mojra afsaneh (Orcid ID: 0000-0001-8790-7016)

Wei Guo-Wei (Orcid ID: 0000-0002-5781-2937)

Wei Guo-Wei (Orcid ID: 0000-0002-5781-2937)

Analysis of thyroid thermographic images for detection

of thyroid tumor: An experimental-numerical study
Farshad Bahramiana,1, Afsaneh Mojraa,2,*

a
Department of Mechanical Engineering, K. N. Toosi University of Technology, 15 Pardis
St., Tehran 1991943344, Iran

*
Corresponding author

1
M.Sc. student of Mechanical Engineering
[email protected]
2
Assistant Professor of Mechanical Engineering
[email protected]

This article has been accepted for publication and undergone full peer review but has not been
through the copyediting, typesetting, pagination and proofreading process which may lead to
differences between this version and the Version of Record. Please cite this article as doi:
10.1002/cnm.3192

This article is protected by copyright. All rights reserved.

Abstract

Thermography is a developing and noninvasive medical imaging technique which can be used
for diagnosis of body disorders based on temperature deviation from normal body temperature.
This research investigates the feasibility of thermography method in conjunction with artificial
neural networks (ANNs) for detection of thyroid tumors. For this purpose, first a three-
dimensional model of the healthy human neck is constructed based on patient-specific
computed tomography (CT) images. This model is used for analyzing bio-heat transfer in the
human neck. The healthy thyroid gland is considered as a heat source and generates heat
according to its temporal temperature. Finite element results verify the thermography potential
for detection of thyroid gland location and estimation of its butterfly shape on the neck
thermogram. The numerical analysis is carried out on 35 models with varying thermo-physical
parameters of the healthy thyroid gland, including heat generation and blood perfusion. The
acquired thermograms are used to develop an ANN for correlating the thermo-physical
parameters of the gland and temperature profile on the neck surface. In the next stage, dynamic
thermal images are captured from 10 healthy and three cancerous human cases. The
experimental thermal images are analyzed by the developed ANN and the corresponding
thermo-physical parameters are obtained. Results show that the estimated heat generation
𝑊 𝑊
values for the healthy cases are about 3,000 while it increases to more than 12,000 𝑚3 for
𝑚3

the cases with tumors. This significant variation confirms the potential of dynamic
thermography in diagnosis of thyroid tumors.

Keywords: Thermography; Thyroid gland tumors; Bio-heat transfer; Artificial neural network;
Finite element analysis.

Highlights

 Considering variable thermo-physical function for heat generation of thyroid gland

 Developing an artificial neural network for thermo-physical parameters estimation
 Conducting dynamic thermographic imaging on healthy and cancerous human subjects
 Assessment of experimental thermograms by means of the developed neural network
 Tumor diagnosis based on considerable elevation of thermo-physical parameters

This article is protected by copyright. All rights reserved.

 1. Introduction

Thyroid gland is a butterfly-shaped endocrine gland which is located at the front of the human’s
neck, below the Adam’s apple. Thyroid gland takes iodine and converts it into specific
hormones which control the metabolism of body organs. Published statistics show that from
1990 to present, the number of women annually diagnosed to have thyroid cancer is tripled.
Published statistics by American Cancer Society show that in year 2017 about 57,000 people
with thyroid cancer have been diagnosed. Today, thyroid cancer is the fifth women’s common
cancer which is predicted to be the fourth and the second women’s common cancer in 2020
and 2030, respectively [1-3].

Over the last two decades, medical image processing has been widely used for diagnosis of the
diseases including cancers. Computer-aided processing has noticeably increased the accuracy
of interpretation of medical images and has facilitated the diagnosis task for the physician [4].
Commonly, finding cancerous tumor in the early stage, increases the chance of patient’s cure
up to 85% while for advanced stages of cancer, the chance of successful treatment reduces to
15% [5]. While most cancers can be found during a routine checkup, there is no recommended
screening test to find thyroid cancer early. Imaging tests are not recommended as screening
tests for thyroid cancer unless a person is at increased risk. Meanwhile, thyroid medical images
by ultrasonography have relatively low quality while it is operator dependent which leads to
improper detection of thyroid nodules [6]. For CT scan, MRI and PET scan imaging,
intravenous contrast agent is injected which can be harmful if it is used periodically. Moreover,
these methods cannot be used during the period of pregnancy [6].

An alternative for imaging tests is infrared thermography which is totally non-invasive to the
body [7]. This method can detect soft tissue defects such as cancerous tumors based on
temperature measurement. In this method, temperature profile on the skin surface is captured
by a thermographic camera to construct a thermogram. The thermogram has a symmetrical
pattern about the sagittal plane. In the case of soft tissue defects, spots with higher temperature
appear in the thermal image hence the symmetrical pattern vanishes [8].

