2023 International Conference on Computer and Applications (ICCA)

A Comprehensive Review of Breast Cancer Early

Detection Using Thermography and Convolutional
Neural Networks
Ahmed Sharaf Eldin
Basant Ali Sayed Doaa Saad Elzanfaly
dept. Information System
dept. Information System dept. Information System
Faculty of Computers & Artificial
Higher Institute of Qualitative Studies, Faculty of Computers & Artificial
Intelligence, Helwan University
Heliopolis Intelligence, Helwan University
Cairo, Egypt
Cairo, Egypt Cairo, Egypt
[email protected]
[email protected] [email protected]
Amr S. Ghoneim
dept. Computer Science
Faculty of Computers & Artificial
Intelligence, Helwan University
Cairo, Egypt
[email protected]

Abstract— Breast cancer is a frequently diagnosed Researchers have achieved promising results using CNNs for
malignancy in the female population, with a rising trend in its breast cancer diagnosis, but their use in thermal imaging has
occurrence. A diverse range of breast cancer screening not been widely implemented in the past. The better
techniques exists, each with unique merits and drawbacks. effectiveness of CNNs compared to textural or statistical
Numerous researchers have devised automated techniques features or the substantial computational demands associated
using deep learning algorithms due to their efficacy and with CNNs may explain this success.
accuracy in forecasting the multiplication of cancer cells via
the use of Thermography technologies. Thermography is Breast cancer diagnosis has traditionally relied on
regarded as a non-invasive and safe technique for the diagnosis mammography, ultrasound, and MRI, but due to limited
of breast cancer. This study provides a thorough examination accessibility in rural areas, alternative methods like
of the research conducted on the analysis of thermal electronic palpation, electrical impedance scanning, and
fluctuations in the breast and the diagnostic approaches used thermal imaging (thermography) have emerged [6].
by different researchers in the detection of breast cancer. Our Thermography, also known as infrared imaging, records
objective is systematic review to conduct a comprehensive temperature distributions of the breast using infrared
examination of the most notable scholarly works concerning cameras. This method can be used to determine the presence
the use of deep learning (DL) models and the datasets of a tumor by detecting increased temperature on the breast
employed in the categorization of breast cancer. Additionally,
surface. Thermography is based on the idea that uncontrolled
we address the challenges and constraints that give rise to the
cellular growth leads to a higher metabolic rate, requiring
most pressing areas of inquiry for future study in this
particular sector.
more blood perfusion. This surplus heat is then directed to
surrounding tissue, causing a rise in breast surface
Keywords— Thermography, Deep Learning, Breast Cancer temperature. This method can be used to identify increased
temperature in the presence of a tumor. This approach is
cost-effective, non-intrusive, and radiation-free [7, 8, 9].
I. INTRODUCTION
Breast cancer is a prevalent form of cancer, with an Breast cancer detection involves thermography, a method
estimated 2.3 million new diagnoses expected in 2020. From that identifies thermal patterns in the breast over a specific
2016 to 2020, 7.8 million women received a breast cancer period. Common features include unequal temperature
diagnosis. The World Health Organization reports a differences between left and right breasts, specific regions of
cumulative number of 685,000 fatalities associated with increased temperature as indicators of abnormalities, changes
breast cancer, compared to other forms like lung, liver, and in hypothermic vascular patterns due to tumor growth, and
stomach cancer. This lower mortality rate suggests that variations in heat patterns within the areolar and periareolar
breast cancer is responsive to therapeutic interventions. The regions [10, 11, 12].
diagnosis of breast cancer is based on tumor stages, which The rest of this paper is organized as follows: Section II
can be identified by factors like tumor size or cancer cell is about thermography in the diagnosis of breast cancer.
presence in lymph nodes [1]. Prompt and accurate detection Section III Deep Learning for Breast cancer Thermogram
is crucial for effective management, a subject with classification. Section IV review on Breast Cancer using
significant scholarly focus [2, 3]. The study supports the idea Thermographs and Deep Learning. Finally, Section V
that breast cancer is responsive to therapeutic interventions. provides the research study conclusion and future work.
Recent studies have focused on using machine learning
methods for breast cancer diagnosis using thermography. II. THERMOGRAPHY IN THE DIAGNOSIS OF BREAST
Some studies have examined tumor size and locations, while CANCER
others examine collection techniques and breast quadrants.
Deep learning, a computational learning methodology, uses
multilayer convolutional neural networks (CNNs) to Thermography has many identifiable properties. One
independently derive characteristics from training data [4, 5]. noteworthy feature is its capacity to identify

