Intelligent Diagnosis of Breast Cancer with Thermograms

using Convolutional Neural Networks

Nurduman Aidossov

, Aigerim Mashekova

, Yong Zhao

, Vasilios Zarikas

Eddie Yin Kwee Ng

and Olzhas Mukhmetov

School of Engineering and Digital Sciences, Nazarbayev University, 53 Kabanbai batyr street, Nur-Sultan, Kazakhstan

School of Mechanical and Aerospace Engineering, Nanyang Technological University, 50 Nanyang Avenue,

639798, Singapore

Keywords: Breast Cancer, Thermography, Convolutional Neural Network, Intelligent Diagnosis.

Abstract: Breast cancer is a serious public health issue among women all over the world. The main methods of breast

cancer diagnosis include ultrasound, mammography and Magnetic Resonance Imaging (MRI). However, the

existing methods of diagnosis are not appropriate for regular mass screening in short intervals. On the other

hand, there is one non-invasive and low-cost method for mass and regular screening which is the so-called

thermography. Recent studies show rapid quality improvement of thermal cameras as well as distinct

development of machine learning techniques that can be combined together to enhance the technology of

breast cancer detection. Machine learning technologies can potentially be used to support the interpretation

of thermal images and help physicians to automatically determine the locations and sizes of tumors, blood

perfusion, and other patient-specific properties of breast tissues. In this study, we aim to develop CNN

techniques for intelligent precision breast tumor diagnosis. The main innovation of our work is the use of

breast thermograms from a multicenter database without preprocessing for binary classification. The results

presented in this paper highlight the usefulness and efficiency of deep learning for standardized analysis of

thermograms. It is found that the model developed can have an accuracy of 80.77%, sensitivity of 44.44 %

and the specificity of 100%.

1 INTRODUCTION

Breast cancer is one of the most serious health

problems with possible fatal consequences for

women in modern times. The risk factors and causes

could include changes in the cell genome, hormonal

dysfunction, family history, hormone therapy,

lifestyle features and undesirable life habits (Francis,

2017; Singh, 2020; NCI, 2014; WHO, 2014).

Currently the mainstream methods of breast

cancer diagnosis include ultrasound, mammography

and MRI. The mammograms also known as gold

standard method of breast cancer diagnosis.

However, it is mainly recommended for women over

https://orcid.org/0000-0002-9555-7818

https://orcid.org/0000-0001-6246-9494

https://orcid.org/0000-0002-9574-4787

https://orcid.org/0000-0002-0419-1858

https://orcid.org/0000-0002-5701-1080

https://orcid.org/0000-0001-7904-0870

40 years old as an invasive method which uses

ionizing radiation (X-rays). Ultrasound is

recommended as a first test to detect whether a lump

is a cyst filled with liquid or a solid tumor. Ultrasound

examination uses high-frequency acoustic waves. It is

useful for younger patients as their mammary glands

have denser structure. However the success of

ultrasound diagnosis critically depends on the

experience of the specialist conducting the test. MRI

is the most effective and accurate method to diagnose

breast cancer or tumor. But up to now this is the most

expensive method, which could only be found in large

and well equipped hospitals.

The existing methods of diagnosis are not

appropriate for regular mass screening in short

598

Aidossov, N., Mashekova, A., Zhao, Y., Zarikas, V., Ng, E. and Mukhmetov, O.

Intelligent Diagnosis of Breast Cancer with Thermograms using Convolutional Neural Networks.

DOI: 10.5220/0010920700003116

In Proceedings of the 14th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2022) - Volume 2, pages 598-604

ISBN: 978-989-758-547-0; ISSN: 2184-433X

intervals. In addition, they are not suited for regular

breast self-examination (BSE) as promoted by WHO

(Sung, 2021; WHO, 2019) for ultimate minimization

of breast cancer fatalities. Thus, a lot of breast cancer

cases are diagnosed in late stages, although early

diagnosis is very important for effective treatment

with good prognosis (NCI, 2014).