Over the last decade, thermography has had progressive elaboration in the medical applications
and has been widely used for detection of skin cancers [9, 10], breast cancers [11, 12], liver
cancers [13], eye diseases [14, 15] and brain tumors [16]. Gautherie [5] used thermography for
early diagnosis of breast cancer. It was observed that patients with asymmetrical thermogram
of the breasts get breast cancer in the next five years. Keyserlingk et al. [17] used clinical

This article is protected by copyright. All rights reserved.

examination, mammography and thermography techniques for detection of breast cancerous
tumors at stages one and two. They concluded that clinical examination, mammography and
thermography have detection accuracy of 61%, 66% and 83%, respectively. They also showed
that the smallest tumor diameter detectable by the thermography is 1.28 cm, while it is 1.66 cm
by the mammography. Helmy et al. [18] conducted a research on detection of thyroid nodules
by means of thermography. They used a simplified geometry of the human neck and
approximated heat generation in the thyroid gland. They observed higher temperature in the
region of nodules on the thermogram. The numerical results were employed to identify the
required accuracy of temperature measurement to detect thyroid nodules. Their numerical
analysis suffered from geometrical simplifications and thermo-physical parameters
approximations for the thyroid and nodules. Furthermore, the hot spots were not necessarily
indicative of nodules. Alves et al. [19] performed medical examinations to affirm higher
accuracy of thermography in comparison to power Doppler sonography for diagnosis of thyroid
nodules. Rizkala et al. [6] conducted a research to compare the results of heat transfer modeling
in a simplified geometry consists of thyroid gland and neck with the thermography images
captured from a prototype. They concluded that the thyroid nodules have higher temperature
in comparison to the healthy thyroid tissue. González et al. [20] investigated the potential of
thermography for detection of thyroid nodules. They analyzed thermograms captured from the
patients’ necks and confirmed the ability of thermography as a cost-effective method for
detection of thyroid nodules.

Considering the thyroid gland as a heat source, its thermo-physical properties have significant
variation by the presence of a tumor. Due to noticeable effect of heat generation amount on the
temperature profile, inclusion of a physical function for heat generation of the thyroid gland is
necessary. This function should consider varying properties for the case of tumor. This has
been underestimated in most of the previous studies. In this research, a real three-dimensional
model of the human neck and its fundamental organs including thyroid gland, trachea, common
carotid artery (CCA) and internal jugular vein (IJV) is constructed based on medical CT-scan
images. The thyroid gland is modelled as a heat source by a thermo-physical function due to
its metabolic activity. Blood perfusion is considered for all neck organs. A thermal analysis is
conducted for the model by solving the Pennes bio-heat transfer equation and thermogram is
attained on the neck surface.

There is knowledge about different metabolic heat generation and blood perfusion of thyroid
tumors compared to healthy thyroid gland. Therefore the numerical analysis is performed for

This article is protected by copyright. All rights reserved.

a wide range of heat generation and blood perfusion values for the purpose of tumor detection.
For each pair of heat generation and blood perfusion values, the temperature profile is
extracted. Based on the results, an artificial neural network (ANN) is developed to find a
correlation between heat generation and blood perfusion values and temperature profile on the
neck surface.

In the next stage, 10 healthy human cases and 3 cases with thyroid tumors are examined with
a thermographic camera through a specified protocol and thermograms are obtained. The
temperature profiles are evaluated by the proposed ANN and heat generation and blood
perfusion values are acquired. These values are compared between the healthy and tumoral
cases to assess their potential in diagnosing thyroid tumor.

2. Material and methods

2.1. Numerical simulation of healthy thyroid gland

2.1.1. Geometry construction

Three-dimensional model of normal human neck and its vital organs including the thyroid
gland, CCA, IJV and the trachea is constructed by using patient-specific CT-scan images
(Figures 1a-1c) on the basis of Helsinki agreement. The patient is a 70-year-old female. CT-
scan images are captured by EMOTION 16 CT-scan system (EMOTION 16, Siemens,
Germany) at AmirAlam hospital (Radiology department, Tehran, Iran). Prior to imaging,
intravenous contrast agent has been injected for increasing the contrast of the thyroid gland,
CCA and IJV.

The CT-scan images are imported to Mimics Medical software (version 19, Materialise NV,
Leuven, Belgium). Then by Materialise 3-Matic Medical software (version 11, Materialise NV,
Leuven, Belgium), the geometry is refined and an appropriate file is provided to import the
geometry into the COMSOL Multiphysics software (version 5.3, Comsol Inc., Burlington,
MA). The resultant 3D model of normal human neck is shown in Figure 1d. In this model, the
human neck has a height of 124 mm and the bottom section of this model can be considered as
a rectangle with dimension of 285 mm × 97 mm. Figure 1e presents the internal organs
included in heat transfer modeling.

This article is protected by copyright. All rights reserved.

 2.1.2. Theoretical equations
For heat transfer modeling in the 3D model, Pennes bio-heat transfer equation is employed [21]
(Eq. (1)).

Tt ( X , t )
t ct  kt Tt ( X , t )  b cbb (Ta  Tt ( X , t ))  Qm ( X , t ) (1)
t
Where 𝑇𝑡 is the tissue temperature versus location (𝑋) and time (𝑡), 𝜌𝑡 , 𝑘𝑡 and 𝑐𝑡 represent
density, thermal conductivity and specific heat of the tissue, respectively, 𝜌𝑏 , 𝑐𝑏 and 𝜔𝑏
represent density, specific heat, and perfusion rate of the blood, respectively, 𝑇𝑎 is temperature
of artery and 𝑄𝑚 is metabolic heat generation.

There is correlation between temporal temperature of the tissue and the amount of metabolic
heat generation [22]. According to this knowledge, the following function is used for metabolic
heat generation (Eq. (2)).

Qm ( X , t )  Qm0  3((T ( X ,t )T0 )/10) (2)

Where 𝑄𝑚0 is amount of heat generation at the core body temperature (𝑇0 = 37 ˚𝐶).

2.1.3. Boundary conditions

Four exterior surfaces including two lateral surfaces besides top and base surfaces are thermally
insulated. The convective thermal boundary condition is used for the neck surface (Eq. (3)).

Tt ( X , t )
 kt  h f (Ts  T f ) (3)
n

Where, ℎ𝑓 represents the coefficient of convective heat transfer between the neck surface and
the surrounding air, 𝑇𝑠 and 𝑇𝑓 are skin and surrounding air temperatures, respectively, and 𝑛 is
unit vector that is normal to the isotherm surface.

