International Communications in Heat and Mass Transfer 75 (2016) 218222

Contents lists available at ScienceDirect

International Communications in Heat and Mass Transfer

journal homepage: www.elsevier.com/locate/ichmt

Identication of a time-dependent bio-heat blood perfusion coecient

Jakub Krzysztof Grabski a , Daniel Lesnic b, * , B. Tomas Johansson c
a
Institute of Applied Mechanics, Poznan University of Technology, Jana Pawa II 24, 60-965 Pozna
n, Poland
b
Department of Applied Mathematics, University of Leeds, Leeds LS2 9JT, UK
c
School of Mathematics, Aston University, Birmingham, UK

A R T I C L E I N F O A B S T R A C T

Available online 30 December 2015 In the paper the identication of the time-dependent blood perfusion coecient is formulated as an inverse
problem. The bio-heat conduction problem is transformed into the classical heat conduction problem. Then
Keywords: the transformed inverse problem is solved using the method of fundamental solutions together with the
Bioheat equation Tikhonov regularization. Some numerical results are presented in order to demonstrate the accuracy and
Blood perfusion coecient the stability of the proposed meshless numerical algorithm.
Method of fundamental solutions 2015 Published by Elsevier Ltd.
Tikhonov regularization

1. Introduction
The mathematical description of the relation between the tissue
temperature and the arterial blood perfusion has received a quite
extensive attention over the years in the literature. Several differ-
ent models for heat transfer in living tissues have been proposed,
e.g. Pennes bioheat transfer model [1], which is the most commonly
used, Wulffs continuum model [2], Klingers continuum model [3],
the continuum model of Holmes and Chen [4] and the WeinbaumJiji
bioheat model [5].
A number of standard numerical methods have been used in
order to nd the heat transfer in living tissues, e.g. the nite element
method [6], the nite difference method [7], the boundary element
method (BEM) [8], the dual reciprocity BEM [9], the Trefftz nite
element method [10] or the Monte Carlo method [11]. Recently,
some papers investigated the application of meshless methods for
solving the bio-heat equation, e.g. Youse in [12] used the Bernstein
Galerkin method for nding the time-dependent blood perfusion
coecient from internal tissue temperature measurements, Cao et
al. in [13] used the method of fundamental solutions (MFS) in com-
bination with radial basis functions for analysing thermal behavior
of skin tissues, whilst the radial basis collocation method was used
by Jamil and Ng in [14] to predict the temperature inside biological
tissues.
In the present paper, the identication of the time-dependent
blood perfusion coecient in bioheat conduction is formulated as an
inverse problem, which is then solved numerically using the MFS.
In order to ensure stability of the solution for noisy data Tikhonov
regularization is employed.
2. Mathematical formulation
The bioheat equation proposed by Pennes [1] can be written as:
T
kt DT wb Cb (T Ta ) + S = qt ct , (1)
t
where kt is the thermal conductivity of the tissue, T is the temper-
ature of the tissue, wb is the blood perfusion rate, Cb is the heat
capacity of blood, Ta is the temperature of the arterial blood, S is a
volumetric source related to heat generation due to metabolism and
heat deposition, qt and ct are, respectively, the density and specic
heat of the tissue, and t is the time.
In one-dimension, Eq. (1) takes the following dimensionless form:
u 2u
(x, t) = 2 (x, t) q (t) u (x, t) + fS , (x, t) (0, 1) (0, T] , (2)
t x
where

x t kt kt (T Ta ) S wb cb L2
Communicated by J. Rose and A. Briggs. x= , t= 2
, u= , fS = , q= .
* Corresponding author.
L L qt ct S0 L2 S0 kt
E-mail address: D.Lesnic@leeds.ac.uk (D. Lesnic). (3)

http://dx.doi.org/10.1016/j.icheatmasstransfer.2015.12.028
0735-1933/ 2015 Published by Elsevier Ltd.
 J. Grabski et al. / International Communications in Heat and Mass Transfer 75 (2016) 218222 219