One of the non-invasive and low cost method for

mass and regular screening is thermography. It is well

known that body temperature is an accurate indicator

of some disorder in the human body. The body

temperature distribution depends on such factors as:

blood perfusion, metabolic rate and ambient

temperature. Any abnormality in the body such as

tumor could be detected by thermography (Melal,

2016; Kandlikar, 2017; Ito, 2019).

Recent studies (Sing, 2019; Kandlikar, 2017;

Jiang, 2010; Bezerra, 2020; Saniei, 2016;

Omranipour, 2016; Sarigoz, 2020; Zeng, 2020)

discovered fast quality improvement of thermal

cameras as well as distinct development of machine

learning techniques that can be used to enhance the

technology of breast cancer detection. Machine

learning algorithms in principle, can be used to

support the interpretation of thermal images and help

physicians to automatically make diagnosis and even

to determine the locations and sizes of tumors, blood

perfusion, and other patient-specific properties of

breast tissues (Khan, 2018).

One recently developed image recognition method

is the so called convolutional neural network (CNN),

which is a deep-learning neural network system that

processes input images by extracting specific training

weights and biases to determine meaningful

characteristics that distinguish one input image from

another (Khan, 2018). Thereby, a diagnosis tool can

be built using CNNs in order to classify “healthy” and

“with-tumor” types of thermograms without any

human experts’ intervention.

In early studies of CNN the segmentation was an

important part of image recognition. There are a lot

of studies dedicated to different methods of image

recognition among them studies (Antonini, 2015;

Dayakshini, 2015; Kermani, 2015; Mahmoudzadeh,

2015; Diaz-Cortes, 2018; Etehad, 2010; Golestani,

2014). Study by Antonini et al. estimated the ability

of thermogram diagnose multicentric or multifocal

breast carcinomas (Antonini, 2017). Study by

Dayakshini segmented thermograms by using

projection profile method and by asymmetry analysis,

comparing the left and right breasts (Dayakshini,

2015). Study by Kermani used Gaussian mixture

model segmentation method (Kermani, 2015). Study

by Mahmoudzadeh used the novel method of Hidden

Markov Model to optimize the segmentation

(Mahmoudzadeh, 2015). Study by Diaz-Cortes

considers the spatial information of the pixel

contained in the image for the segmentation (Diaz-

Cortes, 2018). Further studies (Etehad, 2010;

Golestani, 2014) used and compared k-means, fuzzy

c-means and level set segmentation method to find

out the most accurate.

The classification of the breast images were done

by using feedforward neural network and radial basis

function classification (Ng, 2007). In addition one of

the common method of classification was Support

Vector Machine (SVM), which proved its

effectiveness in different studies (Madhu, 2016;

Milosevic, 2014). Other popular methods of Neural

Network classification are k-nearest neighbors

method and fast fuzzy c-mean method, used in the

studies (Milosevic, 2014; Gaber, 2015).

Our study is one of the next step in the

development of CNN and Thermography. The study

develops an efficient CNN model which uses breast

thermograms for binary classification. The main

innovation of the current work is the use of breast

thermograms with multi-view images from a

multicenter database without preprocessing for

binary classification. The results highlight the

usefulness of deep learning for standardized analysis

of thermograms.

2 MATERIALS AND METHODS

The public Visual Lab database (Visual Lab, 2021),

which contains about 287 thermal images, was used

to extract thermal images as input for our diagnosis

tool. However, for the present study only 76 thermal

images were selected as the most appropriate. These

thermal images were accompanied with doctors’

diagnosis and also had three views: frontal, left and

right.

In addition a second database was used which

consists of thermograms of patients obtained in the

"Multifunctional Medical Center" of the Nur-Sultan

city of Kazakhstan by the authors. 38 thermal images

most suitable for this work were selected. The

database currently includes breast thermograms for

women between the ages of 18 and 80. To protect the

privacy of patients, the nomenclature of breast

thermograms has been designed so that every image

in the database is given a distinguished name.