979-8-3503-0325-4/23/$31.00 ©2023 IEEE 120

 2023 International Conference on Computer and Applications (ICCA)

early indicators of cancer, facilitating the identification of III. DEEP LEARNING TECHNIQUES IN THE
cancer ten years earlier in comparison to other methods CLASSIFICATION OF THERMOGRAMS FOR BREAST
CANCER.
A neural network (NN) is a parallel processor with
simple units that efficiently processes complex information.
It allows for subjective values to be acquired in medical
picture analysis, enabling conclusions about breast
thermogram characterization.
Convolutional neural networks (CNNs) are a deep neural
network technique used to process images. The proposed
methodology uses learning-adjusted parameters like weights
and biases to recognize unique features in photographs. The
output of categorizing breast thermograms into three groups
can be seen as the resultant. This technique is widely used in
Fig1: THERMAL IMAGES CLASSIFICATION BASED ON: (A) TH1- deep neural networks. Unique groups, namely normal,
REPRESENTS NORMAL UNIFORM NON-VASCULAR SAMPLES, (B)
TH2- REPRESENTS VASCULAR ORDINARY UNIFORM SAMPLES, benign, and malignant, is supported by the convolutional
(C) TH3- REPRESENTS SAMPLES THAT ARE AMBIGUOUS OR neural networks (CNNs) shown in Fig 2. Three primary
DOUBTFUL, (D) TH4- ABNORMAL, (E) TH5- REPRESENTS aspects that need prioritization are the readiness of the
SAMPLES THAT ARE SEVERELY ABERRANT. [22] dataset in relation to picture pre-processing, the feature
Such as mammography [13, 14]. The distinguishing learning process, and the classification objective. The
feature of thermography is in its ability to assess various categorization may be divided into three distinct categories,
modalities of cancer therapy. An additional attribute of specifically referred to as "normal, cancer, and malignant."
thermography is its capacity to predict the prospective
condition of the individual [15].
Breast cancer has been identified in 44% of persons who
have shown aberrant thermograms, about five years after the
use of thermography. Based on the thermography data, it has
been shown that individuals who have been diagnosed with
advanced malignancy have a 24% chance of living within a
three-year timeframe. However, when malignancies are
found at earlier stages, the probability of survival increases
significantly, reaching almost 80% [16].
The efficacy of mammography is diminished in younger
individuals and those with a greater quantity of breast tissue.
Factors such as the age of the patient or the density of breast
tissue do not have any impact on thermography [17, 18].
Based on a research, it has been shown that thermography
has the ability to detect indications of breast cancer up to FIG 2: PIPELINE OF CONVOLUTIONAL NEURAL NETWORKS FOR
twelve months earlier than mammography in around 70% of BREAST CANCER DETECTION USING THERMOGRAMS.
cases [13]. The size of undetectable lesions, as measured by In the following sections, we shall analyze the principles
mammography, is around 1.66 cm. However, thermography and corresponding efforts in the use of Convolutional Neural
imposes a more rigorous constraint, limiting the size to 1.28 Networks (CNN) for the categorization of breast
cm [19]. thermograms.
Comparing an abnormal thermogram to a normal
thermogram may serve as a means to identify a significant A. Preprocessing of Images
22-fold increase in the chance of developing breast cancer. Preprocessing is a crucial step in image analysis,
Furthermore, the anomalous thermogram has a level of enhancing the quality and utility of images by reducing noise
significance and influence that is tenfold greater than the and enhancing visual features. It is essential for neural
illness history of the first family member [20]. The field of network algorithms, as the learning process relies on
thermobiology classifies the examination of thermal images extracting relevant features from input images. Preprocessing
into five main categories: The classification of samples into techniques include mean subtraction, standardization, PCA
several categories is as follows: TH1 represents normal brightening, and neighborhood contrast standardization [22].
uniform non-vascular samples, TH2 represents vascular In the processing of breast thermograms, a variety of
ordinary uniform samples, TH3 represents samples that are algorithms are used, including scaling, segmentation of the
ambiguous or doubtful, TH4 represents abnormal samples, "Region of Interest" (ROI), and expansion.
and TH5 represents samples that are severely aberrant. Fig 1
[23] displays the divided and classed pictures of these Pramanik [27] segmentation process involves two phases:
samples. using the MSPSF energy functional and MSPSF-based level
set techniques. The MSPF energy functional-based level set
approach estimates area intensity distribution, generating
discontinuities. The level set approach, consisting of two
parts, automatically initializes breast thermograms' intensity