Generally, a normal mature breathes 15-20 times in one minute, so each inspiration or
expiration lasts about 1.5-2 seconds. During the inspiration, the ambient air with specified
temperature enters the nostril, and during transportation of breathed air from nose to lungs, the
breathed air temperature increases by means of convective heat transfer with the inner surface
of trachea. Therefore, suitable boundary condition for the inner layer of trachea can be
expressed by Eq. (4).

This article is protected by copyright. All rights reserved.

 Ttr ( X , t )
 kt  htr (Ttr  Tair ) (4)
n

Where ℎ𝑡𝑟 denotes the convective heat transfer coefficient between the inner layer of trachea
and the breathed air, 𝑇𝑡𝑟 is the trachea temperature and 𝑇𝑎𝑖𝑟 is the breathed air temperature. The
temperature of nasal cavity and breathed air are assumed to be 37 ˚𝐶 and 34 ˚𝐶, respectively
[23, 24].

The condition for heat transfer between the neck and the internal organs including thyroid
gland, trachea, CCA and IJV are expressed as continuity of temperature and heat flux (Eq. (5)).

Tt Tb
Tt  Tb , kt  kb (5)
n n

Where 𝑇𝑏 and 𝑘𝑏 represent the temperature and thermal conductivity of blood, respectively.

For inner layer of the blood vessels, constant temperature of 37 ˚𝐶 is defined. A schematic view
of the boundary conditions is depicted in Figure 2.

2.1.4. Thermo-physical properties and simulation parameters

Thermo-physical properties of the neck tissue which mainly consists of muscle and its primary
𝑊
organs and the blood are listed in Table 1. Heat generation of 4200 𝑚3 and blood perfusion rate

of 0.0667 ml/s/ml are considered as the base values.

Ambient air temperature and the coefficient of convection heat transfer between the neck
𝑊
surface and surrounding air are considered to be 25 ˚𝐶 and 10 respectively [28]. Nusselt
𝑚2 ℃

number is defined by 𝑁𝑢 = ℎ𝑡𝑟 ∙ 𝑑⁄𝜆 and is used for calculation of convection heat transfer
coefficient in trachea. In definition of the Nusselt number, 𝜆 is thermal conductivity of air and
𝑑 denotes the trachea diameter which has an average value of 2 𝑐𝑚. By considering an average
𝑚 𝑊
air velocity of 2 𝑠 , the average coefficient of convection heat transfer is equal to 35 𝑚2 ℃ for

inspiration and expiration phases.

2.1.5. Simulation procedure

For modeling of heat transfer, COMSOL Multiphysics software is used. The thermal
simulations are carried out by using a system with one quad-core CPU (Intel Core i7 2670QM)
and with 8 GB RAM. The average computational time for 3D simulations is 430 seconds.

This article is protected by copyright. All rights reserved.

The initial temperature of all neck organs are set to 37 ˚𝐶. Temperature variation on the neck
surface is a time-dependent procedure according to the experimental tests performed in the
present study, which would be explained in section 2.3.2. Time duration of transient thermal
analysis is 60 𝑚𝑖𝑛 with precision of 10-6 and time step of 0.005 second. For meshing,
tetrahedral elements are used. For assessing mesh independency, three meshing with 34168,
73715 and 186123 elements are examined and temperature of a reference point on the neck
surface is studied and reported in Table 2. The related computational times are also included.
The reference point is located in front of the thyroid gland where maximum temperature
gradient is observed. Location of the reference point is displayed in Figure 3.

According to the temperature variation for different grids and considering the computational
cost, grid with total element number of 73715 is selected for the numerical modeling.

2.2. Parameter estimation by multi-layer ANN and the procedure flowchart

There is knowledge about different metabolic heat generation and blood perfusion of thyroid
tumors compared to healthy thyroid gland. Therefore, the numerical analysis is carried out for
a wide range of heat generation and blood perfusion values provided in Table 1. According to
this Table, 35 simulations are conducted and for each simulation, the temperature profile is
extracted. In order to provide a correlation between the temperature profile and the
corresponding heat generation and blood perfusion values, an ANN is employed.

In the present study, a four-layer network is developed for estimation of the thyroid gland’s
thermo-physical parameters based on the neck surface temperature. For training the network,
two vertical paths are defined on the neck surface, which are displayed in Figure 4a. The first
path is located on the midplane of body and the second path is located at the right side of the
neck and in front of the right lobe of thyroid gland. A total number of 60 points are defined on
each path for which the temperature data are collected. Therefore, for each pair of heat
generation and blood perfusion values, 120 temperature data are collected. For 35 thermal
analyses, 4200 datasets are presented. These data are divided into 2940 training and 1260
testing datasets. Each set of data includes location, heat generation, and blood perfusion values
as the inputs of the network and temperature on the neck surface as the output. Details of the
employed ANN are listed in Table 3.

Assessment of the proposed 𝐴𝑁𝑁 is carried out by calculating two statistical parameters,
including the efficiency (𝐸) and the regression coefficient for the training and testing

This article is protected by copyright. All rights reserved.

procedures. Efficiency of the network is calculated by using Eq. (6). Theoretically, an ideal
network has efficiency and regression coefficient equal to 1.0.

N N

 ( X d  X s )2   ( X d  X s )2
E i 1
N
i 1
(6)
(X
i 1
d  X s) 2

where, 𝑋 represents the thyroid metabolic heat generation, 𝑋𝑑 denotes the desired or real value
which has been obtained by the thermal analysis, 𝑋𝑠 is the estimated value by the network, and
𝑋̅s is the average value of estimated variables among the whole datasets.