Temperature distributions on the breast skin

surfaces were recorded by the thermal camera IRTIS-

2000 ME, which is used for medical research and the

diagnosis of a wide range of illnesses, including

Intelligent Diagnosis of Breast Cancer with Thermograms using Convolutional Neural Networks

599

oncological diseases. Its temperature resolution for

the entire field of view is 0.02 °C and its temperature

measurement accuracy is 0.1 °C (see Figure 1).

An experimental procedure was developed,

together with the instructions for conducting the

experiment with which the doctor-oncologist was

familiarized. When conducting an experiment,

patients underwent a familiarization procedure with

the research being conducted and gave their consent

to participate in the research, since participation was

voluntary. The study was approved by the

institutional ethics committee of Nazarbayev

University AEO (identification number is

294/17062020).

Figure 1: Clinical office for collecting patient data and IR

camera IRTIS-2000 ME.

A typical breast thermogram used has three RGB

channels with a square size of 224 × 224 × 3 by

numbers of pixels. A breast thermogram should

include half the armpit to analyze the entire breast

tissue and nearby ganglion groups. The area of interest

of the breast thermogram shows a significant increase

in temperature compared to the temperature in the

adjacent area for a patient with breast tumor.

Examples of thermograms used are shown in Figure 2.

Two databases were combined to increase the size

of the dataset, since the format of the thermograms

and their images were similar. After they were mixed,

the dataset was classified into two sets: Training and

Validation ones, containing 88 and 26 images,

respectively.

Image classification is the process of classifying

images according to their visual contents. The

learning process for neural network involves

recognizing breast thermograms with a

predetermined label, for example healthy and sick.

This problem is known as supervised learning

(Simeone, 2018). Thus, in the current study, the

image set was divided into “Healthy” and “Sick” (as

shown in Figure 2). The sigmoid function is a non-

linear continuous function. Classification in CNN is

based on inference, which implies that its output can

be the entire range of x to the domain [0,1] of f(x).

Mathematically, the sigmoid function is defined by

(Sanjeev, 2017):

𝑓



𝑥

















(1)

The parameters α and β define the center and width of

the sigmoid function, respectively.

In the current study, the CNN architecture consists

of 5 layers of convolution and pooling. This is

followed by flattening and 2 fully connected layers

with the latter to obtain a binary output of probability

(Figure 3).

Figure 2: An example of segmentation and division of

thermal images into "Sick" and "Healthy".

Figure 3: Consolidated architecture of the CNN with the

display of parameters at each level.

CNN is a computational model that typically

consists of three types of neural layers: convolution,

pooling and fully connected ones. The convolution

and pooling layers extract features from the input

images, while the fully connected layer converts the

extracted features into final output, such as binary

classification. In the convolution layers have a small

grid filled with parameters called kernel as a feature

extractor, applied at each image position. One layer

ICAART 2022 - 14th International Conference on Agents and Artiﬁcial Intelligence

600

feeds its output into the next layer, thus extracted

features may progressively grow more complex as a

forward process. Then the parameters in kernels can

be optimized through gradient-based optimization

scheme in a backward propagation process called

training, which is performed so as to minimize the

difference between outputs and given labels.

A batch size of 16 was used as a standard number

of training examples utilized in one iteration of

forward/backward pass. Activation function

Rectified Linear Unit (ReLU) was used in the CNN

architecture in the convolution layer. The size of the

kernel is 3×3 pixels. First we began with a filter value

of 32 (32 filters) in the first 3 layers, and the following

two layers had the filter size of 64. After the

convolution, flattening the input after CNN layers is

standard procedure to go with and also adding the

ANN layer as well.

First callback list was defined as follows. For the

model to learn effectively it was necessary to define

the EarlyStopping function. It was used to halt the

epochs on metric of “loss” value and “patience”

value. This function was used to avoid overfitting. In

this CNN model, “loss” value is tracked and

“patience” of 3 epochs is defined. What it means is

that once the loss value reaches the minimum, and in

the next 3 iterations the value of loss increases, then

training will stop at that epoch. Another adherence is

reducing the learning rate. So, once the metric

stagnates, the learning rate reduces. Patience is 2 for

this callback, and if no improvement is detected, then

the learning rate reduces by a factor of 0.3, because in

that way loss value will decrease gradually and finally

arrive to the lowest value.