121
 2023 International Conference on Computer and Applications (ICCA)

distribution and reduces energy consumption. This subset of fully connected layers processes these features,
segmentation approach effectively decreases calculation generating the network's final outputs. These outputs
time. represent probabilities associated with each class in
classification tasks [37].
B. Convolutional Neural Networks (CNNs)
a) Feature Extraction Layer: The feature extraction IV. BREAST CANCER DIAGNOSIS USING
layer encodes input data, specifically photographs, to THERMOGRAPHS AND DEEP LEARNING
generate features. A single picture is encoded as a "feature In a cohort study conducted in the year 2020. The
map" with numerical values describing the image's visual aforementioned study included the categorization of
attributes. This layer consists of convolution and pooling thermography breast pictures into two distinct groups,
components. The convolution layer is organized as a kernel, namely "normal" and "abnormal" [32]. Despite its accuracy
and the breast thermogram has three shading channels (red, rate of 85%, with a sensitivity of 87% and a specificity of
green, and blue). Each channel requires a matching filter. 83%. The research conducted by [61] included a comparison
The convolution layer produces a feature map, which is then of several classifiers, including AdaBoost, Support Vector
passed to the pooling layer. The pooling layer is composed Machines (SVM), k-Nearest Neighbors (KNN), Naive
of a single channel with a pre-determined step size. The Bayes, and neural network probability. The frontal breast
pictures demonstrated the most significant outcomes when
convolutional layer conducts a more comprehensive
the genetic algorithm (GA) reduction technique was
examination of feature maps, while the pooling layer
integrated with the AdaBoost classifier. This combination
conducts a less exhaustive analysis. Two common pooling resulted in an average accuracy of 85.5% and 87.5% for the
techniques are max pooling and average pooling. corresponding experiments. Following that, more
Inside the realm of "learning features," Convolutional methodologies were established, as seen in Table 1.
Neural Networks (CNNs) engage in the replication of
convolution and maximum pooling operations iteratively, TABLE 1: WORKS ON BREAST THERMOGRAM CLASSIFICATION
USING MACHINE LEARNING.
with the ultimate objective of perceiving the distinctive
characteristics or "features" present inside pictures. Figure 3 ref
Methodology Result
illustrates the convolution process of a Convolutional Neural Classifier Feature Acc Sen Spe
Network (CNN), using breast thermograms as the input [28] SVM GLCM 95% 97% 98.6%
Texture and structural
pictures. Given that the inputs are composed of three [33] FANN
features.
88.5% 87% 89%
channels representing the red, green, and blue components [34] GA Statistical features 90.1% 89.3% 91 %
(RGB), it follows that the kernel volume is comprised of
three distinct "2D kernels." Each individual channel was [39] LSSV (RLBP) 89% - -
found to have a correlation with a certain kernel. The M
quantity of feature maps is indicative of the dimensions of Texture and shape
[43] ELM feature from Haralick 88% - -
the kernel. Regarding map characteristics.
and Zernike extractors
[45] SVM Daubechies Wavelet 97.1% - -
[46] SVM Statistical features 89.7% 75% 90.6 %
Texture and statistical
[48] SVM 100% - -
features
[49] RF GLRLM and GLCM 95.4% - 88%
texture features and
[50] ANN 98.9% - -
statistical analysis
Texture feature
[52] KNN 99.2% 98.4% 100%
extraction
LSSV Texture feature
[54] 96% 100% 92%
M extraction
[55] SVM GLCM 97.1% 98% 92.3%
Texture and Wavelet
[56] SVM 91% 87.2% 94.3%
features
Texture and structural
[58] KNN 87.5% 89.1% 93.3%
FIG 3: VISUALIZATION OF CONVOLUTIONAL PROCESS OF features.
Histogram of oriented
BREAST THERMOGRAM [40]. [60] DT 98% 96.6% 100%
gradients

b) Fully-Connected Layer: Convolutional layers In the year 2020, Mishra made applications to many
transform feature maps into a vector for input in a fully classifiers like SVM, decision tree, random forest, K-NN,
connected layer, with weights in a dense layer determining linear regression, and fuzzy logic classifiers [45].The
the association between input and output. Pooling layers research found that random forest with principal component
refine features, processed by a subset of fully connected analysis (PCA) achieved the highest accuracy of 95.45%,
layers, to generate outputs [37]. while the Support Vector Machine (SVM) classifier was
C. Images Classification used [44].
The convolutional layer transforms feature maps into a The research aims to achieve maximum accuracy in
unidimensional array for input in a fully connected layer. breast thermal medical imaging classification using a deep
The association between input and output is determined by neural network model. This decision was made to improve
adjustable weights in a dense layer. Convolutional layers the performance of neural network methodologies using deep
establish features and refine them through pooling layers. A learning techniques. Table 2 presents a comprehensive

