengths the electromagnetic properties of living tissue (6)
allow these to be simplified via the potential equation (e.g., [19], Since v(ξ) = φ(z, ξ)|z =z  =0,σ 1 =0 is the Bessel transform of
also [11], [12] for ∇ ln(σ) = 0) V (ρ) in (5)

∞
i(r)
∇2 Φ(r) + ∇Φ(r)∇ln(σ) = − (1) v(ξ) = 2π ρV (ρ)J0 (ξρ)dρ = φ(z, ξ)|z =z  =0,σ 1 =0
σ 0
where Φ(r) and i(r) denote the electric potential and the cur- I R2 (ξ)r2 (0, ξ) + t2 (0, ξ)
rent source distribution, respectively, at the point r = (ρ, θ, z). =− , (7)
σ2 (0) t2 (0, ξ) + R2 (ξ)r2 (0, ξ)
Considering a piecewise continuous plane stratified conductiv-
ity profile σ(z) and a quasi-static current point-source S where the global reflection coefficients Ri (ξ) are expressed re-
 cursively in (8), as shown at the bottom of the page.
i(r) = Iδ(r − r ) (2)
The two independent solutions of (6), ti (z, ξ) and ri (z, ξ),
 
located at r = (0, 0, z ), we aim to reconstruct σ(z) for z ≥ z1 corresponding to a specified ith layer, denote the so-called
using a noninvasive measurements of the potential Φ(r) at the transmitted and reflected functions, respectively. Both func-
plane z1 = 0, as outlined in Fig. 1. tions ti (z, ξ) and ri (z, ξ) corresponding to piecewise linear and
piecewise constant conductivity profiles, utilized in subsequent
III. FORWARD PROBLEM ANALYTIC SOLUTION reconstruction calculations, are given explicitly in Table I. Note
Piecewise continuous media can be specified by that φ (z, ξ), [ln(σ(z)] , t (z, ξ), and r (z, ξ) denote derivatives
of φ(z, ξ), [ln(σ(z)], t(z, ξ) and r(z, ξ) with respect to z, in
σ(z) = σi (z) (3) (6)–(8).
where z belongs to the ith layer
IV. RECONSTRUCTION PROCEDURE
zi−1 < z < zi , i = 1, 2, . . . , n + 1, z0 = −∞, zn +1 = ∞ .
(4) An efficient reconstruction procedure can be executed by
Assuming: 1) the current point-source is located at the origin, approximating the piecewise continuous profile by a piecewise
i.e., z  = z1 = 0; 2) the measurements are performed on the constant profile, as depicted in Fig. 2.

Ri (ξ)
Ri+1 (ξ) + (ti+1 (zi , ξ)σi (zi)ti (zi , ξ) − ti (zi , ξ)σi+1 (zi)ti+1 (zi , ξ))/(ri+1 (zi , ξ)σi (zi)ti (zi , ξ) − ti (zi , ξ)σi+1 (zi)ri+1

(zi , ξ))
=
1+ Ri+1 (ξ)(ri+1 (zi , ξ)σi (zi)ri (zi , ξ)−ri (zi , ξ)σi+1 (zi)ri+1 (zi , ξ))/(ti+1 (zi , ξ)σi (zi)ri (zi , ξ)−ri (zi , ξ)σi+1 (zi)ti+1 (zi , ξ))
  

 
ri+1 (zi , ξ)σi (zi)ti (zi , ξ) − ti (zi , ξ)σi+1 (zi )ri+1

(zi , ξ)
× − , Rn +1 (ξ) = 0 (8)
ti+1 (zi , ξ)σi (zi )ri (zi , ξ) − ri (zi , ξ)σi+1 (zi )ti+1 (zi , ξ)
DOLGIN AND EINZIGER: STABLE RECONSTRUCTION OF PIECEWISE CONTINUOUS PLANE 1229

(3), namely the jump discontinuity

 
 σi (zi ) − σi+1 (zi ) 
 
 σi (zi ) + σi+1 (zi )   1 (14)

and the gross rate

|{ln [σ(z)]} | (zi − zi−1 )  1 . (15)