Another important parameter to define was class

weight. Since the dataset consisted mostly of patients

who had breast cancer, then it was necessary to assign

higher class weight to minority classes, so it could

learn in a balanced way from all classes.

CNN learning took 23 iterations to reach the

stopping point as mentioned above. Each iteration

took 10-11 seconds on a computer whose technical

specifications are as follows, in Table 1:

Table 1: CPU - Intel® Xeon® Silver 4210 Processor.

Total Cores 10

Total Threads 20

Max Turbo Frequency3 3.20 GHz

Processor Base Frequency2 2.20 GHz

Cache 13.75 MB

RAM size 64GB

Maximum Memory Speed 2400 MHz

In summary, the overall flow diagram of the proposed

study is presented on Figure 4.

Figure 4: The overall flow diagram of the proposed study.

3 RESULTS AND DISCUSSION

During training with stochastic gradient descent

iterations the loss gradually decreased to 0.2151,

while the accuracy of the training data increases to

80.77%%. The learning rate remains relatively low

from 30.0e-05 to 9.0e-05 as seen in Figure 5.

Figure 5: Graph of iteration versus loss, accuracy and

learning rate values.

Intelligent Diagnosis of Breast Cancer with Thermograms using Convolutional Neural Networks

601

The accuracy of prediction of the CNN model is

80.77%, as shown in Figure 6.

2/2 [==============================] – 127

ms / step – loss: 0.3695 - accuracy: 0.8077

Figure 6: Displaying the accuracy of the CNN model.

To analyze the inaccuracies, a confusion matrix was

plotted.

Figure 7: Confusion matrix showing the correctly and

incorrectly predicted test data for each class.

In the proposed CNN study, the network was

trained with thermograms with doctor’s diagnoses so

that it could later be used to give a direct diagnostic

response with a normal or abnormal decision when

fed with a thermogram without the need for manual

preprocessing and training of the features. Thus, it is

easier to analyze thermograms in a standardized way,

as opposed to studies with human judgment.

One limitation of this study is that the number of

cases for analysis is less than the amount of data

typically collected for deep learning. Although data

augmentation can partially solve this problem, this

disadvantage can be eliminated by using data

exchange structures such as those implemented in

neuroimaging (Yamashita, 2018). It can also be

solved if it is combined with physics-driven and

PINN diagnosis (Mukhmetov, 2021; Karniadakis,

2021), which we are working on.

Since thermography is intended as an adjunct to

mammography in breast cancer screening, its most

important value is sensitivity to detect the possible

presence of an abnormality (Lalkhen, 2008). The

study showed a sensitivity of 44.44 %. Along with

this, the performance of the classifier was assessed in

terms of sensitivity, specificity, and accuracy. In

general, the accuracy value was 80.77%, which can

be considered quite high, since it allows detecting

breast cancer with almost 81% confidence in

thermograms. The specificity of the classifier was

100%.

Another metric that was used is ROC (receiver

operating characteristic) curve and AUC value

(Figure 8).

A negative predicted value (NPV) indicates the

likelihood that the patient is not sick if the test is

negative. Figure 7 shows that 5 out of 26 were

incorrectly identified as healthy when in fact they are

sick. The authors acknowledge that the positive

predictive value (PPV) is low and is a metric that we

would like to improve.

Figure 8: ROC curve and AUC value.

Main innovation of our work is the use of breast

thermograms from a multicenter database without

preprocessing the images for efficient and automatic

binary classification. The results presented in this

paper highlight the usefulness of deep learning for

standardized analysis of thermograms with

efficiency. Future work will apply these algorithms in

a longitudinal study without tagged data and evaluate

their effectiveness in comparison with experts.

