01

Thermography
for detecting Breast Cancer
Support vector machine based methodology for classification of
thermal images pertaining to breast cancer.
Author : Shobhana Periyasamy, Aruna Prakasarao, M. Menaka,
B. Venkatraman , M. Jayashree

Presented by : Mr. Inarm Ekakkaraphibal 4949 & Mr. Pasit Boonpasit 4945
 Table of contents 02

Introduction Objective Materials &

Methods
 Table of contents 03

Results and Discussion Conclusion

Introduction 04

Breast Cancer –
Except for skin cancer,
breast cancer is the
most common cancer
in women in the
United States.

Fig(1)- Image show common cancer in female in each country (2020) -shorturl.at/flLU3
Introduction 05
Breast Cancer –
Breast cancer is a disease that
affects the mammary glands,
which are responsible for the
production of milk. It can often
go undetected until it is found
through regular screenings or
when symptoms begin to appear.

Fig(2)- Image show anatomy of Breast -shorturl.at/apsMZ

Introduction 06

Breast Cancer –
Early detection can improve
survival rates.By catching the
cancer early, doctors can often
remove the tumor before it has a
chance to spread to other parts
of the body. Regular screenings,
such as mammograms or
thermogram, are crucial for early
detection of breast cancer.
Fig(3)- Image show a graph which show the chance of surviving after a proper treatment
for 5 years after been diagnosis with breast cancer -shorturl.at/iGM12
 Key Objectives 07
1 2
Analyze Compare
To analyze the use of thermal To compare the effectiveness of
images in detecting breast analyzing the upper outer
cancer. quadrant of the breast versus
analyzing the whole breast.

Employ
Investigate To use statistical and texture features to
To investigate the potential of distinguish between normal and abnormal
thermography to detect breast subjects, and employ ANOVA and SVM-RBF
cancer at an early stage. kernel techniques for classification.

4 3
Material and Method 08
Material –
In order to capture the thermal
image of the breast, we utilized
the FLIR T460 thermal camera,
which boasts a temperature
sensitivity of 0.01 and a spatial
resolution of 320 x 240 pixels.,
however, any infrared camera can
be utilized to capture the image.
The clearer the image, the more
Fig(4)- Image show FLIR T460 Infared Camera -shorturl.at/bFGIO precise the diagnosis.
Material and Method Method Flow Chart
09

Image acquisition Image preprocessing Segmentation

Classification Statistical Analysis Feature Extraction

Material and Method 10

Method – 1. Image acquisition

According to the study, 50 subjects were
put to the test and it is imperative that the
subjects avoid using talcum powder, tight
fitting clothing, cosmetics, underarm
deodorants, ointments, underarm shaving,
and the consumption of tea, coffee or other
beverages for 2 hours prior to the imaging.

Fig(5)- Image show Talcum Powder -shorturl.at/iPU56

Material and Method 11

Method – 1. Image acquisition

The subject will then be placed in a room with an ambient temperature of 20-22
degrees Celsius, and instructed to undress from the waist up.
A thermalization period of 15 minutes will be implemented prior to the imaging,
as thermal imaging can be influenced by various factors such as air drifts and
ambient temperature.During this acclimation period, the subject must maintain a
sedentary position with hands over their head to avoid friction of the body.
Material and Method 12

Method – 1. Image acquisition

To ensure precise absolute temperature
measurement, radiometric corrections for
distance (0.8m), emissivity (approximately
0.98), humidity, and ambient temperature
were added to the thermogram.

Fig(6)- Example of thermal image -shorturl.at/mnEL2

Material and Method 13

Method – 1. Image acquisition

Five different postures of breast image
were acquired, including one anterior
view, two lateral views, and two
oblique views. The anterior view was
used for image analysis. Any unwanted
labels and temperature scale in the
image were removed using the FLIR
Research IR Software.
Fig(7)- Image show FLIR Research IR Software -shorturl.at/eIVX5
Material and Method 14

(a) Right lateral

(b) Right oblique
(c) Left lateral
(d) Left oblique
(e) Anterior
 Material and Method 15
Method – 2. Image preprocessing
In this study, preprocessing has been carried out in three steps.

q 1 First 2 Second 3 Third

Normalization and gray scale The Homomorphic filter (was Adaptive histogram equalization
conversion were applied to applied in the second step to technique was used in the final
all breast thermal images. suppress the multiplicative noise step to enhance edges and
present in the thermal image. improve image contrast for
further analysis
Material and Method 16
Method – 2. Image preprocessing : Normalization and gray scale conversion
Fig.(8) and (9) -shorturl.at/fgpQX