Neglecting, for simplicity, the dimensionless heat source fS we by Cannon et al. [17]. Reduction of the Inverse Problem (4), (5),
consider the inverse problem of nding the temperature u(x, t) and (7), (8) and (10) to a well-posed Volterra integral equation of the
the time-dependent blood perfusion coecient q(t) satisfying the second kind has been proved in [18, 19] and with multi-dimensional
bio-heat conduction equation: extensions given in [20].
In this study, we start with the theoretical analysis for solving the
u 2u alternative Inverse Problem (4), (5), (7)(9).
(x, t) = 2 (x, t) q (t) u (x, t) , (x, t) (0, 1) (0, T] (4)
We rst observe that employing the transformations
t x
subject to the initial condition  t 
r (t) = exp q (t) dt , v (x, t) = r (t) u (x, t) (16)
0
u (x, 0) = u0 (x) , x [0, 1] , (5)
recasts the Problem (4), (5), (7)(9) into the form
the Robin boundary condition at x = 1, namely
v 2v
u (x, t) = 2 (x, t) , (x, t) (0, 1) (0, T] , (17)
au (1, t) + b (1, t) = g (t) , t (0, T] , (6) t x
x
v (x, 0) = u0 (x) , x [0, 1] , (18)
the Dirichlet boundary condition at x = 0, namely

u (0, t) = f (t) , t (0, T] , (7) v (0, t) = r (t) f (t) , v (1, t) = r (t) g0 (t) , t (0, T] , (19)

and the Neumann boundary condition at x = 0, namely v

(0, t) = r (t) h (t) , t (0, T] . (20)
x
u
(0, t) = h (t) , t (0, T] . (8)
x Imposing also the rst three continuity conditions in Eq. (11)
leads to
In Eq. (6), we typically choose a = 1, b = 0, i.e. the Dirichlet
boundary condition
r (0) = 1 . (21)
u (1, t) = g0 (t) , t (0, T] , (9)
Thus using Eq. (16), the bio-heat Eq. (4) transforms into the heat
or a = 0, b = 1, i.e. the Neumann boundary condition Eq. (17). Applying the Green formula to the Problem (17)-(20), a
system is obtained consisting of two integral equations
u  
(1, t) = g1 (t) , t (0, T] . (10) 1 t
G
x r (t) f (t) = r (t) G (0, t; 0, t) h (t) + (0, t; 0, t) f (t)
2 0 n
We also assume the continuity compatibility conditions   t
G
(0, t; 1, t) g0 (t) dt + r (t) g1 (t) G (0, t; 1, t) dt
n 0
 1
u0 (0) = f (0) , u0 (0) = h (0) , u0 (1) = g0 (0) , u0 (1) = g1 (1) .
+ G (0, t; y, 0) u0 (y) dy, t (0, T] , (22)
(11) 0

Instead of the heat ux in Condition (8), we can have the temper-

ature measurement at an interior point X0 (0, 1) giving the data   t
1 G
r (t) g0 (t) = r (t) G (1, t; 0, t) h (t) + (1, t; 0, t) f (t)
2 0 n
u (X 0 , t ) = 0 (t ) , t (0, T] (12)   t
G
(1, t; 1, t) g0 (t) dt + r (t) g1 (t) G (1, t; 1, t) dt
n 0
satisfying the continuity condition  1
+ G (1, t; y, 0) u0 (y) dy, t (0, T] , (23)
0 (0) = u 0 (X 0 ) . (13) 0

where G is the fundamental solution of the one-dimensional time-

Another alternative to Eq. (8) is to prescribe the total mass
dependent heat equation
measurement

 1 H (t t ) (x n )2
u (x, t) dx = E (t) , t (0, T] (14) G (x, t; n, t) =  exp , (24)
0 4p (t t) 4 (t t )

satisfying the continuity condition with H the Heaviside function.