It should be noted that while the approach adopted herein for

layered media [13], [14] leads to an explicit closed form rep-
resentation of Green’s function in terms of the image series
and subsequently its WKB approximation, the potentially gen-
eralized approach for 3-D configurations [20]–[22], utilizing
Fig. 2. Discretization of a piecewise continuous conductivity profile.
Green’s integral identities, is limited to configuration for which
Green’s function can be evaluated explicitly.
Mathematically, we denote the approximated piecewise con-
stant conductivity by σp B. Linear Transformation—Legendre Expansion
σ(z) = σp , z p−1 < z < z p (9) We use shifted Legendre polynomial Ll (1 − 2ρ2 ) [23], de-
fined on a unit window width 0 ≤ ρ ≤ w, w = 1 [unit length],
corresponding to the pth layer via
to expand the scaled potential V (sρ) as
p = 1, 2, . . . , m + 1, z0 = −∞, zm +1 = ∞ . (10)
∞

It is readily noted that while σi (z) in (3) and (4) corresponds V (sρ) = bl (s)Ll (1 − 2ρ2 ) (16)
to the actual piecewise continuous media, σp in (9) and (10) l=0
corresponds to its discretization via piecewise constant media,
where [24]
as shown in Fig. 2.
A reconstruction procedure can now be performed by uti-
w

lizing the previously introduced LE in conjunction with PM, bl (s) = 2(2l + 1) ρV (sρ)Ln (1 − 2ρ2 )dρ
0
which is most effective for WKB piecewise constant conductiv-
sw 
ity profiles [11], [12]. This two-step reconstruction procedure, 2(2l + 1)
= ρV (ρ)Ln 1 − 2(ρ/s)2 dρ
incorporating WKB image series expansion, is compared to s2 0
the alternative, widely used, MLS method. These schemes are 2I 
n  −(2l+1)
briefly reviewed for the sake of clarity and completeness. = Ki 2zi /s + 1 + (2zi /s)2 (17)
πsσ2 i=1

A. Forward Problem—WKB Image Series Expansion and s is a positive real scaling factor. Note that, s can also
Recently, it has been shown [11]–[14] that the WKB solu- be regarded, via the second equality in (17), as scaling factor
tion of (1) in a piecewise constant plane stratified medium [11, for Legendre’s unit window width w = 1 (with respect to a
Fig. 1] is given via the following image series expansion: nonscaled potential V (ρ)). Furthermore, while for i > 1 , Ki =
Ki , for i = 1, K1 = 1/2, and the unknown parameter is σ2 .
VW KB (ρ) = ΦW KB (r)|z =z  =0,σ 1 =0
 
I 1 
m
Kp C. Spectral Domain Estimate—Extended Prony Method
= + (11)
πσ2 2ρ p=2 [ρ2 + (2z p )2 ]1/2 The individual n terms contained in the right-hand side
summation of (17) may be regarded as n distinct expo-
where the intrinsic reflection coefficients Kp , p = 2, 3, . . . , m,
nentials (2(zi /s) + 1 + (2zi /s)2 )−2 of power l, weighted
given by
by 2IKi /πsσ2 ((2zi /s) + 1 + (2zi /s)2 )−1 . Hence, The PM
σp − σp+1 procedure [25] exploiting 2n data points, i.e., bl (s), 0 ≤ l ≤
Kp = , K0 = 0, σ0 = σ1 (12)
σp + σp+1 2n − 1 or, equivalently, 2n inner products, can be utilized to re-
construct n distinct exponentials and weights characterizing the
are assumed to be sufficiently small, i.e., for WKB
stratification geometry and electrical parameters, respectively.
approximation
Also, to avoid dominance of any one of the exponentials con-
|Kp |  1 . (13) tained in the right-hand side of (17) the scaling factor s should
be large enough to support the entire reconstruction range, i.e.,
Note that, in view of the discretization scheme shown in Fig. 2,
σp is given via (9). Furthermore, the WKB constraint (13) im- 2zi
s≥ (18)
poses two WKB constraints on the actual conductivity profile w
1230 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 57, NO. 5, MAY 2010