Training of the network is performed by the temperature results attained form numerical
analyses. The trained network will be subsequently employed for the experimental
temperatures attained from the thermograms in the experiments on human subjects. Through
this, the corresponding heat generations and blood perfusions are estimated. To this end, two
vertical paths are defined on the thermograms similar to the numerical study and temperature
data are collected on similar points. The paths are displayed on the CT-scan image of a case
with thyroid tumor (Figure 4b). As it can be seen in this Figure, the second path is located in
front of the tumoral region. For all cancerous cases, path locations are similar to this case.

The estimated thermo-physical parameters will be compared between the healthy and tumoral
cases in section 3.3. A flowchart of the whole procedure is illustrated in Figure 5.

2.3. Experimental test

An experimental study is conducted on the human subjects to attain the thermogram on the
neck surface. The population study includes of 10 healthy cases and 3 cancerous cases with
thyroid tumors.

2.3.1. Experimental setup

The thermographic experiments are performed by means of an infrared camera, FLIR A655sc
with image resolution of 640×480 and thermal sensitivity of less than 0.03 ℃ (according to
the FLIR A655sc user manual). The procedure of capturing thermal images are performed in a
room with controlled ambient temperature of 25 ℃ ± 1 ℃ and ambient humidity of 20% ±5%.
There are two methods for capturing thermographic images; static imaging and dynamic
imaging. In static imaging, after patient preparation, one image is captured from the intended
organ. In dynamic imaging, prior to the image capturing, skin surface of the intended organ is

This article is protected by copyright. All rights reserved.

cooled down with different methods including cool air stream, covering with wet cotton gauze
and spraying methanol. Subsequently, during the return procedure to thermal balance with the
surrounding, a series of thermographic images are captured. In the present study, a dynamic
imaging is performed on the neck surface of the human subjects.

2.3.2. Experimental procedure

The population study includes of 10 healthy males and 3 male patients with thyroid tumors.
Cases ranged from 22 to 30 years old with an average age of 26. Tumor diameters are 34.19
mm, 18.29 mm, and 25.29 mm for the cancerous cases #1, #2, and #3, respectively. For all
cancerous cases, tumor location is on the right lobe of the thyroid gland. For capturing dynamic
thermal images, after patient preparation, liquid solution consists of water and 75% volume
concentration of medical ethanol is sprayed on the subject’s neck to cool the neck tissue. During
the return procedure to thermal balance, 5-minute infrared movie is recorded with the thermal
camera from the neck surface and specific frames of the thermal video are chosen for further
analysis.

Prior to the analysis of thermal images, first each thermal image is segmented in order to
separate the subjects’ neck form the image background. Subsequently, standard Perona–Malik
anisotropic diffusion filter is used for noise removal from the image [29, 30]. This anisotropic
filter has the ability to reduce speckle and spurious noises and improves edges of the image.
Finally, the grayscale thermal image is converted to a pseudo color thermal image. The steps
of image processing for one sample thermal image by the end of imaging period (at 5 minutes)
are shown in Figure 6a to Figure 6d. The image belongs to one healthy human subject. Two
other frames of the same case at specific times are presented in Figure 6e and Figure 6f.

3. Results and discussion

The temperature data obtained from the numerical thermal analysis for the healthy case is
compared with data provided by Jin et al. [31] in a similar numerical study, in Table 4.
Temperature data belong to the points located on two vertical paths on the neck surface which
have been previously defined in Figure 4a. According to this Table, maximum difference is 2%
which proves the reliability of thermal simulation in the present study.

3.1. Thermal analysis of healthy thyroid gland

The 3D model is thermally analyzed by considering heat generation and blood perfusion base
values provided in Table 1. The resultant thermal map on the neck surface is shown in Figure

This article is protected by copyright. All rights reserved.

7a. Figure 7b presents temperature variation in depth (on the sagittal plane). According to
Figure 7b, significant temperature variation could be observed in front of the thyroid gland as
a result of metabolic heat generation by the gland. Effect of the gland is diminished near the
neck surface but still there is detectable temperature gradient on the skin. Higher metabolic
activity and blood perfusion rate of the thyroid gland compared to the surrounding tissues are
the reasons of temperature rise on the neck surface in front of the gland. Any disorders in
conjunction with elevation of thermo-physical parameters of the gland, increases the amount
of temperature rise. In Figure 7c, the temperature map on four sections parallel with the coronal
plane of the body, in the direction of anterior to posterior of the neck is provided. Thermally
affected areas in the vicinity the gland can be observed on different sections. Shapes of these
areas match with the shape of the gland.

The thermogram on the neck surface not only indicates the location of thyroid gland but also
suggests the butterfly shape of it. An approximate shape of the gland can be inferred from
Figure 7a by considering regions with rather high temperatures. In section 3.2, a correlation is
developed between regions with rather high temperatures on the thermogram and thermo-
physical properties of the gland by employing an ANN. Thyroid disorders may deteriorate the
symmetrical pattern of the thermogram about the sagittal plane. By identifying the shape of
thyroid gland on the thermogram by using the ANN, thyroid nodules become detectable. In
Figure 8, temperature variation is plotted on two vertical paths on the neck surface (the paths
have been defined in Figure 4a). According to this figure, an approximate temperature increase
of 1 ℃ is attained in front of the thyroid gland.

