You are on page 1of 13

25

An improved three-dimensional direct numerical


modelling and thermal analysis of a female breast
with tumour

E Y K Ng* and N M Sudharsan


School of Mechanical and Production Engineering, Nanyang Technological University, Singapore

Abstract: It is well known that malignant tumour tissue generally has higher metabolic and blood
perfusion rates than most normal tissues. The authors aim to show that the tissue temperature prole
within the breast and the surface temperature prole can be quantied to develop an expert system
or diagnostic tool for breast cancer detection. The surface temperature and tissue temperature proles
are analysed for a three-dimensional numerical model of a normal breast and a breast with a tumour.
Tumours of diVerent sizes are placed at various locations. In the model, the tissue temperature prole
is distorted at the tumour location and was found to compare well with in vivo tests. It was also
found that as the tumour was moved to deeper locations its eVect on surface temperature was lower.
It was observed that small tumours in deeper regions do not have a signicant isolated impact on
the surface. The numerical results could also capture a shift in the position of the tumour. For
tumours greater than 10 mm in the supercial regions and of signicant size in deeper regions, it
could be seen that the surface temperature distribution of the breast is directly related to the position
and size of the tumour embedded in it. The feasibility of providing a diagnostic tool in conjunction
with numerical modelling and high-resolution thermograms is also discussed.

Keywords: thermal analysis, tumour, breast cancer diagnostic tool

NOTATION á zenith angle measured from Y (deg )


õ non-dimensional temperature
$ Laplacian operator ô time required for tumour to double its
c specic heat capacity (J/ kg ßC ) volume (days)
C constant ( W day/m3)
D tumour diameter (m)
h combined overall heat transfer coeYcient Subscripts
(convection, radiation and evaporation)
a artery
( W/m2 ßC )
b blood
k thermal conductivity ( W/m ßC )
N e environment
scaling factor
m metabolic
q volumetric heat generation ( W/m3)
R radius, tumour location (m)
T temperature ( ßC )
w blood perfusion rate (kg /s m3 ) 1 INTRODUCTION
x, y, z Cartesian coordinates (computational
domain) (m) An aVordable but sensitive detection tool for breast
X, Y, Z Cartesian coordinates (physical domain) cancer is the need of the hour in view of the ever increas-
(m) ing number of breast cancer cases. Thermography is an
aVordable tool where an abnormality in the surface tem-
The MS was received on 23 December 1999 and was accepted after perature could indicate an underlying tumour. Gautherie
revision for publication on 7 July 2000. [1] observed a higher temperature and blood ow in
* Corresponding author: School of Mechanical and Production
Engineering, Nanyang Technological University, 50 Nanyang Avenue, cancerous tissue. The main drawback of this equipment
Singapore 639798. is that the sensitivity increased with the size of tumour
Downloaded from pih.sagepub.com at University of Birmingham on August 23, 2015
H07399 © IMechE 2001 Proc Instn Mech Engrs Vol 215 Part H
26 E Y K NG AND N M SUHDARSAN

Fig. 1 Cross-sectional view of the model of the breast (physical domain)

and false positive results were very high. However several 0.8 s would yield the result that the diameter of the
investigators [2–4] have found that patients with false detectable lesion should not be one-fourth its depth.
positive thermograms are at a higher risk of developing In the authors’ earlier two-dimensional work [13],
cancer in the long run, requiring long-term follow-up. priority was given to studying the best setting for con-
Siu et al. [5] incorporated the patient’s characteristics ducting thermography. The breast is far from two-
with age and family history in a multivariant risk func- dimensional and requires analysis in three dimensions.
tion, based on the discriminant function analysis of Two-dimensional analysis exploits breast symmetry and
Tasuoka [6 ] and concluded that for women aged 50 and this has the inherent limitation of placing the tumour in
younger, thermography aided with physical examination the Y axis (Fig. 1 ). In reality, the tumour could be of
has a sensitivity close to that of a mammogram. Usuki any size and located anywhere. Hence a full three-
et al. [7] and Cockburn [8] both conducted a cold stress dimensional analysis is essential to understand fully the
test, where vasoconstriction is forced by cooling the eVect of the tumour on the surface temperature. An
breast either by an alcohol mist or by soaking the arms analysis of variance also needs to be performed to com-
and feet in ice-cold water. A dynamic or time-dependent pare this analysis with the results obtained in the earlier
temperature scan was performed in which the presence two-dimensional study [13]. For the purpose of inter-
of tumour could be pin-pointed more accurately, largely preting the thermogram, an ideal solution would be to
eliminating false positive results. simulate the presence of a tumour in a breast, by merely
Chen et al. [9] reduced the bioheat equation to one observing the surface temperature distribution. This is,
dimension and showed that the blood perfusion term in fact, an inverse heat transfer problem. Moreover, it
can be extracted from time-dependent thermograms and, would require time-dependent analysis with the incor-
on the other hand, that the metabolic heating term poration of vasomotor behaviour in the model. In order
cannot be determined using a similar procedure. Osman to see if this ideal state can be reached, the scope of this
and Afy [10, 11] performed a steady state direct study is as follows:
numerical simulation of a three-dimensional breast
1. To determine whether the presence of a tumour can
model with an extremely coarse grid system and con-
be numerically simulated in three dimensions and if
cluded that the surface temperature may not be directly
the surface temperature is related to the underlying
related to an underlying tumour. The present authors
tumour.
believe that numerical simulation may provide an
2. To perform an analysis of variance to see the sensi-
additional tool to combine with the existing high-
tivity of the model.
resolution thermogram in order to improve sensitivity.
3. To analyse the eVect of tumours of diVerent sizes at
However, there is rst a need to address the concern of
diVerent locations on surface temperature.
Chen et al. [9] and Osman and Afy [11].
4. To discuss the feasibility of conducting a time-
In previous work the present authors developed a two-
dependent analysis and the possibility of reverse
dimensional breast model [12] and found that the surface
engineering.
temperature was aVected by an underlying tumour. The
model was studied using analysis of variance ( ANOVA)
in order to arrive at the best parametric optimization for
tumour identication [13]. Chen et al. [9] observed that 2 BREAST CANCER STATISTICS
for a camera resolution of 0.1 ßC and a scan time of 4 s,
the diameter of the detectable lesion should not be less Breast cancer is the number one killer disease among
than one-third its depth. However, present-day thermo- women in Singapore. As many as 800 cases are detected
grams have an accuracy of 0.03 ßC and a scan time of each year with an increasing rate of 3.3 per cent per
Downloaded from pih.sagepub.com at University of Birmingham on August 23, 2015
Proc Instn Mech Engrs Vol 215 Part H H07399 © IMechE 2001
AN IMPROVED THREE-DIMENSIONAL DIRECT NUMERICAL MODELLING AND THERMAL ANALYSIS 27