Fig. 3. Reconstruction of oscillating conductivity profile [z2 /w = Fig. 4. Reconstruction of monotonically decreasing conductivity profile
1, z3 /w = 2, b2 = 10 S/m, b3 /b2 = 0.6, l2 = 5 m, and l3 = −3 m, (29)] [z2 /w = 0.5, z3 /w = 2, b2 = 10 S/m, b3 /b2 = 0.95, l2 = 5 m, and l3 =
via the three methods: PM (w = 1, s = 4), PMMLS [α = 2.2, z 2 /w = 0.2, 9.5 m, (29)] via the three methods: PM(w = 1, s = 4), PMMLS [α = 1.75,
(28)], and ESMLS [∆z/w = 0.51, (27)]. z 2 /w = 0.15 (28)], and ESMLS [∆z/w = 0.54, (27)].

leading to the inclusion of V (ρ) well within the window 0 ≤ D. Alternative Reconstruction Procedure—MLS Method
ρ/w ≤ s. Note, however, that the selection s  2zn /w, results
in almost linearly dependant exponentials and, thus, should be Upon setting prescribed values for z p , the MLS error function
avoided. ε, given via
Following the PM procedure, the n unknown exponentials L −1
((2zi /s) + 1 + (2zi /s)2 )−2 are roots of the n-order polyno- 1 
ε= (VW KB (ρl ) − V (ρl ))2 (24)
mial equation L
l=0


n is minimized by differentiation with respect to the normalized
cp (µ2 )p = 0 (19) intrinsic reflection coefficients Kq , as follows:
p=0
∂ε I IKq
= 0, K1 = , Kq = , q = 2, 3, . . . , m . (25)
where cn = 1. Let µ2i denote the ith root of (19), then ∂Kq 2πσ2 πσ2
 −1
 2 Note that VW KB (ρ) and V (ρ) are given in (11) and (5), respec-
2zi 2zi
µi =  + 1+  (20) tively. The linear optimization problem, explicitly expressed as
s s m
−1 


L
Kp ρ ρ
− V (ρl ) = 0,
leading to p=1
2
ρl + (2z p )2 2
ρl + (2z q )2
l=0

µ2i − 1 q = 1, 2, 3, . . . , m (26)
zi = s. (21)
4µi
is solved by a matrix inversion. Subsequently, σp are recon-
The polynomial coefficients denoted by the vector C = structed from (12). Two z p sets are selected for the MLS proce-
[c0 , c1 , . . . , cn −1 ]T are solution of the linear system dure (26): 1) the standard equispaced set, referred as ESMLS,
given by
AC = B , C = A−1 B (22)
z p = (p − 1)∆z, p = 1, 2, . . . , m (27)
where
and 2) a nonhomogenous set, motivated via PM reconstruction
Aj,k = bj +k −1 , j, k = 0, 1, . . . , n − 1, results and referred as PMMLS

B = −[bn , bn +1 , . . . , b2n −1 ]T z p+2 − z p+1 = α(z p+1 − z p ), p = 1, 2, . . . , m − 2 (28)

and T denotes transpose operation. Finally, the where α is a numerically measured constant.
weight coefficients, denoted by the vector K = 2I/
πsσ2 [µ1 K1 , µ2 K2 , . . . , µn Kn )]T , are also solution of
the linear system
V. RECONSTRUCTION RESULTS
MK = D , K = M−1 D (23)
The simulations performed herein utilize three distinct
where Mi,j = (µ2j )i and D = [b0 , b1 , . . . , bn −1 ]T . four layered, n + 1 = 4 [(3)–(4)], continuous piecewise linear
DOLGIN AND EINZIGER: STABLE RECONSTRUCTION OF PIECEWISE CONTINUOUS PLANE 1231

Fig. 5. Reconstruction of monotonically decreasing non-WKB conductivity Fig. 7. Noisy measurements reconstruction (SNR|ρ = s w = 60 dB, [12]) of
profile [z2 /w = 0.5, z3 /w = 2, b2 = 10 S/m, b3 /b2 = 0.8, l2 = 10/6 m, monotonically decreasing conductivity profile [z2 /w = 0.5, z3 /w = 2, b2 =
and l3 = 4 m, (29)] via the three methods: PM (w = 1, s = 4), PMMLS 10 S/m, b3 /b2 = 0.95, l2 = 5 m, and l3 = 9.5 m, (29)] via the three methods:
[α = 1.76, z 2 /w = 0.12, (28)], and ESMLS [∆z/w = 0.63, (27)]. PM (w = 1, s = 4), PMMLS [α = 1.75, z 2 /w = 0.15, (28)], and ESMLS
[∆z/w = 0.54, (27)].