 Using Eq. (24), Eqs. (22) and (23) can be rewritten as a linear
1
E (0) = u0 (x) dx. (15) system of Volterra integral equations of the second kind, namely,
0
   
r (t ) t
g 0 (t ) 1
Under certain additional conditions, the uniqueness of the Inverse pr(t) f (t) = h (t) +g1 (t) + exp dt
0 tt 2 (t t) 4 (tt)
Problems (4), (5), (7), (9) and (12) or (14) has been proved by Prilepko  1 
and Solovev [15] and Lin [16], respectively. Morever, both these 1 y2
+ u0 (y) exp dy, t (0, T] , (25)
inverse problems, also in multi-dimensions, have been investigated t 0 4t
220 J. Grabski et al. / International Communications in Heat and Mass Transfer 75 (2016) 218222

    
r (t)t
f (t ) 1
pr (t) g0 (t) = h (t) +g1 (t) + exp dt
0 tt 2 4 (tt)
 t 
1 (1 y)2
+ u0 (y) exp dy, t (0, T] .
t 0 4t
(26)

Given the input data h, g0 , u0 , and f, the unknowns are r(t), r

C1 [0, T], r(0) = 1, r (t) > 0, and the product r(t)g1 (t).
Once r(t) and g1 (t) have been uniquely determined, the tempera-
ture is explicitly obtained from

 t  
1 G
u (x, t) = r (t) G (x, t; 0, t) h (t) + (x, t; 0, t) f (t) dt
r (t ) 0 n
 t  
G
+ r (t) G (x, t; 1, t) g1 (t) (x, t; 1, t) g0 (t) dt
0 n
 
1
+ G (x, t; y, 0) u0 (y) dy , (x, t) (0, 1) (0, T] . (27)
0

Also, from Eq. (16) formally, we can write

r  (t )
q (t ) = , t [0, T] . (28)
r (t )

Prior to this study, Trucu et al. [21] applied the BEM for solving
the system of boundary integral Eqs. (25) and (26). In this study, we
apply a different, but related meshless MFS for solving the inverse Fig. 1. Distribution of source and collocation points.
Problem (17)-(20) for nding the solution pair (v(x, t), r(t)).
where rk = r(tk ) are unknown quantities for k = 1, M1 .
3. The method of fundamental solutions (MFS) We can then order and rewrite the system of Eqs. (31) (33) of
(3M1 + N + 1) linear equations with (4M + M1 ) unknowns (C, r),
The MFS for the parabolic heat Eq. (17) that we shall use was where r = (rk )k=1,M , in the explicit form:
1
developed by Johansson and Lesnic in [22] based on results of
Kupradze [23]. In this method, we seek an approximation for the

M
(0) 
M
(1)
solution of the Heat Eq. (17) as a linear combination of fundamental cm G (0, tk ; y0 , tm ) + cm G (0, tk ; y1 , tm )
solutions m=M+1 m=M+1

rk f (tk ) = 0, k = 1, M1 , (34)

M
(0) 
M
(1)
vM (x, t) = cm G (x, t; y0 , tm ) + cm G (x, t; y1 , tm ) ,
m=M+1 m=M+1

M 
M
(x, t) [0, 1] [0, T] , (29) (0)
cm G (1, tk ; y0 , tm ) +
(1)
cm G (1, tk ; y1 , tm )
m=M+1 m=M+1
where y0 = d, y1 = 1 + d, and rk g0 (tk ) = 0, k = 1, M1 , (35)