3.2. ANN analysis for estimation of thermo-physical parameters

The four-layer ANN is developed to correlate the surface temperature with the corresponding
values of heat generation and blood perfusion rate. It is observed that heat generation has
predominant effect on the temperature profile compared to blood perfusion rate of the thyroid
gland. Figure 9 shows the desired values of heat generation which correspond to the input
values of the numerical analysis, versus estimated values by the network. Figure 9a and Figure
9b present the desired and estimated data on path number 1 and path number 2, respectively.

For the proposed network, efficiency (E) is 0.93 and 0.96, and R-squared values are 0.93 and
0.97 for path #1 and path #2, respectively. The resultant values support the ANN high potential
in estimation of heat generation of the thyroid gland.

This article is protected by copyright. All rights reserved.

According to the bio-heat transfer equation (Eq. (1)), heat generation and blood perfusion rate
have controversial effects on the temperature profile. Sensitivity analysis indicates that by
300% increase of 𝑄𝑚 and 13000% increase of 𝜔𝑏 , the temperatures vary about 0.3% and 1%,
respectively; which are rather high according to the normal body temperature. Combination of
both 𝑄𝑚 and 𝜔𝑏 indicates more significant temperature rise; as the maximum temperature in
𝑊
front of the thyroid gland at 𝑄𝑚 = 13000 & 𝜔𝑏 = 0.0667 𝑚𝑙/𝑠/𝑚𝑙 increases by 2.3%
𝑚3
𝑊
from the corresponding temperature at 𝑄𝑚 = 3000 & 𝜔𝑏 = 0.0005 𝑚𝑙/𝑠/𝑚𝑙.
𝑚3

3.3. Experimental thermal imaging of healthy and cancerous cases

The thermograms are captured from 10 healthy cases and three cancerous cases by the thermal
camera. Figure 10 shows the processed pseudo color thermal image for 13 attended cases.
Approximate locations of the thyroid glands with rather high temperatures are indicated in the
thermograms of healthy cases.

Temperature variation on path number 1 and path number 2, for the healthy cases and the
cancerous cases, are provided in Figure 11a and Figure 11b, respectively. The distance between
𝑥1 to 𝑥2 corresponds to the neck region where the thyroid gland is located behind the neck
surface and it has been indicated in Figure 4b. According to this Figure, this region is also in
front of the tumoral tissues for the cancerous cases. As can be seen in Figure 11, the
temperatures increase significantly in front of the tumoral regions on both paths for the
cancerous cases. The amount of temperature increase is the most for the first cancerous case
due to the largest thyroid tumor compared to the other cancerous cases.

Temperature data for all cases on two defined paths on the distance between 𝑥1 to 𝑥2 are given
to the developed ANN and the corresponding heat generation and blood perfusion rate are
calculated. Figure 12a is a plot of heat generation values on path number 1 for a sample healthy
case and Figure 12b is the same plot for a sample cancerous case. Figure 12c and Figure 12d
are the plots for the same healthy and cancerous cases on path number 2. For better observation
of heat generation difference between healthy and tumoral cases, heat generation values for the
sample healthy case and the cancerous case are plotted together on path number 1 and path
number 2 in Figure 12e and Figure 12f, respectively. Significant elevation of heat generation
by the tumoral thyroid compared to the healthy thyroid can be inferred from these plots.
The average of heat generation values is calculated among all points for each case and reported
in Table 5. In this Table, the average values are reported for each path, separately.

This article is protected by copyright. All rights reserved.

 𝑊
From this Table, the average of heat generation is approximately 3022 ± 52 among the
𝑚3
𝑊
healthy cases, while it rises to an average value of 12400 among the cancerous cases.
𝑚3

Therefore, in the tumoral thyroid gland, heat generation is up to 4-fold more than the healthy
case. Given that tumor is the result of abnormal increase of cells, the elevated metabolic heat
could be inferred.

According to the pathological findings, tumor growth has two distinct phases which are the
avascular phase and the vascular phase. During the avascular phase of tumor growth, a limited
number of cancer cells aggregates in the healthy tissue. In the experimental models, maximum
tumor size in the avascular phase is considered 1-2 mm [32]. The next stage of the tumor growth
is the vascular phase in which new blood vessels are recruited to nourish the increased number
of tumor cells [33]. It is affirmed that by the time a tumor is approximately 2-3 mm, it will
contain more than 1 million cells and it has already entered the vascular phase of growth [34].
In this phase, the cell metabolism is increased enormously which results in the elevation of heat
generation by the tumoral tissue. Findings of the present study for the tumor in the vascular
phase show that the increased metabolic heat generation has measurable effect on the
temperature profile of the neck surface.

It is worth to mention that the physiological age and gender may have effects on the
functionality of thyroid gland [35]. However to the best of the authors’ knowledge, there is no
published quantitative relationship between the thermo-physical parameters of thyroid gland
and age and gender. Therefore, results of the present study are more suitable for the age group
of young adults.

4. Conclusions
The main objective of the present study is to identify thermal map of healthy thyroid gland
with the purpose of detecting thyroid tumors. To this end, at first stage of the study, a thermal
analysis is conducted for three-dimensional model of healthy thyroid gland. At this stage,
thermo-physical properties of the gland including the blood perfusion rate and metabolic heat
generation are varied over a rational range reported for the healthy tissue in the literature. The
main output of the thermal analysis is temperature profile on the neck surface in front of the
thyroid gland. Investigation of the obtained thermal image indicates the region in front of the
gland with rather high temperatures. Moreover, the butterfly shape of the gland can be implied
from the thermal image.

This article is protected by copyright. All rights reserved.