Fig. 2 Number of breast cancer deaths per 100 000 women world wide

annum. Figure 2 presents the number of breast cancer breast cancer. However, mammograms have their draw-
deaths per 100 000 women world-wide. Around 180 000 backs; the mammogram sensitivity is higher for older
breast cancers are diagnosed annually in the United women (age group 60–69 years) at 85 per cent compared
States. The number of women aVected by breast cancer with younger women (<50 years) at 64 per cent [14].
for every 100 000 is 39 in Singapore, 88 (among whites) Moreover, the mammogram is painful and the harm due
in the United States, 21.2 in China, 32.2 in Hong Kong to the X-ray is equivalent to smoking three cigarettes
and 19.9 in India. Out of every 100 000 women, 15.1 die [16 ].
of breast cancer in Singapore compared with 11.1 in The Singapore Breast Cancer Screening Project
Hong Kong and 33 in the United States [14]. In Europe, Working Committee conducted a randomized trial of
out of 100 000 women, 29.3 die of breast cancer in mammography screening for the prevalence and detec-
England and Wales, 28.3 in Denmark, 27.5 in Scotland, tion of breast carcinoma. In their published work (Ng
27 in Ireland, 25 in Switzerland, 23.2 in Germany, 19.6 et al. [17]) it was found that the cumulative percentage
in France and 17.4 in Sweden [15]. Family history plays of women with carcinomas of size 11–15 mm and less
a signicant role; the risk factor is four times higher for was around 52 per cent; 62 per cent of women had carci-
a woman whose mother and sister have breast cancer. nomas of size 16–20 mm and less and 95 per cent of
Other risk factors are age (greater than 35 years), early women screened had carcinomas of size 50 mm and less.
menarche (under 12 years), late menopause (after 50 It was found that the cumulative percentage of women
years), childless or rst child after 30 years and a diet with carcinomas of 50 mm or less in the ‘not invited for
high in animal fat [14]. screening’ group was 74 per cent. This proves that sys-
Cancer grows from a single cell invading locally and tematic screening helps in identifying more cancers and
may spread to all parts of the body in advanced stages. eventually in reducing mortality rates. Cady et al. [18]
The growth of the tumour mass up to a diameter of less drew similar conclusions from studies conducted in the
than 20 mm is classied as the preclinical stage and the United States, that the mean diameter of cancer detected
clinical phase starts from 20 mm and above. The various was reduced over a period of years (20–21 mm diameter
stages are as follows: during the period 1989–93, down from 30 mm during
the period 1984–89). They have also presented the lack
Stage 0 Non-invasive
of eVectiveness of mammographic screening in reducing
Stage I <20 mm
the mortality for younger age groups, i.e. women 40–49
Stage II Spreads to armpit
years old. This may be attributed to the dense breast
Stage III >50 mm and locally invasive (broken
tissues, which reduces the sensitivity of the mammogram.
through skin)
Stage IV Spreads to other parts of the body
If detected at Stage 0, the prognosis (outcome) is as 3 BIOHEAT EQUATION
good as for a normal patient in a ten-year period. If
cancer is detected early (tumour size <10 mm) the The governing diVerential equation presented by Pennes
patient has an 85 per cent chance of cure as opposed to [19] is used in this model. Equation ( 1) can describe a
10 per cent if the cancer is detected late. The mammo- quantitative relation in a human tissue and includes the
gram is presently the standard tool for detection of eVect of blood ow in tissue temperature on a continuum
Downloaded from pih.sagepub.com at University of Birmingham on August 23, 2015
H07399 © IMechE 2001 Proc Instn Mech Engrs Vol 215 Part H
28 E Y K NG AND N M SUHDARSAN