(2m 1) T
tm = , m = M + 1, . . . , M. (30)
2M

M
(0) G 
M
(1) G
cm (0, tk ; y0 , tm ) + cm (0, tk ; y1 , tm )
Here d > 0 is an MFS parameter to be preassigned representing the m=M+1
x m=M+1
x
distance between the source points and the boundary; the distribu-
rk h (tk ) = 0, k = 1, M1 , (36)
tion of the source points and the collocation points is presented in
Fig. 1.
In Eq.  (29), the 4M unknown real coecients C =
(0) (1)
cm , c m are to be determined by collocating the Initial 
M
(0) 
M
(1)
m=M+1,M cm G (xl , 0; y0 , tm ) + cm G (xl , 0; y1 , tm )
Condition (18) and the Boundary Conditions (19) and (20). We m=M+1 m=M+1
choose xl = l/N for l = 0, N, and tk = kT/M1 for k = 1, M1 , to be = u0 (xl ) , l = 0, N. (37)
the collocation points. Applying Eqs. (18)(20) at these collocation
points result in A necessary condition for a unique solution is

vM (xl , 0) = u0 (xl ) , l = 0, N, (31) 3M1 + N + 1 4M + M1 or, 2M1 + N + 1 4M. (38)

vM (0, tk ) = rk f (tk ) , vM (1, tk ) = rk g0 (tk ) , k = 1, M1 , (32) The system of linear Eqs. (34)(37) be written more compactly in
the matrix form as
vM
(0, tk ) = rk h (tk ) , k = 1, M1 , (33) AX = b (39)
x
 J. Grabski et al. / International Communications in Heat and Mass Transfer 75 (2016) 218222 221

with the obvious notation. Since the MFS matrix is ill- Eq. (47) we employ the Tikhonov regularization method, as described
conditioned [24], we apply the zeroth-order Tikhonov regularization in Eq. (40), to yield the regularized solution
method for solving the system of Eq. (39). This yields the stable
 1
solution [25], r = Ktr K + kDtr D Ktr R, (48)
 1
X = Atr A + KI Atr b, (40) where k 0 is a regularization parameter to be prescribed according
to some criterion, e.g. the L-curve. Here, compared with Eq. (40), we
where I is the identity matrix of order 4M+M1 . In Eq. (40), K 0 rep- use a regularizing matrix D possibly different from I, to capture the
resents the regularization parameter which can be selected using the smoothness of r for example for rst-order regularization,
L-curve criterion, see e.g. [26]. Once the vector X = (C, r) has been

determined, the interior temperature is explicitly obtained, using 1 1 0 0 . . . 0
Eqs. (16) and (29), as 0 1 1 0 . . . 0
D=
....................... (49)

1 
M 
M 0 0 ... 0 1 1
cm G(x, tk ; y1 , tm ) ,
(0) (1)
u(x, tk ) = cm G(x, tk ; y0 , tm ) +
rk
m=M+1 m=M+1 of dimension (M1 1) M1 .
x (0, 1) , k = 1, M1 . (41) Once r is found, the perfusion coecient is obtained as

Also from Eq. (28), we obtain r  (t k )

q ( tk ) = , k = 1, M1 , (50)
r (t k )
r (tk ) r (tk1 )
q ( tk ) = , k = 1, M1 (42) with r given by Eq. (43) using the above approximation of the
(tk tk1 ) r (tk )
integral.

using forward nite-differences to approximate the derivative r (tk ).
However, in order to obtain a stable solution for q, one can mollify the 4. Numerical results and discussion
noisy data r. An alternative to the mollication method is to recast
the ill-posed problem of nding the rst-order derivative of the noisy For simplicity, we take T = 1. The parameters used in the pro-
function r as a Fredholm integral equation of the rst-kind, see e.g. posed method of solution are: M = M1 = 90, N = 2M1 1,
Trucu et al. [27]. Specically, the equation to be solved is then d = 0.4. The parameters K = 10 5 and k = 10 3 of the Tikhonov
regularizations Eqs. (40) and (48) were chosen using the L-curve
 T criterion.
r (t ) = 1 + K (t, t) r (t) dt, t [0, T] , (43) In the numerical simulation we consider a typical example with
0
the Input Data (5), (7)(9) given by
where the kernel is given by
u (x, 0) = u0 (x) = 3(x 2)2 , x [0, 1] (51)

0, 0 t < t T, 
K (t, t) =
1, 0 t t T. t3
u (0, t) = f (t) = 6t exp t t2 , t (0, 1] (52)
3