From the numerical analysis, it is also inferred that while increase of heat generation gives rise
to the temperature value, increase of the blood perfusion may have an opposite effect. In order
to clarify the correlation between thermo-physical parameters and the thermal map, a four-
layer ANN is employed. Assessment of the network shows that for higher values of heat
generation, performance of the network attenuates.

In the next stage of the study, experiments are conducted on 10 healthy human cases and 3
cancerous cases through a specified protocol. The main output of the experiments is the
thermograms of the neck surface captured by an infrared camera during a dynamic imaging
procedure. The network which is developed by the numerical temperatures is used by the
experimental temperature data to estimate the corresponding thermo-physical parameters of
the thyroid glands. Results of this estimation for the healthy thyroid glands are very close
together and very close to the thermo-physical properties assumed for the healthy thyroid gland
as the inputs of the numerical study. Meanwhile, results for the cancerous cases show
considerable deviation from the healthy cases' results. Therefore, assessment of thermo-
physical parameters can be considered as a suitable tool for screening health status of the
thyroid gland. It can be also used as a screening tool for diagnosis of thyroid tumors or any
types of disorders which affect the thermal performance of the gland.

Funding
This research did not receive any specific grant from funding agencies in the public,
commercial, or not-for-profit sectors.

Conflict of interest
The authors declare that they have no conflict of interest.

Ethical approval
All procedures followed were in accordance with the ethical standards of the responsible
committee on human experimentation (institutional and national) and with the Helsinki
Declaration of 1975, as revised in 2000 (5). Informed consent was obtained from all patients
for being included in the study.

This article is protected by copyright. All rights reserved.

References
1. Amirican Cancer Society; www.cancer.org
2. Morris, L.G., R.M. Tuttle, and L. Davies, Changing Trends in the Incidence of Thyroid
Cancer in the United States. JAMA Otolaryngol Head Neck Surg, 2016. 142(7): p. 709-
11.
3. Rahib, L., B.D. Smith, R. Aizenberg, A.B. Rosenzweig, J.M. Fleshman, and L.M.
Matrisian, Projecting cancer incidence and deaths to 2030: the unexpected burden of
thyroid, liver, and pancreas cancers in the United States. Cancer Res, 2014. 74(11): p.
2913-21.
4. Goceri, E. and S. C., Biomedical Information Technology: Image Based Computer
Aided Diagnosis Systems, in International Conference on Advanced Technologies. 28
April – 1 May 2018: Antalya, Turkey. p. 132.
5. Gautherie, M., Thermobiological assessment of benign and malignant breast diseases.
American Journal of Obstetrics and Gynecology, 1983. 147(8): p. 861-869.
6. Rizkalla, J., W. Tilbury, A. Helmy, V.K. Suryadevara, M. Rizkalla, and M.M.
Holdmann, Computer simulation/practical models for human thyroid thermographic
imaging. Biomedical Science and Engineering, 2015. 8: p. 246-256.
7. Ring, F., Thermal imaging today and its relevance to diabetes. Journal of Diabetes
Science and Technology 2010. 4: p. 857-862.
8. Gavriloaia, G., A.M. Ghemigian, and A.E. Hurduc, Early cancer diagnosis by image
processing sensors measuring the conductive or radiative heat. 13th International
Conference on Biomedical Engineering, 2009: p. 427-430.
9. Bhowmik, A., R. Repaka, R. Mulaveesala, and S.C. Mishra, Suitability of frequency
modulated thermal wave imaging for skin cancer detection-A theoretical prediction. J
Therm Biol, 2015. 51: p. 65-82.
10. Cholewka, A., A. Stanek, S. Kwiatek, A. Sieron, and Z. Drzazga, Does the temperature
gradient correlate with the photodynamic diagnosis parameter numerical colour value
(NCV)? Photodiagnosis Photodyn Ther, 2013. 10(1): p. 33-8.
11. Saniei, E., S. Setayeshi, M.E. Akbari, and M. Navid, vascular network matching in
dynamic thermography for breast cancer detection. Quantitative InfraRed
Thermography Journal, 2015. 1(12): p. 24-36.
12. Shi, G., F. Han, C. Liang, L. Wang, and K. Li, A novel method of thermal tomography
tumor diagnosis and its clinical practice. Applied Thermal Engineering, 2014. 73(1):
p. 408-415.
13. Keangin, P., T. Wessapan, and P. Rattanadecho, Analysis of heat transfer in deformed
liver cancer modeling treated using a microwave coaxial antenna. Applied Thermal
Engineering, 2011. 31(16): p. 3243-3254.
14. Gonnermann, J., J.P. Klein, M.K. Klamann, A.K. Maier, U. Pleyer, A.M. Joussen, and
E. Bertelmann, Dry eye symptoms in patients after eyelid reconstruction with full-
thickness eyelid defects: using the Tomey TG-1000 thermographer. Ophthalmic Res,
2012. 48(4): p. 192-198.
15. Tan, J.-H., E.Y.K. Ng, U. Rajendra Acharya, and C. Chee, Infrared thermography on
ocular surface temperature: A review. Infrared Physics & Technology, 2009. 52(4): p.
97-108.
16. Sadeghi-Goughari, M. and A. Mojra, Intraoperative thermal imaging of brain tumors
using a haptic-thermal robot with application in minimally invasive neurosurgery.
Applied Thermal Engineering, 2015. 91: p. 600-610.

This article is protected by copyright. All rights reserved.