1 First

Before After
Material and Method 17
Method – 2. Image preprocessing : Homomorphic filter
Fig.(9) and (10) -shorturl.at/cdrEL

2 Second

Before After
 Material and Method 18
Method – 2. Image preprocessing : Adaptive histogram equalization technique

Fig.(10) and (11) - shorturl.at/xyJ23

3 Third

Before After
Material and Method 20
Method – 3. Segmentation
The region of interest in an image is
segmented using a graph cut method,
which employs graph theory and
transforms each pixel into a node
connected by weighted edges, and it's
used to perform efficient segmentation,
by cutting along frail edges, further the
image is divided into four quadrants by
selecting the xy coordinates using two
Fig.(12)- Schematic diagram of breast showing various ROI’s in front view
planes intersecting perpendicularly and
– From paper considering the nipple as centroid.
Material and Method 21
Method – 4. Feature extraction
This study used a combination of
statistical and texture features
extracted from segmented
thermograms to distinguish normal
and abnormal subjects, with first
order statistics giving information
about the distribution of gray levels
and second order statistics providing
insight into the relative positions of
these distributions.
 Material and Method 22

Method – 5. Statistical Analysis

ANOVA is a statistical approach used
to determine if there is a significant
difference between groups by
evaluating the p-value (0.5) and
either accepting or rejecting the null
hypothesis (no difference) or
alternative hypothesis (difference).
Fig.(13)- Example of ANOVA Testing –shorturl.at/fkmwL
Material and Method 23

Method – 6. Classification
Classification is an important task to accurately distinguish the difference
between two classes. The authors used a well-known machine learning
technique called support vector machine (SVM) classifier to classify the two
categories, normal and abnormal, by feeding significant features from the
whole breast and their quadrant breast into the classifier and comparing the
results, using the Gaussian radial bias function kernel as the input features
were non-linearly separable data.
Material and Method 24
Method – 6. Classification

Flow chart of SVM grid search >

Results and Discussion 25
Fig.(14)- Image before and after preprocess

Cancer
Results and Discussion 26
Segmented into 4 quadrants

Fig.(15)- Image after segmentation process

Results and Discussion 27

Fig.(16)- Table 1 show ANOVA results for whole breast

Can’t use
Exceed 0.05
Correlation
Results and Discussion 28
Can’t use
Correlation and Contrast (Exceed 0.05)
Fig.(17)- Table 2 show ANOVA results for upper outer quadrant
Results and Discussion 29
Fig.(18)- Graph show comparison between analyzing Whole Breast vs UOQB
Results and Discussion Whole Breast comparison to other work 30
Accuracy Sensitivity Specificity

100.00%
71.43%
85.71%
90.00%

93.00%

87.00%
90.48%
85.71%
81.07%

Acharya et al. (2012) Usha et al. (2019) This Study (Whole Breast)
Using Texture Features Using Tamura & Using Texture Feature &
and SVM Classifier. Statistical Feature and Statistical Feature and
SVM Classifier. SVM Classifier.
Results and Discussion UUQ comparison to other work 31
Accuracy Sensitivity Specificity

100.00%

85.71%
92.50%
90.00%

93.00%

87.00%
90.48%
85.71%
81.07%

Acharya et al. (2012) Usha et al. (2019) This Study (UOQ)

Using Texture Features Using Tamura & Using Texture Feature &
and SVM Classifier. Statistical Feature and Statistical Feature and
SVM Classifier. SVM Classifier.
Conclusion 32
-Thermography has potential to detect breast cancer at an early stage

-The study used two approaches for analyzing the images, one using features from the whole
breast and another using features from just the upper outer quadrant of the breast, using
features from the whole breast had a higher accuracy rate of 92.86% compared to the upper
outer quadrant approach which had an accuracy rate of 85.71%.

-Statistical analyses were done in order to find the appropriate feature on the whole and
quadrant breast and were given as input for the Support Vector Machine Radial Basis Function
(SVM – RBF) classifier with grid search method.

-The accuracy and reliability of the system could be improved by using more images for
classification.
 33

Q&A
Thanks For Your Attention
Presented by : Mr. Inarm Ekakkaraphibal 4949 & Mr. Pasit Boonpasit 4945