We have r(t0 ) = r(0) = 1 and t3
u (1, t) = g0 (t) = (3 + 6t) exp t t2 , t (0, 1] , (53)
 3
T
r k = r (t k ) = 1 + K (tk , t) r (t) dt, k = 1, M1 . (44)
0 
u t3
Dene fj = T
+ jT
for j = 0, M1 , giving tj = (fj1 + fj )/2 for
(0, t) = h(t) = 6(x2) exp tt2 (1+qg (t)) , t (0, 1] ,
2M1 M1 x 3
j = 1, M1 . Then we approximate the integral in Eq. (44) as (54)

M1
 
  fj
where in the heat ux data we added q percentage of noise with g(t)
r  tj K (tk , t) dt = rk 1, k = 1, M1 . (45)
fj1 as a random variable in [1, 1]. It can easily be checked that the ana-
j=1
lytical solution of the Inverse Problem (4), (51)-(54) (with q = 0) is

    given by
Dening R = (rk 1)k=1,M
, r = r tj j=1,M and
1 1

 t3
 fj
t fj1 u (x, t) = 3(x2)2 +6t exp t t2 , (x, t) (0, 1) (0, 1] ,
0, 3
Kkj = K (tk , t) dt = t fj1 , fj1 < t < fj (46)

(55)
fj1 f f , fj t
j j1

q (t) = 1 + 2t + t2 , t (0, 1] . (56)