Fig. 6. Noisy measurements reconstruction (SNR|ρ = s w = 60 dB, [12]) of

oscillating conductivity profile [z2 /w = 1, z3 /w = 2, b2 = 10 S/m, b3 /b2 =
0.6, l2 = 5 m, and l3 = −3 m, (29)] via the three methods: PM (w = 1, s = 4),
PMMLS [α = 2.2, z 2 /w = 0.2, (28)], and ESMLS [∆z/w = 0.51, (27)].
Fig. 8. Noisy measurements reconstruction (SNR|ρ = s w = 60 dB, [12])
of monotonically decreasing non-WKB conductivity profile [z2 /w =
conductivity profiles, given via 0.5, z3 /w = 2, b2 = 10 S/m, b3 /b2 = 0.8, l2 = 10/6 m, and l3 = 4 m, (29)]
 via the three methods: PM (w = 1, s = 4), PMMLS (α = 1.76, z 2 /w = 0.12,
 b2 (1 − z/l2 ), z1 ≤ z ≤ z2 , i = 2 (28)], and ESMLS [∆z/w = 0.63, (27)].
σi (z) = b3 (1 − z/l3 ), z2 ≤ z ≤ z3 , i = 3 . (29)

b3 (1 − z3 /l3 ), z3 ≤ z, i=4

Figs. 3–8 and Table II readily demonstrate the potential

promise and the successful implementation of the PM method on the relative error (30) here, than in [11] and [12], since the
for reconstruction of continuous conductivity profiles, signifi- noise associated with the discretization of the continuous profile
cant extension of the previous reconstruction results for piece is dominant (see Table II). Finally, the discretization noise can
wise constant layered media [11], [12]. Furthermore, profound be further reduced in PMMLS by increasing the layers number
analysis of the depths relations, provided by the PM scheme, in (28).
reveals that there is an almost constant ratio between the thick- Reconstruction results for three distinct conductivity profiles
ness of the layers [difference between the subsequent depths utilizing PM, PMMLS, and ESMLS methods, are shown in
(28)]. Incorporating this observation in PMMLS, indeed leads Figs. 3–8. Relative errors associated with the pth layer recon-
to stable and accurate reconstructions, whereas the selection of struction, defined via
a standard equispaced z p set (27), results in poor and unstable  
zp
reconstructions. It is noted that both PM and PMMLS methods z p −1 |σp − σ(z)| dz
Error = 100  zp  [%] (30)
perform very well even for non-WKB conductivity profile (see σ(z)dz
z p −1
Figs. 5 and 8, respectively) and noisy measurements (see Figs.
6–8 and Table II). The measurements noise have a lesser effect are specified in Table II, including noisy data reconstructions.
1232 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 57, NO. 5, MAY 2010