Furthermore, an integrated methodology that will

combine Bayesian Networks (Zarikas, 2015;

Zholdasbayeva, 2020) and CNNs will be developed

in order not only to improve the diagnosis but also to

dig out the key factors that determine a successful

diagnosis. It is important to remember that the current

deep learning methods cannot yet replace the

clinician when making a clinical diagnosis, but it can

help the clinician make more accurate diagnoses and

treatment recommendations. Furthermore, before

widespread adoption in clinical practice, deep

learning models should be tested on representative

datasets of different communities in order to solve

generalization problems for new populations.

4 CONCLUSIONS

Early detection of breast cancer remains an important

part of the fight against breast cancer. World Health

ICAART 2022 - 14th International Conference on Agents and Artiﬁcial Intelligence

602

Organization recommends regular self-examination

to detect the breast cancer at early stages. The review

of the previous studies shows that thermography is a

promising supplementary tool for breast cancer

detection at early stages. The combination of

thermography and computer technology can

considerably enhance breast cancer detection at early

stages. Modern models of neural networks have led to

an increase in the accuracy of classification of breast

cancer thermograms, especially in distinguishing

between healthy and deceased cases.

In the present study, a successful diagnosis tool is

presented using convolutional neural network (CNN)

to implement and validate the deep learning model.

This algorithm could accurately classify breast cancer

thermograms as “Healthy” and “Sick” using two

databases and utilizing multi-view images. Moreover,

our results were calculated automatically without any

image pre-processing to obtain perspective sensitivity

values, thus reducing human error and bias and

improving efficiency. Reason of that is usage of Data

Augmentation technique that is artificially enlarging

the dataset size that helps for CNN to better learn and

distinguish in binary classification. The limitation of

the present study is that the patients’ data available

for the analysis were less than the amount of data

typically collected for deep learning. In addition, the

positive predictive value (PPV) is still considered

low, which can be further improved via physics-

informed Neural Network (PINN) models in the

future which are being developed by us.

ACKNOWLEDGEMENTS

The authors are grateful to the Ministry of Education

and Science of the Republic of Kazakhstan for

financing this work through the grant for the

“Application of artificial intelligence to complement

thermography for breast cancer prediction”

(AP08857347) and Nazarbayev University for

managing the research project.

REFERENCES

Antonini S., Kolarić D., Herceg Ž., et al. “Thermographic

visualization of multicentric breast carcinoma,” in

Proceedings of the 2015, 57th International Symposium

ELMAR (ELMAR), pp. 13–16, IEEE, Zadar, Croatia,

September 2015.

Bezerra, L.A.; Ribeiro, R.R.; Lyra, P.R.M.; Lima, R.C.F.

An empirical correlation to estimate thermal properties

of the breast and of the breast nodule using

thermographic images and optimization techniques. Int.

J. Heat Mass Transf. 2020, 149, 119215.

Breast Cancer Treatment (PDQ®). NCI. 23 May 2014.

Archived from the original on 5 July 2014. Retrieved

29 June 2014. https://www.cancer.gov/types/breast/

hp/breast-treatment-pdq Accessed on May 2021

Dayakshini, K. Surekha, K. Prasad, and K. V. Rajagopal,

“Segmentation of breast thermogram images for the

detection of breast cancer: a projection profile

approach,” International Journal of Image and

Graphics, vol. 3, no. 1, pp. 47–51, 2015.

Díaz-Cortés M.-A., Ortega-Sánchez N., Hinojosa S., et al.

“A multi-level thresholding method for breast

thermograms analysis using dragonfly algorithm,”

Infrared Physics & Technology, vol. 93, pp. 346–361,

2018.

EtehadTavakol M., Sadri S., Ng E. Y. K. “Application of

K-and fuzzy c-means for color segmentation of thermal

infrared breast images,” Journal of Medical Systems,

vol. 34, no. 1, pp. 35–42, 2010.

Francis, S.V.; Sasikala, M.; Jaipurkar, S.D. Detection of

Breast Abnormality Using Rotational Thermography.

In Application of Infrared to Biomedical Sciences;

Springer: Singapore, 2017; pp. 133–158.