122
 2023 International Conference on Computer and Applications (ICCA)

overview of the use of convolutional neural networks in the and DIT), Image Dimensions (3D, 2D), number image in
context of breast cancer diagnosis, specifically focusing on each class (Total (T) and Database Used. However, prior
breast thermogram data. studies were limited to altering the quantity of thermal
pictures mostly obtained from the DMR-IR database.
A unique methodology, referred to as "InceptionV3-
KNN" and including a module named "Stage Cancer," was TABLE 3: COMPARISON OF RELATED WORK USING
used in a research [64]. Was used to classify breast cancer THERMOGRAPHIC IMAGES
into four phases: T1 (non-invasive breast illness), T2 (tumor General Image Image Num
Camera
size of 2 cm), T3 (tumor size exceeding 5 cm), and T4 Ref
type
Goal Protocol Dimension image Dataset
(malignant growth covering the whole breast). The study Scr Diag Sit Dit 3D 2D T
utilized the "Linear Support Vector Machine" and [28] SC620 - × × × - × 193 DMR-IR
[43] SC620 × - - - - × 23 DMR-IR
"Convolutional Neural Network" models to determine the [23] SC620 - × × × × - 3411 DMR-IR
decision boundary [65], effectively optimizing equipment SC620 80 DMR-IR
utilization using a powerful classifier in medical practice. [40] - × × × × -
S45 98 UFPE
[39] SC620 × - × × × - 107 DMR-IR
A separate research investigation was conducted to [58] SC620 × - - - × - - DMR-IR
examine the performance of several convolutional neural [59] SC620 - × - × × - 208 DMR-IR
network architectures, including ResNet101, DenseNet, [56] SC620 - × - - × × 1874 DMR-IR
MobileNetV2, and ShuffleNetV2. The performance of these [57] SC620 - × - × × - 1874 DMR-IR
models was assessed using the ImageNet database, and they [38] SC620 - × - - - × 336 DMR-IR
showed exceptional capabilities in picture categorization. [41] SC620 - × - × - × 196 DMR-IR
Thermal breast pictures obtained from the Database of [32] SC620 - × - - - × 1345 DMR-IR

Mastology Research (DMR) were used for training the [45] SC620 - × - - - × 56 DMR-IR
[44] SC620 - × - × - × 68 DMR-IR
model. [62] SC620 - × - - - × 57 DMR-IR
In the year 2021, Torres used a deep convolutional neural [63] SC620 - × - × - × 681 DMR-IR

network (CNN) using transfer learning techniques to [61] SC620 - × - × - × 140 DMR-IR
DBT-TU-
autonomously categorize thermograms into two distinct T650sc × - - - - × 70 JU
[67]
categories, namely normal and aberrant. The ResNet-101 SC620 × - - - - × 80 DMR-IR
model, which was trained via transfer learning, exhibited a
sensitivity of 92.3% and a specificity of 53.8% [55]. Other V. CHALLENGES OF BREAST CANCER IN EARLY
CNN models that were included in the study include DTECTION USING THERMALGRAPHY AND DEEP LEARNING
inception V4, inception MV4 [56], deep convolution neural Breast cancer is a substantial global health issue,
network (DCNN) [57], and inception V3 [58]. necessitating the utmost importance of timely identification
TABLE 2: WORKS ON BREAST THERMOGRAM CLASSIFICATION in order to enhance patient prognoses. Thermography and
USING DEEP LEARNING Convolutional Neural Networks (CNNs) have been
investigated as two potential approaches for early detection.
Methodology Result
Ref
Classifier Feature Acc Sens Spec
Nevertheless, three approaches encounter distinct obstacles.
Matrix feature Thermography is a non-invasive technique used for the
[23] ShuffleNet 100% 100% 100%
extraction
[24] CNN Texture feature 100% - 100% early identification of breast cancer, which does not need the
extraction administration of any sort of energy into the human body
[32] ELM Geometrical and 100% - - [50]. Despite its considerable potential, a primary obstacle
Textural Feature associated with thermography lies in its inherent inability to
Extraction
[33] V-Net High -level Feature 100% - -
discern the underlying factors contributing to elevated
[34] V-net High-level features 99.6% 99.6% - thermal patterns. The presence of warmth in the breast region
[35] C-DCNN
Automatic Feature
- - -
might potentially serve as an indicator for the presence of
Selection breast cancer. However, it is important to note that this
[55] ResNet-101 - 0.749 92.6% 53.8% symptom can also be associated with non-cancerous
Inception
[58] V3
statistical features 85% - - conditions, such as mastitis [83]. The presence of
ResNet50 Hu Moments 96% - - inflammation in the breast caused by bacterial or viral
Sparse Latent Space
- -
infection might lead to an elevation in tissue temperature,
Deep Thermomis hence influencing the outcomes of thermography.
[59] SPAER 78.2%
Low-Rank Matrix
- -
Approximation According to Pramanik’s study that was conducted in
98.5 - - 2016, from Advantage the study, utilized a high-resolution
[60] InceptionV3 Extraction features
%
[61] CNN Extraction features 98.9% - - thermal camera [33], [84], [87] to achieve a high level of
Texture and classification accuracy [26], [59], [79] [81], [88], [74],
[62] CNN 92% 91% 94%
statistical features [81].The sensitivity ratio has a high. Disadvantage of these
studies, used Sensitivity, Sample size, accuracy, AUC are not
Previous study has shown the considerable endeavors of given. Sample size is small Thermal camera has low
researchers in using several artificial neural network (ANN) resolution[38], using sample size is small
models, and more recently, deep learning techniques, for the dataset[26],[32],[33],[38],[52],[54],[57],[58],[59],[60],[66]
purpose of breast cancer diagnosis. Table 3 displays the
pertinent information of the published studies, encompassing Recent advancements in the field of medical technology
the year of publication, Camera type, overarching objective have led to the development of Niramai's Thermalytix
(screening and diagnosis), image acquisition protocol (SIT solution in India. This approach relies on the identification of