TABLE II
RECONSTRUCTION RESULTS FOR THREE DISTINCT CONDUCTIVITY PROFILES UTILIZING PM, PMMLS, AND ESMLS METHODS

VI. CONCLUSION [5] P. Linderholm, J. Vannod, Y. Barrandon, and P. Renaud, “Bipolar re-
sistivity profiling of 3D tissue culture,” Biosens. Bioelectron., vol. 22,
The EIT algorithms outlined herein for piecewise continuous pp. 789–796, 2007.
layers perform accurately for shallow reconstruction, even for [6] S. Levy, D. Adam, and Y. Bresler, “Electromagnetic impedance tomogra-
phy (EMIT): A new method for impedance imaging,” IEEE Trans. Med.
noisy data. This feature has already been established recently for Imag., vol. 21, no. 6, pp. 676–687, Jun. 2002.
piecewise constant layered media [11], [12]. Thus, our approach [7] S. Zlochiver, M. Rosenfeld, and S. Abboud, “Induced-current electrical
is effective for applications such as breast and skin cancer detec- impedance tomography: A 2-D theoretical simulation,” IEEE Trans. Med.
Imag., vol. 22, no. 12, pp. 1550–1560, Dec. 2003.
tion [4] as well as biofilms thickness monitoring [5]. It should be [8] J. S. Kallman and J. G. Berryman, “Weigthed least-squares criteria for
noted, however, that only inhomogeneous depth discretization, electrical impedance tomography,” IEEE Trans. Med. Imag., vol. 11,
dictated by PM, leads to stable and accurate results, including no. 2, pp. 284–292, Jun. 1992.
[9] M. Soleimani, C. E. Powell, and N. Polydorides, “Improving the for-
the MLS procedures, and thus rendering regularization tech- ward solver for the complete electrode model in EIT using algebraic
niques unnecessary. Furthermore, the reconstructed media has multigrid,” IEEE Trans. Med. Imag., vol. 24, no. 5, pp. 577–583, May
to be only locally layered, i.e., it can slowly vary laterally as 2005.
[10] L. Borcea, “Nonlinear multigrid for electrical impedance tomography,” in
compared with Legendre finite-extent window width. In view Proc. SPIE—Int. Soc. Opt. Eng., vol. 5016, no. 1, 2003, pp. 49–58.
of these observations, applying our physically motivated inho- [11] M. Dolgin and P. D. Einziger, “Image series expansion for electri-
mogeneous coordinate discretization in other EIT algorithms cal impedance tomography,” Phys. Rev. Lett., vol. 93, no. 14, pp.
148101-1–148101-4, Oct. 2004.
may result in an inherent stabilization of these reconstruction [12] M. Dolgin and P. D. Einziger, “Reconstruction of layered biological tissues
algorithms. via electrical impedance tomography,” IEEE Trans. Biomed. Eng., vol. 53,
Finally, the impulse response in (5)–(8) and (11)–(15), cor- no. 12, pp. 2464–2471, Dec. 2006.
[13] P. D. Einziger, L. M. Livshitz, and J. Mizrahi, “Rigorous image-series
responding to the exact potential and its WKB approximation, expansions of quasi-static Green’s functions for regions with planar strati-
respectively, is valid as long as the electrodes size (e.g., the di- fication,” IEEE Trans. Antennas Propag., vol. 50, no. 12, pp. 1813–1823,
ameter of electrodes S and P in Fig. 1) is much smaller than Dec. 2002.
[14] P. D. Einziger, L. M. Livshitz, and J. Mizrahi, “Image theory and appli-
both the layers thickness [(3) and (4)] and [(9) and (10)] and the cations in bioelectromagnetics,” in Biomathematics: Modeling and Stim-
window width w [(16) and (17)]. While the effect of electrodes ulation, J. C. Misra, Ed. Singapore: World Scientific, 2005, pp. 315–359.
size is neglected throughout this paper, it can be accounted [15] P. M. Morse and H. Feshbach, Methods of Theoretical Physics. New
York: McGraw-Hill, 1953.
for [26] and [27], particulary, in conjunction with the image [16] M. Goharian, M. Bruwer, A. Jegatheesan, G. R. Moran, and J. F. Mac-
series expansion [28]. Gregor, “A novel approach for EIT regularization via spatial and spectral
principal component analysis,” Physiol. Meas., vol. 28, no. 9, pp. 1001–
REFERENCES 1016, Sep. 2007.
[17] V. Kolehmainen, M. Vauhkonen, P. A. Karjalainen, and J. P. Kaipio, “Spa-
[1] D. S. Holder, Ed., Electrical Impedance Tomography, Methods, History tial inhomogeneity and regularization in EIT,” in Proc. 19th Annu. Int.
and Applications. Bristol, U.K.: IOP, 2005. Conf. IEEE Eng. Med. Biol. Soc., Chicago, IL, 30 Oct.–2 Nov.1997, vol.
[2] B. H. Brown, “Electrical impedance tomography (EIT): A review,” J. 1, pp. 449–452.
Med. Eng. Tech., vol. 27, no. 3, pp. 97–108, 2003. [18] M. Arad, S. Zlochiver, T. Davidson, Y. Shoenfeld, A. Adunsky, and S. Ab-
[3] W. R. Lionheart, “EIT reconstruction algorithms: Pitfalls, challenges and boud, “The detection of pleural effusion using a parametric EIT tech-
recent developments,” Physiol. Meas., vol. 1, no. 25, pp. 125–142, Feb. nique,” Physiol. Meas., vol. 30, no. 4, pp. 421–428, Apr. 2009.
2004. [19] Electromagnetic Fields and Energy, H. A. Haus and J. R. Melcher, Eds.
[4] M. Assenheimer, O. Laver-Moskovitz, D. Malonek, D. Manor, I. Na- Englewood Cliffs, NJ: Prentice-Hall, 1989.
haliel, R. Nitzan, and A. Saad, “The T-SCANT M technology: Electrical [20] J. Lehr, “A vector derivation useful in impedance plethysmographic field
impedance as a diagnostic tool for breast cancer detection,” Physiol. calculations,” IEEE Trans. Biomed. Eng., vol. 19, no. 6, pp. 156–157,
Meas., vol. 22, no. 1, pp. 1–8, Feb. 2001. Mar. 1972.
DOLGIN AND EINZIGER: STABLE RECONSTRUCTION OF PIECEWISE CONTINUOUS PLANE 1233