Francis S. V., Sasikala M., Bhavani Bharathi G., Jaipurkar

S. D. “Breast cancer detection in rotational

thermography images using texture features,” Infrared

Physics & Technology, vol. 67, pp. 490–496, 2014.

Fan, Guo-Feng & Yu, Meng & Dong, Song-Qiao & Yeh,

Yi-Hsuan & Hong, Wei-Chiang, 2021. "Forecasting

short-term electricity load using hybrid support vector

regression with grey catastrophe and random forest

modeling," Utilities Policy, Elsevier, vol. 73(C).

Golestani N., Etehad Tavakol M., Ng E. Y. K. “Level set

method for segmentation of infrared breast

thermograms,” EXCLI Journal, vol. 13, pp. 241–251,

2014.

Gaber T., Ismail G., Anter A., et al. “Thermogram breast

cancer prediction approach based on neutrosophic sets

and fuzzy c-means algorithm,” in Proceedings of the

2015 37th Annual International Conference of the IEEE

Engineering in Medicine and Biology Society (EMBC),

pp. 4254–4257, IEEE, Milan, Italy, August 2015.

Ito, K.; Asnida, A.W.; Daud, S.A.; Ng, E.Y.K. Thermal

analysis on 3D breast cancer model. In Computational

Modelling and Simulation for Biomedical

Applications; Wahab, A.S., Mohd, A.S.; Eds.; Penerbit

UTM Press: Skudai, Malaysia, 2019; pp. 165–186.

Jiang, L.; Zhan, W.; Loew, M.H. Modeling static and

dynamic thermography of the human breast under

elastic deformation. Phys. Med. Biol. 2010, 56, 187–

202, doi:10.1088/0031-9155/56/1/012.

Kandlikar, S.G.; Perez-Raya, I.; Raghupathi, P.A.;

Gonzalez-Hernandez, J.-L.; Dabydeen, D.; Medeiros,

L.; Phatak, P. Infrared imaging technology for breast

cancer detection – Current status, protocols and new

directions. Int. J. Heat Mass Transf. 2017, 108, 2303–

2320, doi:10.1016/j.ijheatmasstransfer.2017.01.086.

Khan S., Rahmani H., Shah S. A. A., Bennamoun M. A

guide to convolutional neural networks for computer

Intelligent Diagnosis of Breast Cancer with Thermograms using Convolutional Neural Networks

603

vision // Synth. Lect. Comput. Vis., 2018 – Vol. 8, No.

1 – P. 1 – 207.

Kermani S., Samadzadehaghdam N., EtehadTavakol M.

“Automatic color segmentation of breast infrared

images using a Gaussian mixture model,” Optik, vol.

126, no. 21, pp. 3288–3294, 2015.

Karniadakis, G.E., Kevrekidis, I.G., Lu, L. et al. Physics-

informed machine learning. Nat Rev Phys 3, 422–440

(2021). https://doi.org/10.1038/s42254-021-00314-5.

Lalkhen A.G., McCluskey A. Clinical tests: sensitivity and

specificity. Contin Educ Anaesth Crit Care Pain. 2008;

8: 221–223.

Melal, I.; Kengne, E.; el Guemhioui, K.; Lakhssassi, A. 3D

Modelling using the finite element method for

directional re-moval of a cancerous tumor. J. Biomed.

Sci. 2016, 5, 1–8.

Mahmoudzadeh E., Montazeri M. A., Zekri M., Sadri S.

“Extended hidden Markov model for optimized

segmentation of breast thermography images,” Infrared

Physics & Technology, vol. 72, pp. 19–28, 2015.

Madhu H., Kakileti S. T., Venkataramani K., Jabbireddy S.

“Extraction of medically interpretable features for

classification of malignancy in breast thermography,”

in Proceedings of the 2016 38th Annual International

Conference of the IEEE Engineering in Medicine and

Biology Society (EMBC), pp. 1062–1065, IEEE,

Orlando, FL, USA, August 2016.

Milosevic M., Jankovic D., Peulic A. “Thermography based

breast cancer detection using texture features and

minimum variance quantization,” EXCLI Journal, vol.