basis. The equation includes the heat transfer by conduc- mesh structure. The choice was then to increase the mesh
tion through the tissue, the volumetric metabolic heat concentration on the boundary or to modify the compu-
generation of the tissue and the volumetric blood per- tational domain in a suitable manner. The boundary
fusion rate, whose strength was considered to be pro- layer of the breast, i.e. the subcutaneous fat layer with
portional to the arterial/venous temperature diVerence. overlying skin as shown in Fig. 1, is expanded by a factor
The temperature of the artery is approximated as the of 10. Figure 3 depicts the cross-sectional view of the
core temperature of the body and that of the vein computational domain with an expanded boundary. The
approximated as the local tissue temperature: gure also shows a central tumour located along the
Y axis whose centre is 49.5 mm from the base (inset).
k$2T Õc w (T ÕT ) +q = 0 (1)
b b a m The regions are meshed in such a way that the nodal
The boundary conditions involved are Õk$T = connectivity is maintained and the information is not
h(T ÕT ) on the surface and T =T in the core distorted during computation. It was found that the
e a
(thoracic wall ). above ‘skin treatment’ could capture the gradient and
Although the limitations of the Pennes equation are greatly minimized the numerical instability. As it can be
well known, it was found from the justications pre- seen that the space coordinates in the computational
sented in reference [12] that the Pennes equation did domain are diVerent from the physical domain, the equa-
provide reasonable agreement with in vivo studies. tions need to be modied to preserve the original form.
Hence, the traditional bioheat equation presented by The equation for the computational domain is manipu-
Pennes [19] is adopted. lated by incorporating the transformation function for
the space coordinates. In this case the transformation
function is
4 MATHEMATICAL MODELLING
x=NX, y=NY, z=NZ
A three-dimensional model was developed which was a After the simulation, the results from the computational
close representation of the actual shape of the breast, domain are transformed back to the physical domain for
with various tissue layers of unequal thickness having analysis. The above manipulation is detailed in
diVerent thermo-physical properties [20]. This model has Sudharsan and Ng [13].
the exibility that allows the various thermo-physical The model was discretized using four-noded tetra-
parameters of the tissue regime to be changed and hence hedrons and a mesh less-sensitive approximation was
permits better control over the simulation. However, obtained. The advantage of this model over others is
modelling the individual layers had some computational that it is a closer representation of the actual breast and
drawbacks. Although the grid topology was carefully mesh discretization is done for each layer as opposed to
chosen to provide reasonable accuracy (Ng and a multicomponent element (1380 tetrahedrons [10]).
Sudharsan [21]), the cross-sectional temperature prole Therefore, the properties and parameters such as the
had a reasonable distribution but the surface tempera- blood perfusion term and tumour metabolic heat can be
ture had numerical noise. The skin conductivity was very varied on an individual basis for the purposes of simu-
low and this caused a large step gradient across the lation. The number of elements in this model is over
boundary. The gradient was not captured by a regular 37 000 and satises the mesh-independent solution.

Fig. 3 Cross-sectional view of the computational domain for the three-dimensional model of the breast with
expanded boundary (inset, tumour at glandular region)
Downloaded from pih.sagepub.com at University of Birmingham on August 23, 2015
Proc Instn Mech Engrs Vol 215 Part H H07399 © IMechE 2001
AN IMPROVED THREE-DIMENSIONAL DIRECT NUMERICAL MODELLING AND THERMAL ANALYSIS 29

5 SOLUTION METHOD Therefore a tumour of diameter 10 mm will yield ô= 50


days from equation (3) and q = 65.46103 W/m3 from
m
The model was developed using FEMAP [22] and the equation (2). Similarly for a 15 and 30 mm tumour, the
mesh generated was interfaced with a PDE (partial tumour doubling time will be 240 and 565 days respect-
diVerential equation) calculator, FASTFLO [23]. The ively and the metabolic rate will be 13.66103 and
governing equation and boundary conditions were pro- 5.796103 W/m3 respectively. In this paper the tumour
grammed using FASTTALK macrolanguage of sizes considered are 10, 12.5, 15, 17.5, 20 and 30 mm.
FASTFLO. The solution obtained is transformed back The appropriate metabolic rates were calculated and
to the physical domain by a user-written program in C. simulated.
The results are then plotted with TECPLOT [24] post- The next important parameter is the choice of the
processing software. It is to be noted that the FASTFLO blood perfusion term. Gautherie [1] conducted in vivo
software is not a standard nite element method ( FEM ) studies on breast tumours and has estimated that the
package and it has to be programmed by the user. mean eVective thermal conductivity, which includes the
Algorithms and manipulation of the equation and enhancement in conductivity due to blood perfusion, is
boundary conditions must be written by the user. The 0.511 W/m ßC. With an assumption that the thermal con-
FEM model was benchmarked with known analytical ductivity of the tumour tissue matches that of the
solutions to arrive at the optimum nodal distance [21]. glandular tissue at 0.48W/m ßC, the enhancement in con-
The simulation was carried out on an SGI O2 machine ductivity is 0.03 W/m ßC. From Priebe [27] it is seen that
and the solution time was 0.05 s per mesh point. The a variation in blood ow of 150 ml/100 g per min equates
total time for solution was 6 min. to 0.05 W/m ßC. Therefore in this study it would yield a
mean blood ow of 90 ml/100 g per min. Referring to
the work of Vaupel [28] it was observed that the varia-
6 CHOICE OF PARAMETERS bility in tumour blood ow in the breast would be from
8 to 80 ml/100 g per min. However, here a blood ow of
The values of thermal conductivity and metabolic rates 60 ml/100 g per min is assumed to verify whether the
for the various layers are taken from Werner and Buse numerical model can capture the eVect of a tumour on
[25] and are summarized in Table 1. It is noted that these the surface for a owrate that is lower than the mean
values are for generic tissues. However, the tumour perfusion source term.
blood ow and metabolic heat are obtained from the The boundary conditions for this study are: on the
in vivo studies of Gautherie [1] and Gautherie et al. [26 ]. skin surface, the heat transfer due to convection, radi-
The blood owrates to tissues are given in terms of milli- ation and evaporation; in the core of the body (i.e. the
litres of blood per 100 grams of tissue per minute;. thoracic wall ) the temperature is maintained at 37 ßC;
1 ml/100 g tissue per min is converted to 1 kg of blood and the artery temperature is at the core temperature.
per unit volume of tissue (m3 ) per second using the The subcutaneous tissue and the skin overlying it pro-
density of blood and tissues (1060 and 1080 kg /m3 gress continuously throughout the body. As the breast
respectively). This mass owrate is multiplied by the is simulated as an independent entity, the subcutaneous
specic heat capacity of blood ( 4200 J/ kg ßC ) to yield region at the breast base (a ring with inner and outer
0.86103 W/m3 ßC. From Gautherie et al. [26 ] the diameters of 134 and 144 mm respectively as indicated
tumour doubling time and the metabolic rate are related in Fig. 1) is set to be adiabatic. The breast is assumed
by a hyperbolic function. to be in supine condition to account for the validity of
the hemispherical model. There is no clothing on the
q ô= C ( W day/m3) (2) surface of the breast. As most of the thermographic
m
where C is a constant, 3.276106 W day/m3, and ô is the measurement is for the purpose of correlation in future
time required for the tumour to double its volume. The works and it is expected to take place in a controlled
tumour diameter is related to ô as follows: constant temperature environment of 21 ßC, the same
value is used as the environment temperature for the
D = exp[0.002134(ô Õ50)] 10Õ2 (m) (3) boundary condition.
At temperatures between 20 and 30 ßC the body loses
Table 1 Values of blood perfusion, thermal conductivity and heat by insensible evaporation. This requires the incor-
metabolic heat production [25, 26 ] poration of the mass equation due to loss of water from
c w k the body by phase change. However, in order to avoid
b b
Layers ( W/m3 ßC ) ( W/m ßC ) q ( W/m3) adding complexity to the equation, and as the thermo-
m
graphic measurements are to be taken under strict proto-
Subcutaneous fat 800 0.21 400
Gland 2400 0.48 700 cols, the heat loss due to evaporation may be taken as
Muscle 3600 0.48 700 negligible. Osman and Afy [10] adopted the heat trans-
Tumour 486103 0.48 13.66103 (15 mm) fer coeYcient for convection from Pennes [19] and evalu-
5.796103 (30 mm)
ated the combined heat transfer coeYcient due to the
Downloaded from pih.sagepub.com at University of Birmingham on August 23, 2015
H07399 © IMechE 2001 Proc Instn Mech Engrs Vol 215 Part H
30 E Y K NG AND N M SUHDARSAN