123
 2023 International Conference on Computer and Applications (ICCA)

cancer cells exhibiting heightened activity, which 2017].

subsequently leads to the generation of localized regions [5] Gaber T and Hassanien, AE. Digital mammography image analysis
system based on mathematical morphology. In: IEEE computer society
with elevated temperatures. This phenomenon is mostly 7th Iternational Conference on Intelligent Engineering Systems INES;
attributed to the expansion or dilatation of blood vessels, as Assiut-Luxor, Egypt. 2003.
well as the formation of novel vascular lesions. NIRAMAI is [6] S.G. Kandlikar, I. Perez-Raya, P.A. Raghupathi, J.S. Gonzalez-
an acronym that stands for Non-Invasive Risk Assessment by Hernandez, D. Dabydeen, L. Medeiros, P. Phatak. s.l. Infrared Imaging
Machine Intelligence [84]. This integrated system for breast technology for breast cancer detection – Current status, protocols and
new directions. International Journal of heat and mass transfer, 2017,
health assessment and diagnostics use thermal imaging Vol. 108, pp. 2303-2320.
technology in conjunction with artificial intelligence. The [7] An improved three-dimensional direct numerical modelling and thermal
infrared camera is a convenient and comparatively affordable analysis of a female breast with tumour. E.Y.K. Ng, N.M. Sudharsan.
alternative to conventional two-dimensional mammography 2001, Proc. Inst. Mech. Eng., Vol. 215 (1), pp. 25-37. 25-37.
equipment. The utilization of this imaging technique enables [8] J.R. Keyserlingk, P.D. Ahlgren, E. Yu, N. Belliveau, M. Yassa, IEEE Eng,
the visualization of temperature fluctuations, hence revealing Med. Biol.Mag. 19 (3) (2000) 30–41.

the heat distribution inside the breast tissue. Subsequently, [9] Relationship between microvessel density and thermographic hot areas
in breast cancer.T. Yahara, T. Koga, S. Yoshida, S. Nakagawa, H.
the thermalytix artificial intelligence system is employed to Deguchi, K. Shirouzu. 2003, Surg. Today, Vol. 33 (4), pp. 243-248. 243-
evaluate these temperature changes and identify any 248.
anomalies [85]. [10] 2016 International Conference on Advances in Computing,
Communications and Informatics, ICACCI 2016. S. Pramanik, D.
Bhattacharje, M. Nasipuri. 2016. Texture analysis of breast thermogram
VI. CONCLUSION for differentiation of malignant and benign breast. pp. 8-14.