[21] D. B. Geselowitz, “An application of electrocardiagraphic lead theory to Madlena Dolgin received the B.Sc, M.Sc., and Ph.D. degrees in electrical
impedance plethysmography,” IEEE Trans. Biomed. Eng., vol. BME-18, engineering from the Technion, Israel Institute of Technology, Haifa, Israel, in
no. 1, pp. 38–41, Jan. 1971. 2000, 2002, and 2009, respectively.
[22] P. Metherall, D. C. Barber, R. H. Smallwood, and B. H. Brown, She is currently a Faculty Member at the Engineering School, Kinneret
“Three-dimensional electrical impedance tomography,” Nature, vol. 380, College on the Sea of Galilee, Tsemach, Israel. Her research interests includes
no. 6574, pp. 509–512, Apr. 1996. bioelectromagnetics and medical microwave imaging.
[23] M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions.
New York: Dover, 1970.
[24] A. Erdelyi, Tables of Integral Transforms vol. II. New York: McGraw-
Hill, 1954.
[25] R. W. Hamming, Numerical Methods for Scientists and Engineers. New
York: Dover, 1986.
[26] K. S. Cheng, D. Isaacson, J. C. Newell, and D. G. Gisser, “Electrode
models for electric current computed tomography,” IEEE Trans. Bio.
Eng., vol. 36, no. 9, pp. 918–924, Sep. 1989. Pinchas D. Einziger received the B.Sc. and M.Sc. degrees in electrical engi-
[27] B. S. Kim, G. Boverman, J. C. Newell, G. J. Saulnier, and D. Isaacson, neering from the Technion, Israel Institute of Technology, Haifa, Israel, and the
“The complete electrode model for EIT in a mammography geometry,” Ph.D. degree in electrophysics from the Polytechnic University, Brooklyn, NY,
Physiol. Meas., vol. 28, no. 7, pp. S57–S69, 2007. in 1976, 1978 and 1981, respectively.
[28] L. M. Livshitz, P. D. Einziger, and J. Mizrahi, “A model of finite electrodes Since 1981 he has been a Faculty Member in the Department of Electrical
in layered media: Hybrid image series and momemt method scheme,” Engineering, Technion, Israel Institute of Technology. His current research in-
ACES J. , Spec. Issue Comput. Bioelectromagn., vol. 16, no. 2, pp. 145– terests include electromagnetic wave theory, nonlinear wave phenomena, and
154, 2001. bioelectromagnetics.