eVects of convection, radiation and evaporation, h, to as breast with oVset tumour. Two more models are
be 13.5 W/m2 ßC. This value is used in this study. developed for a breast with a 15 mm tumour located at
(X, Y: 30, 40) and (X, Y : 30, 47.5). The surface tempera-
ture pattern is analysed for these small shifts in position.
7 SIZE AND POSITION OF TUMOUR

On the eve of the Breast Cancer Awareness Month 8 NORMAL BREAST


(October 1998), free mammograms were conducted by
the Breast Cancer Foundation on 1000 women. The The four perfusion sets considered are:
average size of lump detected was 30 mm. From Ng et al.
Perfusion set 1, Normal perfusion as in Table 1
[17], it was found that 135 cancers were detected among
(c w )
the 28 231 women screened; 52 per cent of these women b b 1
Perfusion set 2, The perfusion to the subcutaneous
had invasive carcinomas of 11–15 mm or less. Six
(c w ) layer alone is increased to
tumour sizes in the range 10–30 mm are considered. b b 2
1.66103 W/m3 ßC
Taking a 15 mm tumour as the reference, it was
Perfusion set 3, The perfusion to the gland layer
decided to place the tumour at a location where its sur-
(c w ) alone is increased to
face is one diameter below the skin surface. As the total b b 3
3.66103 W/m3 ßC
height of the areola region is 72 mm from the base, the
Perfusion set 4, The perfusion to both gland and
centre of the tumour will be 49.5 mm from the base along
(c w ) subcutaneous region is increased by
the Y axis. The six models are generated with tumour b b 4
the above values
sizes of 10, 12.5, 15, 17.5, 20 and 30 mm at this location.
This model is designated as the breast with central The above increase is based on the range of values as
tumour. The tumour is now moved one radius (15 mm specied and tabulated in Werner and Buse [25].
reference) below the original location of 49.5 mm. These The change in perfusion has an eVect on the surface
six new sets of models with tumours located 42 mm from temperature and this eVect is analysed in terms of the
the base are designated as breasts with a deep central change in surface area for a certain temperature.
tumour. The 15 mm tumour at the location 49.5 mm Figure 4 presents the change in area under a particular
from the base is then assumed to move along a circular surface temperature (27.2 ßC ) on a normal breast for the
path. The new tumour position is R=49.5 mm and á= various perfusion terms. A contour labelled 27.2 means
22.5ß and a model is generated. This model is designated that the area within that region is at a temperature of

Fig. 4 Change in area under isotherm 27.2 ßC on a three-dimensional normal breast model for various blood
perfusion sets
Downloaded from pih.sagepub.com at University of Birmingham on August 23, 2015
Proc Instn Mech Engrs Vol 215 Part H H07399 © IMechE 2001
AN IMPROVED THREE-DIMENSIONAL DIRECT NUMERICAL MODELLING AND THERMAL ANALYSIS 31