The early diagnosis of breast cancer remains a [11] Supportive Noninvasive Tool for the Diagnosis of Breast Cancer Using a
Thermographic Camera Sensor. M.A. Garduno-Ramon, S. G. Vega-
fundamental aspect of managing breast illness. Breast self- Mancilla, L.A. Morales-Henandez, R.A. Osornio-Rios. s.l. :
examination is advocated by the World Health Organization doi:10.3390/s17030497, 2017, Sensors, Vol. 17, p. 497.
as a means to enhance women's awareness on the dangers [12] Breast thermography and cancer risk prediction. Gautheria M., Gros C.M.
associated with breast cancer. Thermography is designed for 1980, Cancer, Vol. 45, pp. 51-56. 51-56.
the purpose of early detection via screening approaches, and [13] Head J, Wang F, Lipari C, Elliott R. The important role of infrared
imaging in breast cancer. IEEE Eng Med Biol Mag 2000; 19: 52-57.
it is believed to provide a possible enhancement for a self-
screening approach capable of identifying breast cancer [14] Benard A, Desmedt C, Smolina M, Szternfeld P, Verdonck M. Infrared
imaging in breast cancer: automated tissue component recognition and
during its first phases. Based on the breast thermogram spectral characterization of breast cancer cells as well as the tumor
outline, the detection of imbalanced thermal conveyances microenvironment.Analyst 2014; 139: 1044-1056.
between the breasts may serve as an early indicator for the [15] Keyserlingk JR, Yassa M, Ahlgren P, Belliveau N, editors. Preliminary
presence of breast cancer. The analysis of the imbalanced evaluation of preoperative chemo hormonotherapy-induced reduction of
the functional infrared imaging score in patients with locally advanced
thermal distribution seen on breast thermograms would be breast cancer. Engineering in Medicine and Biology Society, 2001
conducted utilizing computer-assisted innovation. The Proceedings of the 23rd Annual International Conference of the IEEE;
implementation of this invention has the potential to mitigate 2001: IEEE.
mistakes. Our analysis revealed that the existing neural [16] Kim YS, Cho YE, editors. Pre-and postoperative thermographic imaging
network models have shown enhanced performance in of lumbar disc herniations. International Symposium on Advanced
Techniques and Clinical Applications in Biomedical Thermology; 1995:
accurately classifying breast cancer thermograms, Taylor & Francis.
specifically in differentiating between normal and high-risk [17] Head JF, Wang F, Elliott RL. Breast thermography is a noninvasive
instances. However, there is a need to improve the prognostic procedure that predicts tumor growth rate in breast cancer
effectiveness of the neural networks (NNs) model. Future patients. Ann N Y Acad Sci 1993; 698: 153-158.
study should strive to enhance the categorization of breast [18] Qi H, Diakides NA. Thermal infrared imaging in early breast cancer
detection-a survey of recent research. Engineering in Medicine and
thermograms. This necessitates the provision of Biology Society, 2003 Proceedings of the 25th Annual International
representative datasets, the preparation of exceptional return Conference of the IEEE; 2003: IEEE.
on investments (ROIs), the allocation of exceptional [19] Ohsumi S, Takashima S, Aogi K, Usuki H. Prognostic value of
computational resources, and the implementation of thermographical findings in patients with primary breast cancer. Breast
"lightweight" convolutional neural network (CNN) models. Cancer Res Treat 2002; 74: 213-220.
The attainment of these goals will result in a reduction in the [20] R. Roslidar, M. Syaryadhi, K. Saddami, B. Pradhan, F. Arnia, M. Syukri
and K. Munadi, "BreaCNet: A high-accuracy breast thermogram
computational time needed for convolution calculations and classifier based on mobile convolutional neural network," Mathematical
an improvement in the accuracy rates. Subsequently, a self- Biosciences an Engineering, 2022.
breast detecting device may be developed in the first phases [21] Watmough D, Fowler PW, Oliver R. The thermal scanning of a curved
without the need for physical integration, using a isothermal surface: implications for clinical thermography. Phys Med Biol
thermography-based technique for free detection. 1970; 15: 1.
[22] S. Khan, H. Rahmani, S. A. A. Shah, and M. Bennamoun, ``A guide to
convolutional neural networks for computer vision,'' Synth. Lect. Comput.
REFERENCES Vis., vol. 8, no. 1, pp. 1207, Feb. 2018.
[1] Institute, National Cancer. National Cancer Institute. [Online] NIH [23] Thermobiological 2015. Available from: http://
Turning Discovery into Health. [Cited: August 2, 2020.] www.breastthermography.com/ breast_thermography_proc.htm.
https://www.cancer.gov/types/ breast/patient/breast-treatment- [24] S. Tello-Mijares, F. Woo and F. Flores, "Breast Cancer Identification via
pdq#section/all. Thermography Image Segmentation with a Gradient Vector Flow and a
[2] World Health Organization (2021). Breast Cancer. Retrieved from Convolutional Neural Network," Journal of Healthcare Engineering, 2019.
https://www.who.int/news-room/fact-sheets/detail/breast-cancer. [25] D. Sathish, S. Kamath, K. Prasad, R. Kadavigere and R. J. Martis,
[3] A.B. Mariotto, K. Robin Yabroff, Y. Shao, E.J. Feuer, M.L. Brown, JNCI, "Asymmetry analysis of breast thermograms using automated
J. Natl.Cancer Inst 103 (2) (2011) 117–128. segmentation and texture features," springer link, 2016.
[4] Wehle H. Machine learning, deep learning, and AI: What’s the [26] R. R. Devi and G. S. Anandhamala, "Analysis of Breast Thermograms
difference? [Online]. 2017. Available: Using Asymmetry in Infra-Mammary Curves," Journal of Medical
https://www.researchgate.net/publication/31890021_Machine_Learning_ Systems, 2019.
Deep_Learning_and_AI_What’s_the_Difference (2007). [Accessed July [27] S. Pramanik, D. Bhattacharjee and M. Nasipuri, "MSPSF: A Multi-Scale

124
 2023 International Conference on Computer and Applications (ICCA)