for k, j = 1, M1 , Expression (45) recasts as the M1 M1 linear system
of equations
Kr = R.
However, this system is ill-conditioned since the differentiation of
a noisy function is an ill-posed problem. Consequently, to stabilize
The numerically reconstructed blood perfusion coecient for
various percentages of noise q {1%, 5%, 10%} is presented in Fig. 2
(47) in comparison with the exact solution from Eq. (56). From this
gure it can be seen that stable numerical solutions are obtained
which become increasingly more accurate as the percentage of noise
decreases.
222 J. Grabski et al. / International Communications in Heat and Mass Transfer 75 (2016) 218222
Fig. 2. The reconstructed blood perfusion coecient q(t).
5. Conclusions
In this paper, the inverse problem of determining the time-
dependent blood perfusion coecient by means of the method of
fundamental solutions (MFS) combined with the Tikhonov regular-
ization was proposed. It was illustrated that this combination pro-
duces accurate and stable numerical reconstructions. Although the
description has mainly been explained in the one-dimensional set-
ting the MFS has also potential and is feasible in higher dimensions,
see [28].
Acknowledgments
The work of the rst author was supported by the project
Engineer of the Future. Improving the didactic potential of the Poznan
University of Technology POKL.04.03.00-00-259/12, implemented
within the Human Capital Operational Programme, co-nanced by
the European Union within the European Social Fund.
References
[1] H.H. Pennes, Analysis of tissue and arterial blood temperatures in the resting
human forearm, J. Appl. Physiol. 1 (1948) 93122.
[2] W. Wulff, The energy conservation equation for living tissue, IEEE Trans.
Biomed. Eng. 21 (1974) 494495.
[3] H. Klinger, Heat transfer in perfused biological tissue-I: general theory, Bull.
Math. Biol. 36 (1974) 403415.
[4] K.R. Holmes, M.M. Chen, Microvascular contributions in tissue heat transfer,
Ann. N. Y. Acad. Sci. 335 (1980) 137150.
[5] S. Weinbaum, L.M. Jiji, A new simplied bioheat equation for the effect of blood
ow on local average tissue temperature, J. Biomech. Eng. Trans. ASME 107
(1985) 131139.
[6] Y. He, H. Liu, R. Himeno, H. Sunaga, N. Kakusho, H. Yokota, Finite element anal-
ysis of blood ow and heat transfer in an image-based human nger, Comput.
Biol. Med. 38 (2008) 555562.
[7] J.J. Zhao, J. Zhang, N. Kang, F. Yang, A two level nite difference scheme for
one dimensional Pennes bioheat equation, Appl. Math. Comput. 171 (2005)
320331.
[8] C.L. Chan, Boundary element method analysis for the bioheat transfer equation,
J. Biomech. Eng. 31 (1992) 494500.
[9] W.Q. Lu, J. Liu, Y. Zeng, Simulation of the thermal wave propagation in bio-
logical tissues by the dual reciprocity boundary element method, Eng. Anal.
Boundary Elem. 22 (1998) 167174.
[10] J. Jirousek, Q.H. Qin, Application of hybrid-Trefftz element approach to tran-
sient heat conduction analysis, Comput. Struct. 58 (1996) 195201.
[11] Z.S. Deng, J. Liu, Mathematical modeling of temperature mapping over skin sur-
face and its implementation in thermal disease diagnostics, Comput. Biol. Med.
34 (2004) 495521.
[12] S.A. Youse, Finding a control parameter in a one-dimensional parabolic
inverse problem by using the Bernstein Galerkin method, Inverse Prob. Sci. Eng.
17 (2009) 821828.
[13] L. Cao, Q.H. Qin, N. Zhao, An RBF-MFS model for analysing thermal behaviour of
skin tissues, Int. J. Heat Mass Transf. 53 (2010) 12981307.
[14] M. Jamil, E.Y.K. Ng, Evaluation of meshless radial basis collocation method
(RBCM) for heterogeneous conduction and simulation of temperature inside
the biological tissues, Int. J. Therm. Sci. 68 (2013) 4252.
[15] A.I. Prilepko, V.V. Solovev, Solvability of the inverse boundary-value problem
of nding a coecient of a lower-order derivative in a parabolic equation, Diff.
Equat. 23 (1987) 101107.
[16] Y. Lin, An iverse problem for a class of quasi-linear parabolic equations, SIAM J.
Math. Anal. 22 (1991) 146156.
[17] J.R. Cannon, Y. Lin, S. Wang, Determination of source parameter in parabolic
equations, Meccanica 27 (1992) 8594.
[18] D. Lesnic, Identication of the time-dependent perfusion coecient in bio-heat
conduction equation, J. Inverse and Ill-Posed Prob. 17 (2009) 112.
[19] D. Lesnic, M. Ivanchov, Determination of the time-dependent perfusion coe-
cient in the bio-heat equation, Appl. Math. Lett. 39 (2015) 96100.
[20] S.G. Pyatkov, Solvability of some inverse problems for parabolic equations, J.
Inverse and Ill-Posed Prob. 12 (2004) 397412.
[21] D. Trucu, D.B. Ingham, D. Lesnic, Inverse time-dependent perfusion coecient
identication, J. Phys. Conf. Ser. 124 (2008) 012050 (28 pages).
[22] B.T. Johansson, D. Lesnic, A method of fundamental solutions for transient heat
conduction, Eng. Anal. Boundary Elem. 32 (2008) 697703.
[23] V.D. Kupradze, A method for the approximate solution of limiting problems in
mathematical physics, USSR Comput. Math. Math. Phys. 4 (1964) 199205.
[24] P.A. Ramachandran, Method of fundamental solutions: singular value decom-
position analysis, Commun. Numer. Methods Eng. 18 (2002) 789801.
[25] D.L. Phillips, A technique for the numerical solution of certain integral
equations of the rst kind, J. Assoc. Comput. Mach. 9 (1962) 8497.
[26] P.C. Hansen, Analysis of discrete ill-posed problems by means of the L-curve,
SIAM Rev. 34 (1992) 561580.
[27] D. Trucu, D.B. Ingham, D. Lesnic, Reconstruction of the space- and time-
dependent blood perfusion coecient in bio-heat transfer, Heat Transfer Eng.
32 (2011) 800810.
[28] B.T. Johansson, D. Lesnic, T. Reeve, A method of fundamental solutions for
two-dimensional heat conduction, Int. J. Comput. Math. 88 (2011) 16971713.