13, pp. 1204–1215, 2014.

Mukhmetov, O.; Mashekova, A.; Zhao, Y.; Midlenko, A.;

Ng, E.Y.K.; Fok, S.C. Patient/Breast-Specific

Detection of Breast Tumor Based on Patients’

Thermograms, 3D Breast Scans, and Reverse Thermal

Modelling. Appl. Sci. 2021, 11, 6565.

https://doi.org/10.3390/app11146565.

Ng E. Y. K., Acharya U. R., Keith L. G., Lockwood S.

“Detection and differentiation of breast cancer using

neural classifiers with first warning thermal sensors,”

Information Sciences, vol. 177, no. 20, pp. 4526–4538,

2007.

Omranipour, R.; Kazemian, A.; Alipour, S.; Najafi, M.;

Alidoosti, M.; Navid, M.; Alikhassi, A.; Ahmadinejad,

N.; Bagheri, K.; Izadi, S. Comparison of the Accuracy

of Thermography and Mammography in the Detection

of Breast Cancer. Breast Care 2016, 11, 260–264,

doi:10.1159/000448347.

Saniei, E.; Setayeshi, S.; Akbari, M.E.; Navid, M.

Parameter estimation of breast tumour using dynamic

neural network from thermal pattern. J. Adv. Res. 2016,

7, 1045–1055.

Sarigoz, T.; Ertan, T. Role of dynamic thermography in

diagnosis of nodal involvement in patients with breast

cancer: A pilot study. Infrared Phys. Technol. 2020,

108, 103336, doi:10.1016/j.infrared.2020.103336.

Singh, D.; Singh, A.K. Role of image thermography in early

breast cancer detection- Past, present and future.

Comput. Methods Programs Biomed. 2020, 183,

105074, doi:10.1016/j.cmpb.2019.105074.

Simeone O. A brief introduction to machine learning for

engineers // Found. Trends Signal Process., 2018 – Vol.

12, No. 3 – 4 – P. 200 – 431.

Sung, H.; Ferlay, J.; Siegel, R.L.; Laversanne, M.;

Soerjomataram, I.; Jemal, A.; Bray, F. Global cancer

statistics 2020: GLOBOCAN estimates of incidence

and mortality worldwide for 36 cancers in 185

countries. CA Cancer J. Clin. 2021, 71, 209–249,

doi:10.3322/caac.21660.

Sanjeev K., Mahesh Ch. Detection of Microcalcification

Using the Wavelet Based Adaptive Sigmoid Function

and Neural Network. Journal of Information Processing

Systems, vol. 13, no. 4, pp. 703–715, Aug. 2017.

https://doi.org/10.3745/JIPS.01.0007

Visual Lab DMR database. Available online

http://visual.ic.uff.br/dmi/

World Cancer Report 2014; World Health Organization:

Geneva, Switzerland, 2014; Chapter 5.2, ISBN 978-92-

832-0429-9.

World Health Organization. Breast Cancer: Prevention and

Control; World Health Organization, Geneva 2019.

Yamashita R., Nishio M., Do R. K. G., Togashi K.

Convolutional neural networks: An overview and

application in radiology // Insights Imag., 2018 – Vol.

9, No. 4 – P. 611 – 629.

Zeng, J.; Lin, L.; Deng, F. Infrared thermal imaging as a

nonradiation method for detecting thermal expression

characteristics in normal female breasts in China.

Infrared Phys. Technol. 2020, 104, 103125,

doi:10.1016/j.infrared.2019.103125.

Zarikas, V., Papageorgiou, E., Regner, P. Bayesian network

construction using a fuzzy rule based approach for

medical decision support (2015) Expert Systems, 32

(3), pp. 344-369.

Zholdasbayeva M., Zarikas, V., Poulopoulos, S. Bayesian

networks based policy making in the renewable energy

sector (2020) ICAART 2020 - Proceedings of the 12th

International Conference on Agents and Artificial

Intelligence, 2, pp. 453-462.

ICAART 2022 - 14th International Conference on Agents and Artiﬁcial Intelligence

604