less than or equal to 27.2 ßC. Considering the tempera- that the skin temperature results primarily from blood
ture distribution due to normal perfusion (c w ) perfusion from the tissues and the local metabolic rate
b b 1
(dashed line), it can be seen that the area enclosed by is a second-order eVect. The numerical prediction com-
temperature 27.2 ßC and less, is very large. On increasing pares well with the above study. The eVects of tumour
the blood perfusion to the supercial region (c w ) (rep- metabolic and tumour perfusion rates are the least sig-
b b 2
resented as a thick solid line) it can be observed that the nicant and with very low orders of magnitude. The
area with a temperature of 27.2 ßC or less has shrunk tumour metabolic rate acts as a heat source, causing the
considerably. In other words there is a larger region local tissue temperature to increase. The temperature of
where the breast temperature is greater than 27.2 ßC. the venous return blood is therefore higher, as the local
When the perfusion to the glandular region alone is tissue temperature is approximated to that of the venous
increased (c w ) (dotted line) it can be seen that the return blood in the Pennes equation. This hotter blood
b b 3
area at a temperature of 27.2 ßC or less has expanded, is convected to supercial regions causing a regional
however it is found to be slightly warmer than the eVect on the skin surface and was experimentally
normal one. When the blood perfusion to both the skin observed by Gautherie [30]. Gautherie [1] observed that
and glandular region (c w ) is increased the coldest tem- the venous drainage was hotter than its arterial supply
b b 4
perature, i.e. the tip of the areola, is at 27.2 ßC (thin line) in a cancerous tissue and therefore the numerical
and the rest of the region is warmer. One important prediction compares well with this observation.
observation is that perfusion to the supercial region
plays an important role in increasing the breast
temperature. 9 RESULTS AND DISCUSSION
Based on the 27-design ANOVA conducted in the two-
dimensional model [13], it was decided in the present Figure 5 represents the simulated surface temperature
paper to conduct a 25 design by assuming a 15 mm distribution of a normal breast and a breast with a cen-
tumour at a constant location (49.5 mm from base) in tral tumour (schematically represented as a grey circle).
the three-dimensional model. The range of parameters In this gure the contour levels are presented in a non-
used is summarized in Table 2. The parameters that have dimensional form multiplied 100 times for clarity. For
not been incorporated are thermal conductivity, tissue instance a contour level 4 means that the area under this
thickness and the volume of the breast. It is to be noted circle has a temperature less than or equal to 38.5. The
that the tissue thermal conductivity is very low and the diVerence between any two contour levels is 0.5/100
eVective thermal conductivity depends on the blood per- (non-dimensional temperature). In absolute terms, this
fusion term. Hence the blood perfusion term alone is deviation is dõ (T ÕT ) = 0.5(37 Õ21)/100=0.08 ßC.
a e
considered. With regard to tissue thickness and volume, The sensitivity of the digital infrared thermal camera
it was discussed in reference [13] that these are uncon- ( Varioscan 3011-ST, Jenoptik AG [31]) is Ô0.03 ßC.
trollable factors, as there is bound to be wide variability Shrinkage in this circle means a reduction in the cold
from one subject to another, and hence they are not area, or conversely an increase in the warmer area. From
considered. It is to be noted that the blood perfusion Fig. 5 it can be seen that the normal breast has a larger
source term depends on the volume of tissue and any colder area when compared with a breast with a central
analysis by varying the blood perfusion source term tumour. It can also be observed that contour level 4 (in
indirectly provides information regarding the change in the vicinity of the tumour) has a larger interval, com-
volume, i.e. a higher perfusion for a thin tissue would pared with a smaller interval for contour levels away
theoretically be equivalent to a lower perfusion for a from the tumour.
thicker tissue. Figure 6 represents the detailed view of contour
It was observed that the predominant factor is level 4. It can be seen that the dashed line representing
subcutaneous perfusion followed by the heat transfer a normal breast has the largest cold area. As tumour
coeYcient, whose eVect is negative. This means that a size increases, this cold region shrinks in a concentric
higher heat transfer coeYcient reduces the skin tempera- fashion. Consistent shrinkage in this area implies a
ture. The metabolic rate of the subcutaneous skin layer reduction in the cold region or conversely an increase in
is the next most important factor. Love [29] observed the warm area. It can be observed that the interval
between a normal breast and one with a 10 mm tumour
Table 2 High and low values for the various factors
is wider than that of a breast with a 10 and 12.5 mm
tumour. Similarly, the interval between a 20 mm tumour
Factor Description Low High Units and its neighbouring 17.5 mm tumour is smaller than
the interval between 20 and 30 mm tumours.
A Subcutaneous metabolic rate 800 1200 W/m3
B Tumour metabolic rate 700 65 400 W/m3 Next, the tumour is shifted to a deeper location 42 mm
C Subcutaneous blood perfusion 800 2400 W/m3 ßC from the base (deep central tumour). Logically a lesser
D Tumour blood perfusion 32 000 64 000 W/m3 ßC eVect of the tumour on the skin isotherm pattern would
E Heat transfer coeYcient 10.8 16.2 W/m2 ßC
be expected. The enlarged view of contour level 4 for a
Downloaded from pih.sagepub.com at University of Birmingham on August 23, 2015
H07399 © IMechE 2001 Proc Instn Mech Engrs Vol 215 Part H
32 E Y K NG AND N M SUHDARSAN

Fig. 5 Predicted surface temperature distribution for a normal breast and breasts with central tumours of
various sizes. Location, 49.5 mm from base; for key to tumour sizes see Fig. 6