Local Intensity Measurement Function for Segmentation of Breast [49] S. Min, J. Heo, Y. Kong, Y. Nam, P. Ley, B.-K. Jung, D. Oh and W. Shin,
Thermogram," IEEE Transactions on Instrumentation and Measurement, "Thermal Infrared Image Analysis for Breast Cancer Detection," ksii
2019. transactions on internet and information systems, 2017.
[28] L. J. K. S. C. M. H. M. Mathew Jose Mammoottil, "Detection of Breast [50] U. R. Gogoi, M. K. Bhowmik, A. K. Ghosh, D. Bhattacharjee and G.
Cancer from Five-View Thermal Images Using Convolutional Neural Majumdar, "Discriminative Feature Selection for Breast Abnormality
Networks," Hindawi Journal of Healthcare Engineering, p. 15, 2022. Detection and Accurate Classification of Thermograms," in International
[29] V. Lessa and M. Marengoni, "Applying Artificial Neural Network for the Conference on Innovations in Electronics, Signal Processing and
Classification of Breast Cancer Using Infrared Thermographic Images," Communication (IESC), 2017.
in Computer Vision and Graphics, 2016. [51] S. Pramanik, D. Bhattacharjee and M. Nasipuri, "Texture Analysis of
[30] S. Pramanik, D. Banik, D. Bhattacharjee, M. Nasipuri, M. K. Bhowmik Breast Thermogram for Differentiation of Malignant and Benign Breast,"
and G. Majumdar, "Suspicious-region segmentation from breast in Intl. Conference on Advances in Computing, Communications and
thermogram using DLPE-based level set method," IEEE Transactions on Informatics (ICACCI), 2016.
Medical Imaging, 2018. [52] U. Raghavendra, U. R. Acharya, E. Y. K. Ng, J.-H. Tan and A. Gudigar,
[31] D. Sánchez-Ruiz, I. Olmos-Pineda and J. Olvera-López, "Automatic "An integrated index for breast cancer identification using histogram of
region of interest segmentation for breast thermogram image oriented gradient and kernel locality preserving projection features
classification," Pattern Recognition Letters, 2020. extracted from thermograms," Quantitative InfraRed Thermography
Journal, 2016.
[32] F. AlFayez, M. W. A. El-Soud and T. Gaber, "thermogram Breast Cancer
Detection: A Comparative Study of Two Machine Learning Techniques," [53] A. Lashkari, F. Pak and M. Firouzmand, "Full Intelligent Cancer
applied sciences, 2020. Classification of Thermal Breast Images to Assist Physician in Clinical
Diagnostic Applications," Journal of Medical Signals & Sensors, 2016.
[33] S. T. Kakileti, G. Manjunath and H. J. Madhu, "Cascaded CNN for View
Independent Breast Segmentation in Thermal Images," in 41st Annual [54] J. C. Torres-Galván, E. Guevara, E. S. Kolosovas-Machuca, A.
International Conference of the IEEE Engineering in Medicine and Oceguera-Villanueva, J. L. Flores and F. J. González, "Deep
Biology Society (EMBC), 2019. convolutional neural networks for classifying breast cancer using infrared
thermography," Quantitative InfraRed Thermography Journal, 2021.
[34] S. T. Kakileti, A. Dalmia and G. Manjunath, "Exploring deep learning
networks for tumour segmentation in infrared images," Quantitative [55] M. A. S. A. Husaini, M. H. Habaebi, T. S. Gunawan, M. R. Islam, E. A. A.
InfraRed Thermography Journal, 2019. Elsheikh and F. M. Suliman, "Thermal-based early breast cancer
detection using inception V3, inception V4 and modified inception MV4,"
[35] S. Guan, N. Kamona and M. Loew, "Segmentation of Thermal Breast Neural Computing and Applications, 2021.
Images Using Convolutional and Deconvolutional Neural Networks," in
Guan, S., Kamona, N., & Loew, M. (2018). Segmentation of Thermal [56] M. A. S. A. Husaini, M. H. Habaebi, T. S. Gunawan and M. R. Islam,
Breast Images Using Convolutional and Deconvolutional Neural "Self-Detection of Early Breast Cancer Application with Infrared Camera
Networks.IEEE Applied Imagery Pattern Recognition Workshop (AIPR), and Deep Learning," electronics, 2021.
2018. [57] p.kanimozhi, s.sathiya, m.balasubramanian, p.sivaguru and p.sivaraj,
[36] V. Madhavi and T. C. Bobby, "Thermal Imaging Based Breast Cancer "evaluation of machine learning and deep learning approaches to
Analysis Using BEMD and Uniform RLBP," in 3rd International classify breast cancer using thermography," psychology and education,
Conference on Biosignals, images and instrumentation (ICBSII), 2017. 2021.