17. Keyserlingk, J.R., P.D. Ahlgren, E. Yu, and N. Bellivean, Infrared imaging of breast:
Initial reappraisal using high-resolution digital technology in 100 successive cases of
stage I and II breast cancer. The Breast Journak 1998. 4: p. 241-251.
18. Helmy, A., M. Holdmann, and M. Rizkalla, Application of Thermography for Non-
Invasive Diagnosis of Thyroid Gland Disease. IEEE Transactions on Biomedical
Engineering, 2008. 55: p. 1168-1175.
19. Alves, M.L. and M.H. Gabarra, Comparison of power Doppler and thermography for
the selection of thyroid nodules in which fine-needle aspiration biopsy is indicated.
Radiol Bras, 2016. 49(5): p. 311-315.
20. González, J.R., É.O. Rodrigues, C.P. Damião, C.A.P. Fontes, A.C. Silva, A.C. Paiva,
H. Li, C. Du, and A. Conci, An Approach for Thyroid Nodule Analysis Using
Thermographic Images. Application of Infrared to Biomedical Sciences, 2017: p. 451-
475.
21. HH, P., Analysis of tissue and arterial blood temperatures in the resting human
forearm. J Appl Physiol., 1948. 1: p. 93-122.
22. Xu, X., P. Tikuisis, and G. Giesbercht, A mathematical model for human brain
coolingduring cold-water near-drowning. J Appl Physiol., 1999. 86: p. 265-272.
23. Rouadi, P., F. Baroody, D. Abbott, E. Naureckas, J. Solway, and R. Nacleiro, A
technique to measure the ability of the human nose to warm and humidify air. Journal
of Applied Physiology 1999. 87(1): p. 400-406.
24. Webb, P., Air temperatures in respiratory tracts of resting subjects in cold. Journal of
applied physiology, 1951. 4(5): p. 378-382.
25. Deng, Z.S. and J. Liu, Mathematical modeling on temperature mapping over
skinsurface and its implementation in disease diagnostics. Computers in Biology and
Medicine, 2004. 34(6): p. 495-521.
26. Niemz, M., Laser-tissue interactions: fundamentals and applications. Springer, 2002.
27. Werner, J. and M. Buse, Temperature Profiles with Respect to Inhomogeneity and
Geometry of Human Body. Journal of Applied Physiology, 1988. 65: p. 1110–1118.
28. Jin, C., Z. He, and J. Liu, MRI-based finite element simulation on radiofrequency
ablation of thyroid cancer. Comput Methods Programs Biomed, 2014. 113(2): p. 529-
38.
29. Perona, P. and J. Malik, Scale-space and edge-detection using anisotropic diffusion.
IEEE Transactions on Pattern Analysis and Machine Intelligence, 1990. 12(7): p. 629 -
639.
30. Goceri, E., B. Goksel, J.B. Elder, V.K. Puduvalli, J.J. Otero, and M.N. Gurcan,
Quantitative validation of anti-PTBP1 antibody for diagnostic neuropathology use:
Image analysis approach. International Journal of Numerical Methods in Biomedical
Engineering, 2017. 33(11): p. 1-14.
31. Jin, C., Z. He, Y. Yang, and J. Liu, MRI-based three-dimensional thermal physiological
characterization of thyroid gland of human body. Medical Engineering and Physics,
2014. 36: p. 16-25.
32. Folkman, J., What is the evidence that tumors are angiogenesis dependent? JNCI J Natl
Cancer Inst Citeseer, 1990. 82: p. 4-7.
33. Mascheroni, P., C. Stigliano, M. Carfagna, D. Boso, L. Preziosi, P. Decuzzi, and B.
Schrefler, Predicting the growth of glioblastoma multiforme spheroids using a
multiphase porous media model. 2016. 15: p. 1215–1228.
34. Weidner, N., J. Semple, W. Welch, and J. Folkman, Tumor angiogenesis and
metastasis—correlation in invasive breast carcinoma. N Engl J Med Mass Medical
Soc, 1991. 324: p. 1-8.

This article is protected by copyright. All rights reserved.

35. Da costa, V., D. Moreira, and D. Rosenthal, Thyroid function and aging: gender-related
differences. Endocrinol, 2001. 171: p. 192-198.

This article is protected by copyright. All rights reserved.

Figure 1 Different view of the neck tissue, (a) coronal view, (b) axial view, (c) sagittal view,
(e) geometry of normal human neck, (b) geometry of neck primary organs, C1 to C4 represent
thyroid gland, trachea, CCA and IJV, respectively.

This article is protected by copyright. All rights reserved.

 Figure 2 Human neck boundary conditions: (a) blue surface has convective heat transfer;
other surfaces are insulated; (b) inner layer of blood vessels in blue color have constant
temperature of 37 C; (c) inner layer of trachea in blue color has convective heat transfer with
the breathed air.

This article is protected by copyright. All rights reserved.

Figure 3 Location of the reference point on the neck model meshed with tetrahedral elements

This article is protected by copyright. All rights reserved.

Figure 4 Location of two vertical paths for temperature measurement, (a) healthy case in the
numerical study, (b) cancerous case in the experimental study

This article is protected by copyright. All rights reserved.

 Correlation between thermo-physical parameters of the thyroid gland and
the thermogram on the neck surface

Development of ANN Estimation by the developed ANN

Numerical modeling of healthy Thermal experiments on 10 healthy

thyroid gland with specified thermo- cases and 3 cases with thyroid
physical parameters tumor

Obtain thermogram on Acquire thermograms on

the neck surface in front the neck surface in front
of thyroid gland of the thyroid gland

Training of a 4-layer ANN

Execute image
Inputs: Thermo-physical parameters processing procedure for
of the gland including metabolic noise reduction
heat generation, and blood
perfusion rate along with the
location of each point on the
thermogram Estimation of thermo-physical
Output: Temperature of each point parameters by using the developed
on the thermogram ANN

Comparing the obtained

parameters between
healthy cases and
between healthy and
tumoral cases

Figure 5 Flowchart of thermo-physical parameters estimation for the thyroid gland

This article is protected by copyright. All rights reserved.