Fig. 6 Enlarged view of contour level 4 (27.16 ßC ) for a normal breast and breasts with central tumours of
various sizes. Location, 49.5 mm from base
Downloaded from pih.sagepub.com at University of Birmingham on August 23, 2015
Proc Instn Mech Engrs Vol 215 Part H H07399 © IMechE 2001
AN IMPROVED THREE-DIMENSIONAL DIRECT NUMERICAL MODELLING AND THERMAL ANALYSIS 33

normal breast and one with deep central tumours of six were compared. The central tumour (0, 49.5, 0), the deep
sizes is presented in Fig. 7. The shrinkage of the cold central tumour (0, 42, 0) and tumours at locations
zone is self-explanatory. However, the shrinkage is less (30, 40, 0) and ( 30, 47.5, 0) were considered. Figure 9
pronounced than that observed for central tumours presents the detailed view of isotherm level 4. The deep
( Fig. 6). It may also be seen at this location that the central tumour is by far the coldest, encompassing a
eVect of a 10 mm tumour is less than that observed for large area (thin line). The tumour at location (30, 40, 0)
tumours nearer the surface. Moreover, the eVect of all is warmer than the deep central tumour ( long dashed
the tumours is lower and the isotherms cluster towards line). This is because it is closer to the skin than the deep
the normal one. Interestingly, the diameter to depth ratio central tumour. One other observation is that the iso-
for a 10 mm tumour located 42 mm from the base (i.e. therm is elliptic in prole with a larger shrinkage at the
30 mm from the areola the region) is 1 : 3. It would positive X axis (in the vicinity of the tumour) compared
appear that any increase in depth of the 10 mm tumour with the opposite side. The next in order is the central
would cause the loss of tumour signature. Comparing tumour 49.5 mm from the base (dotted line). It can
Figs 6 and 7, it can also be observed that any given be observed that the shrinkage is concentric with the
contour pattern can be caused by a large deep tumour deep central tumour. Finally, the tumour at location
or a small supercial tumour. Hence, for a given contour (30, 47.5, 0 ) shows the maximum reduction in cold zone
pattern several combinations of tumour size and depth area (solid line). It can be clearly seen that the shrinkage
exist. This observation is supported by the one- is more pronounced at the positive X axis in comparison
dimensional analysis performed by Chen et al. [9]. with the opposite side. These results indicate that the
Figures 8a to d show the surface mesh, surface tem- location of the tumour does aVect the surface tempera-
perature distribution and tissue temperature proles in ture distribution contrary to the observation by Osman
XY and XZ planes respectively for a breast with an oVset and Afy [11]. However, it should be pointed out that
tumour. An increased temperature at the tumour limitations exist in this observation. It is important to
location can be observed from Figs 8c and d. The tem- note that the relationship is valid for tumours of larger
perature corresponding to the tumour location is in the size and for smaller tumours close to the surface. It was
range 36–36.5 ßC. For the purpose of comparison the also found that the regional eVect of a hot spot was
in vivo tumour temperatures measured by Gautherie predominantly due to the convective heat transfer of
[32] were compared and found to be in the range blood. The higher metabolic activity causes the tumour
35.5–36.5 ßC. The numerical results compare well with tissue temperature to increase, causing the increase in
experimental observations in the literature. However, it the venous return temperature. This heat is convected to
is to be noted that it is diYcult to obtain the tumour the supercial regions and was also observed experimen-
temperature directly using thermography. tally by Gautherie [30]. Therefore it may be concluded
As a next step, 15 mm tumours at various locations that the vascular architecture has an important role to

Fig. 7 Enlarged view of contour level 4 (27.16 ßC ) for a normal breast and breasts with deep central tumours
of various sizes. Location, 42 mm from base
Downloaded from pih.sagepub.com at University of Birmingham on August 23, 2015
H07399 © IMechE 2001 Proc Instn Mech Engrs Vol 215 Part H
34 E Y K NG AND N M SUHDARSAN

Fig. 8 Various views of a breast with oVset tumour: (a) surface mesh, (b) surface temperature distribution
and tissue temperature proles in (c) XY and (d) XZ planes

Fig. 9 Comparison of the surface temperature distribution for a 15 mm tumour at various locations (contour
level 4, 27.16 ßC )
Downloaded from pih.sagepub.com at University of Birmingham on August 23, 2015
Proc Instn Mech Engrs Vol 215 Part H H07399 © IMechE 2001
AN IMPROVED THREE-DIMENSIONAL DIRECT NUMERICAL MODELLING AND THERMAL ANALYSIS 35

play. A hot spot may be seen at a location away from ation using ANOVA and the Taguchi method [13], it
the tumour site, due to the vascular architecture. was observed that lower skin perfusion provides ideal
However in general, vascular changes are elicited in high- conditions for identifying the presence of a tumour. This
resolution thermography [33] and would enable the observation is in line with the results obtained for the
clinician to make a qualitative assessment. full three-dimensional model.
From ANOVA results it was observed that the tem- The resolution of the FEM model depends on the
perature of the surface increased due to the blood con- number of elements and the aspect ratio of the tetra-
vection. This would mean that localized eVects would hedrons. However, a ner grid structure with a good
depend on vascular architecture. It must be pointed out aspect ratio is expected to increase the computation time
that localized warm areas in reality would not necessarily and would also require faster processors and a huge
indicate an underlying tumour, but fortunately in certain amount of memory. The limitation of this model is that
cases the venous supply route can also be visualized with the shape and tumour location would vary from individ-
the high-resolution thermogram and this would help in ual to individual. This would require a wide range of
making a qualitative judgement. numerical models. However, a single computational
Tumours are seldom spherical and to simulate an arbi- domain can represent many physical models by the
trary shape, a model was generated by fusing two spheri- correct choice of transformation functions.
cal volumes of 15 mm diameter. The fused volume was Now that a very exible model has been developed, it
located at (30, 40, 0) and the orientation was initially has to be acknowledged that this model cannot be used
assumed to be vertical. A second model was generated as a stand-alone tool to provide a unique solution giving
by rotating this volume by 45ß to investigate the eVect the size and location of the lesion. Thermography can
of tumour orientation. Figure 10 compares the tempera- be successful only when large amounts of thermographic
ture distribution in a normal breast and in a breast with data can be obtained to build in intelligence and perform
an arbitrarily shaped tumour. As expected the surface image/pattern recognition to conduct a multivariant
temperature is distorted due to the presence of the approach to improve its sensitivity. Numerical simu-
tumour. However, the shift in orientation caused very lation can be performed by changing a wide range of
little eVect on the surface temperature distribution. It parameters, to eliminate certain possibilities such as
was also observed that the variation between the two background noise, reducing false positive results. For
orientations was observed only when skin perfusion was example a hot spot in a thermogram may be due to a
at its minimum and the deviation was lost when skin tumour or to enhanced blood perfusion. By feeling the
perfusion was increased. Based on parametric optimiz- breast, some idea may be obtained about the presence /