[37] R. Yamashita, M. Nishio, R. K. G. Do, and K. Togashi, ‘‘Convolutional [58] B. Yousefi, H. Akbari, M. Hershman, S. Kawakita, H. C. Fernandes, C.
neural networks: An overview and application in radiology,’’ Insights Ibarra-Castanedo, S. Ahadian and X. P. V. Maldague, "SPAER: Sparse
Imag., vol. 9, no. 4, pp. 611–629, Aug. 2018. Deep Convolutional Autoencoder Model to Extract Low Dimensional
Imaging Biomarkers for Early Detection of Breast Cancer Using Dynamic
[38] A. L. R. d. Silva, M. A. d. Santana, C. L. d. Lima, J. F. S. d. Andrade, T. Thermography," applied sciences, 2021.
K. S. d. Souza, M. B. J. d. Almeida, W. W. A. d. Silva and R. D. C. F. d.
Lima, "Features Selection Study for Breast Cancer Diagnosis Using [59] S. S. Yadav and S. M. Jadhav, "Thermal infrared imaging based breast
Thermographic Images, Genetic Algorithms, and Particle Swarm cancer diagnosis usingmachine learning techniques," Multimedia Tools
Optimization," International Journal of Artificial Intelligence and Machine and Applications, 2020.
Learning, 2021. [60] S. Ekicia and H. Jawzalb, "Breast cancer diagnosis using thermography
[39] N. Pawar, P. Dixit, S. Gangwal and Y. Kewlani, "Thermal Anomaly and convolutional neural networks," Medical Hypotheses, 2020.
detection and Breast Cancer Risk Prediction: Using DITI and Deep [61] J. Zuluaga-Gomez, Z. A. Masry, K. Benaggoune, S. Meraghni and N.
Learning Approach," International Journal of Engineering Research and Zerhouni, "A CNN-based methodology for breast cancer diagnosis using
Applications, 2021. thermal images," Computer Methods in Biomechanics and Biomedical
[40] R. Resmini, L. Silva, A. S. Araujo, P. Medeiros, D. Muchaluat-Saade and Engineering: Imaging & Visualization, 2020.
A. Conci, "Combining Genetic Algorithms and SVM for Breast Cancer [62] S. Mishra, A. Prakash, S. K. Roy, P. Sharan and N. Mathur, "breast
Diagnosis Using Infrared Thermography," sensors, 2021. cancer detection using thermal images and deep learning," in 7th
[41] H. G. zadeh, B. Binazir, A. Fayazi and M. Yargholi, "Breast Cancer International Conference on Computing for Sustainable Global
Diagnosis Based on Feature Extraction Using Dynamic Models of Development (INDIACom), 2020.
Thermal Imaging and Deep Auto-encoder Neural Networks," Journal of [63] S. Mambou, O. Krejcar, P. Maresova, A. Selamat and K. Kuca, "Novel
Testing and Evaluation, 2020. Four Stages Classification of Breast Cancer Using Infrared Thermal
[42] S. Dey, R. Roychoudhury, S. Malakar and R. Sarkar, "Screening of Imaging and a Deep Learning Model," in Bioinformatics and Biomedical
breast cancer from thermogram images by edge detection aided deep Engineering, 2019.
transfer learning model," Multimedia Tools and Applications, 2021. [64] H. T. Iqbal, B. Majeed, U. Khan and M. A. B. Altaf, "An Infrared High
[43] T. A. E. d. Silva, L. F. d. Silva, D. C. Muchaluat-Saade and A. Conci, "A classification Accuracy Hand-held Machine Learning based Breast-
Computational Method to Assist the Diagnosis of Breast Disease Using Cancer Detection System," in IEEE Biomedical Circuits and Systems
Dynamic Thermography," sensors, 2020. Conference (BioCAS), 2019.

[44] V. Mishra and S. K. Rath, "Detection of breast cancer tumours based on [65] R. Roslidar, K. Saddami, F. Arnia, M. Syukri and K. Munadi, "A Study of
feature reduction and classification of thermograms," Quantitative Fine-Tuning CNN Models Based on Thermal Imaging for Breast Cancer
InfraRed Thermography Journal, 2020. Classification," in IEEE International Conference on Cybernetics and
Computational Intelligence, 2019.
[45] A. Hakim and R. N. Awale, "Thermal Imaging - An Emerging Modality for
Breast Cancer Detection: A Comprehensive Review," Journal of Medical [66] Artificial intelligence in cancer imaging: clinical challenges and
Systems, 2020. applications. Bi WL, Hosny A, Schabath MB, et al. CA Cancer J Clin.
2019;69:127–157.
[46] J. Ma, P. Shang, C. Lu, S. Meraghni, K. Benaggoune, J. Zuluaga, N.
Zerhouni, C. Devalland and Z. A. Masry, "A portable breast cancer [67] U. R. Gogoi, M. K. Bhowmik, A. K. Ghosh, D. Bhattacharjee and G.
detection system based on smartphone with infrared camera," Journal of Majumdar, "Discriminative Feature Selection for Breast Abnormality
Vibroengineering, 2019. Detection and Accurate Classification of Thermograms," in International
Conference on Innovations in Electronics, Signal Processing and
[47] J. S. Jeyanathan, A. Shenbagavalli and B. Venkatraman, "Analysis of Communication (IESC), 2017.
Transform-Based Features on Lateral View Breast Thermograms,"
Circuits, Systems, and Signal Processing, 2019.
[48] V. Madhavi and C. B. Thomas, "Multi-view breast thermogram analysis
by fusing texture features," Quantitative InfraRed Thermography Journal,
2019.

125