 Figure 6 Thermal images of the human neck extracted from the thermal video after cooling
with ethanol spray; (a) raw thermal image after 5 minutes, (b) same image after segmentation,
(c) same image after noise removal by Perona–Malik filter [29, 30], (d) same image after
conversion to pseudo color thermal image, (e) pseudo color image at 0 minute, (f) pseudo
color image at 2 minutes

This article is protected by copyright. All rights reserved.

 Figure 7 (a) Thermogram on the neck surface for healthy thyroid gland, (b) temperature
variation on the sagittal plane, (c) temperature variation on four coronal planes, temperature
values are in C.

This article is protected by copyright. All rights reserved.

Figure 8 Temperature variations on two vertical paths for healthy thyroid gland

This article is protected by copyright. All rights reserved.

Figure 9 Comparison of thyroid gland heat generation estimated by the ANN (cross marks)
with desired values (circles), (a) on the first path, (b) on the second path

This article is protected by copyright. All rights reserved.

Figure 10 Thermograms obtained from dynamic thermal imaging, (a)-(j) healthy cases, (k)-
(m) cancerous cases, for healthy cases location of the right lobe, left lobe and the center of
thyroid gland are indicated with plus marks. Temperature values are in ℃.

This article is protected by copyright. All rights reserved.

Figure 11 Temperature variations on the neck surface, (a) on the first path, (b) on the second
path

This article is protected by copyright. All rights reserved.

 Figure 12 Estimation of heat generation by the developed ANN, (a) for a healthy thyroid
gland on path #1, (b) for a tumoral thyroid gland on path #1, (c) for a healthy thyroid gland
on path #2, (d) for a tumoral thyroid gland on path #2, (e) for healthy and tumoral thyroid
glands on path #1, (f) for healthy and tumoral thyroid glands on path #2

This article is protected by copyright. All rights reserved.

 Table 1 Material properties used in the thermal analysis [25, 26, 27]
𝒌𝒈 𝑱 𝑾 𝑾
 (𝒎𝟑 ) 𝒄(
𝒌𝒈.𝑪
) 𝒌(
𝒎.𝑪
) 𝝎𝒃 (ml/s/ml) 𝑸𝒎𝟎 (
𝒎𝟑
)

Neck tissue (muscle) 1050 3600 0.5 0.0005 420

0.0005-0.0064-0.0133- 3000-4200-5000-7000-
Thyroid gland 1050 3600 0.5
0.0252-0.0667 9000-11000-13000
Blood 1050 3600 0.5 - -

Table 2 Mesh independency assessment through temperature measurement of the reference

point

Model# Grid #1 Grid #2 Grid #3

Number of elements 34168 73715 186123
Computational time (s) 242 434 907
Reference point temperature (C) 35.110 35.1075 35.1074
Variation (%) - 0.007 0.003

Table 3 Properties of the four-layer ANN

Performance Number Number Transfer Number of neurons Number

function of inputs of outputs function 1st layer 2nd layer 3rd layer of epochs

MSE 3 1 Tansing 2 22 2 1000

This article is protected by copyright. All rights reserved.

Table 4 Comparison between numerical temperature data obtained from the present study
and study by Jin et al. [31]

Distance on Present study Study by Jin et al. [30] Error (%)

path (cm) Path #1 Path #2 Path #1 Path #2 Path #1 Path #2
0 34.23 34.19 34.51 34.40 0.80 0.60
1 34.50 34.38 34.66 34.72 0.46 0.96
2 35.02 34.65 34.93 34.96 0.26 0.90
3 35.27 34.85 35.10 35.26 0.49 1.15
4 35.43 35.08 35.38 35.46 0.13 1.09
5 35.43 35.18 35.55 35.49 0.32 0.86
6 35.41 34.95 35.35 35.23 0.18 0.79
7 35.30 34.77 34.78 34.97 1.50 0.58
8 35.12 34.43 34.41 34.56 2.05 0.40
9 34.51 34.18 33.85 34.27 1.94 0.27
10 34.28 34.03 33.74 33.96 1.59 0.22

Table 5 Average values of heat generation for each case on two predefined paths
Average of heat
Health
𝑾
Case condition of generation ( )
𝒎𝟑

number the thyroid

Path #1 Path #2
gland

#1 Healthy 3010.4 3002.8

#2 Healthy 3006.8 3004.5
#3 Healthy 3006.2 3005.5
#4 Healthy 3006.4 3002.9
#5 Healthy 3179.4 3169.6
#6 Healthy 3005.8 3002.0
#7 Healthy 3006.3 3001.9
#8 Healthy 3002.5 3000.9
#9 Healthy 3004.1 3005.4
#10 Healthy 3006.1 3007.2
#11 Cancerous 12997.6 12997.6
#12 Cancerous 12821.0 11242.6
#13 Cancerous 12872.6 11312.8

This article is protected by copyright. All rights reserved.

 Graphical Abstract
Variable metabolic heat generation of the thyroid gland according to its temporal temperature
is the crucial factor in differentiating a healthy gland from a tumoral one which has been
underestimated in previous studies. This study is focused on developing a correlation between
temperature profile on the neck surface in front of the thyroid gland and amount of heat
generation by the gland. Results are employed for healthy and cancerous human subjects and
existence of tumor is verified.

This article is protected by copyright. All rights reserved.