Fig. 10 Surface temperature distribution for a normal breast and breasts containing a tumour of arbitrary
shape with diVerent orientations
Downloaded from pih.sagepub.com at University of Birmingham on August 23, 2015
H07399 © IMechE 2001 Proc Instn Mech Engrs Vol 215 Part H
36 E Y K NG AND N M SUHDARSAN

absence of a lump. Numerical simulation can help in ACKNOWLEDGEMENTS


generating a set of patterns by varying the parameters
to create a database as a supplement to experimental The authors would like to thank Dr F. C. Ng FRCR,
results. It is also necessary to know the length of time Senior Consultant Radiologist, Singapore General
the lump has existed, the size and history of the lump, Hospital and Dr E. H. Ng FRCS, Consultant
a history of surgical intervention and/or other treatment, Oncologist, Mount Elizabeth Medical Centre,
for example radiotherapy, as well as proper physical Singapore, for their valuable discussions and input to
examination before a conclusion can be made based this project. The authors would also like to thank
on the imaging modality. The simulation can be an Professor Ritchie Alastair for providing expert editing
additional input to the physician in interpreting the and proof-reading of the manuscript.
thermogram. The selected output from numerical simu-
lation may also be an input to diagnosis using articial
neural networks and measured thermograms.
REFERENCES

1 Gautherie, M. Thermopathology of breast cancer: measure-


ment and analysis of in vivo temperature and blood ow.
10 CONCLUDING REMARKS Ann. New York Acad. Sci., 1980, 335, 383–415.
2 Isard, J. H., Becker, W., Shilo, R. and Ostrum, B. J. Breast
The three-dimensional numerical model was simulated thermography after four years and 10,000 studies. Am.
J. Roentgenology, Radiat. Therapy Nuclear Med., 1972,
with tumours of various sizes and locations. The tumour
115(4), 811–821.
temperature correlated well with the in vivo studies of
3 Isard, J. H., Sweitzer, C. J. and Edelstien, G. R. Breast
Gautherie [32]. The surface temperature distribution thermography. A prognostic indicator for breast cancer
was found to change with the size and depth of the survival. Cancer, 1988, 62, 484–488.
tumour. Osman and Afy [11] observed that the surface 4 Gautherie, M. and Gros, C. M. Breast thermography and
temperature may not be related to the embedded tumour cancer risk prediction. Cancer, 1980, 45, 51–56.
beneath it. It was observed that a direct relationship 5 Siu, O. T., Ghent, W. R., Colwell, B. T. and Henderson,
existed between the two. It was also observed that, for J. C. Thermogram aided clinical examination of the
a given temperature contour, several combinations of breast—an alternative to mammography for women 50 or
tumour depth and size exist. It would therefore be younger. Can. J. Publ. Health, 1982, 73(4), 232–235.
6 Tasuoka, M. M. Multivariate Analysis, 1971 (John Wiley,
diYcult to predict the size and position of a tumour
New York).
just by a single steady state temperature observation.
7 Usuki, H., Teramoto, S., Komatsubara, S., Hirai, S. I.,
Although Chen et al. [9] have made some preliminary Misumi, T., Murakami, M., Onoda, Y., Kawashima, K.,
inroads in extracting the blood perfusion term from Kino, K., Yamashita, K. I. and Matsubara, J. Advantages
time-dependent analysis, the model did not incorporate of subtraction thermography in the diagnosis of breast dis-
the vasodilation and vasoconstriction, which might ease. Biomed. Thermology, 1991, 11(4), 286–291.
entirely alter the ow pattern, and hence the model is 8 Cockburn, W. Thermal image analysis: thermal challenge
far from a straightforward inverse heat transfer problem. test. 1999, Isabella Tracey Memorial Breast Health
A pilot study was conducted in Nanyang Technological Foundation, Whittier, California. <http://home.earthlink.
University and 34 volunteers participated. The volunteers net/~docbill/challenge.htm>.
9 Chen, M. M., Pederson, C. O. and Chato, J. C. On the
were aged 30–47 years. The numerical model correlated
feasibility of obtaining three dimensional information from
reasonably with experiments with a careful choice of par-
thermographic measurements. J. Biomech. Engng, 1977,
ameters, boundary conditions and scaling factor. Further 99(2), 58–64.
eld data collection in the Department of Diagnostic 10 Osman, M. M. and Afy, E. M. Thermal modelling of the
Radiology at Singapore General Hospital is in progress, normal woman’s breast. Trans. ASME, J. Biomech. Engng,
for quantitative validation. The possibility of combining 1984, 106, 123–130.
a numerical simulation and an articial intelligence 11 Osman, M. M. and Afy, E. M. Thermal modelling of the
approach adjunct to thermography will be analysed. A malignant woman’s breast. Trans. ASME, J. Biomech.
more exible model with diVerent layers and four quad- Engng, 1988, 110, 269–276.
rants is being developed to address the shape issues and 12 Sudharsan, N. M., Ng, E. Y.-K. and Teh, S. L. Surface
will be compared with eld data. A database will be gen- temperature distribution of a breast with/without tumor.
Int. J. Comput. Meth. Biomechanics Biomed. Engng, 1999,
erated by proper characterization using tissue thickness,
2(1), 187–199.
breast diameter and blood perfusion (roughly based on 13 Sudharsan, N. M. and Ng, E. Y.-K. Parametric optimisation
age and activity of breast). However, it is to be acknowl- for tumour identication: bioheat equation using ANOVA
edged that this characterization will be on a case-to-case and Taguchi method. Proc. Instn Mech. Engrs, Part H,
basis, as parameters are unique to individuals. This data- Journal of Engineering in Medicine, 2000, 214(H5),
base will be accessed for comparing experimental results. 505–512.
Downloaded from pih.sagepub.com at University of Birmingham on August 23, 2015
Proc Instn Mech Engrs Vol 215 Part H H07399 © IMechE 2001
AN IMPROVED THREE-DIMENSIONAL DIRECT NUMERICAL MODELLING AND THERMAL ANALYSIS 37

14 Singapore Cancer Registry, 1999, Ministry of Health, Victoria, Australia). <http//www.cmis.csiro.au/index.


Singapore. <http//www.gov.sg/moh/hfacts/hfacts.html>. htm>.
15 National Cancer Institute, National Institutes of Health, 24 TECPLOT User’s Manual Version 7.0, 1997 (Amtec
Bethesda, Maryland. <http://rex.nci.nih.gov/NCI_Pub_ Engineering Inc., Bellevue, Washington). <http//
Interface/raterisk/rates42.html>. www.amtec.com>.
16 Hoe, A. L. Breast cancer kills. Does early detection make 25 Werner, J. and Buse, M. Temperature proles with respect
a diVerence? In Public Symposium on Cancer, 1998 to inhomogeneity and geometry of the human body.
(Singapore Cancer Society, Singapore). J. Appl. Physiol., 1988, 65(30), 1110–1118.
17 Ng, E. H., Ng, F. C., Tan, P. H., Chiang, G., Tan, K. P., 26 Gautherie, M., Quenneville, Y. and Gros, C. M. Metabolic
Seow, A., Emmanuel, S., Tan, C. H. and Ho, G. H. Results heat production, growth rate and prognosis of early breast
of intermediate measures from a population-based, ran- carcinomas. Biomedicine, 1975, 22, 328–336.
domized trial of mammographic screening prevalence and 27 Priebe, L. Heat transport and specic blood ow in homo-
detection of breast carcinoma among Asian women—the geneously and isotropically perfused tissue. In Physiological
Singapore breast screening project. Cancer, 1998, 82(8), and Behavioural Temperature Regulation ( Eds J. D. Hardy,
1521–1528. A. P. Gagge and A. J. Stolwijk), 1970, pp. 272–280
18 Cady, B., Stone, M. D., Schuler, J. C., Thakur, R., Wanner, (Charles C. Thomas, Springeld, Illinois).
M. A. and Lavin, P. T. The new era in breast cancer, 28 Vaupel, P. Tumor blood ow. In Blood Perfusion and
invasion, size and nodal involvement dramatically decreas- Microenvironment of Human Tumors: Implications for
ing as a result of mammographic screening. Arch. Surg., Clinical Radiooncology (Eds M. Molls and P. Vaupel ),
1996, 131, 301–308. 1998, pp. 41–46 (Springer-Verlag, Berlin).
19 Pennes, H. H. Analysis of tissue and arterial blood tempera- 29 Love, T. J. Thermography as an indicator of blood per-
ture in resting human forearm. J. Appl. Physiol., 1948, 2, fusion. Ann. New York Acad. Sci., 1980, 335, 383–415.
93–122. 30 Gautherie, M. Thermobiological assessment of benign and
20 Sudharsan, N. M., Ng, E. Y.-K. and Teh, S. L. Direct simu- malignant breast diseases. Am. J. Obstet. Genecology, 1983,
lation of breast abnormality. In Proceedings of the ASME 147, 861–869.
Bioengineering Conference, Big Sky, Montana, 1999, 31 Varioscan 3011-ST Technical Data, Jenoptik, Laser, Optik,
Vol. 42, pp. 397–398. Systeme GmbH, Optical Systems Engineering, Jena,
21 Ng, E. Y.-K. and Sudharsan, N. M. Benchmarking the Germany. <http://www.jenoptick.com.ag>.
bioheat equation for nite element analysis. In Proceedings 32 Gautherie, M. Temperature and blood ow patterns in
of the Fourth International Symposium on Computer breast cancer during natural evolution and following radio-
Methods in Biomechanics and Biomedical Engineering, theraphy. In Biomedical Thermology, 1982, pp. 21–64 (Alan
Lisbon, Portugal, 13–16 October 1999. R. Liss, Inc., New York).
22 FEMAP User Guide Version 5, 1998 (Enterprise Software 33 Keyserlingk, J. R., Ahlgren, P. D., Yu, E. and Belliveau, N.
Products, Exton, Pennsylvania). <http//www.femap.com>. Infrared imaging of the breast: initial reappraisal using
23 FASTFLO Tutorial Guide Version 2.1, 1998 (CSIRO high-resolution digital technology in 100 successive cases
Mathematical and Information Services (CMIS ), Carlton, of stage I and II breast cancer. Breast J., 1998, 4, 245–251.

Downloaded from pih.sagepub.com at University of Birmingham on August 23, 2015


H07399 © IMechE 2001 Proc Instn Mech Engrs Vol 215 Part H

You might also like