FOR LUNG CANCER SCREENING

IMPLEMENTATION OF ARTIFICIAL NEURAL NETWORK ALGORITHM FOR CLASSIFICATION OF NORMAL AND CRACKLES RESPIRATORY SOUNDS

FOR LUNG CANCER SCREENING

BY

NURFATIHAH BINTI SHAFIAN

A dissertation submitted in fulfilment of the requirement for the degree of Master of Science (Computer and Information

Engineering)

Kulliyyah of Engineering

International Islamic University Malaysia

MARCH 2019

ii

ABSTRACT

The mortality rate of lung cancer is increasing year by year. As reported by the International Agency for Research on Cancer (IARC) in 2014, lung cancer is the primary death cause worldwide and ranked 3^rd in Malaysia as the most common type of cancer. Methods used for lung cancer screening are expensive, contain radiation exposure or invasive. Thus, this study proposed an implementation of Artificial Neural Network (ANN) algorithm for classification of normal and crackles respiratory sounds in lung cancer patients. This method is safe and non-invasive. A total of 23 healthy subjects and 23 lung cancer patients were recruited in this study. The data collected was extracted via a Discrete Wavelet Transform that is based on two different mother- wavelets which are Daubechies 7 (db7) and Symlet 7 (sym7) and Fast Fourier Transform (FFT). Seven statistical features which are mean, variance, minimum amplitude, maximum amplitude, minimum energy, maximum energy and mean of energy were extracted. ANN was used to classify respiratory sound signals as normal and crackle sounds. The results displayed that db7 and sym7 have achieved classification accuracies of 99.0%, while FFT achieved 85.0% classification accuracy.

This shows that db7, sym7, and FFT features and ANN algorithm can be used in classifying respiratory sound signals in lung cancer patients.

iii

ثحبلا ةصلاخ

ناطرسلبا ةصالخا ثوحبلل ةيلودلا ةلاكولا تدافأ .ماع دعب اًماع ةئرلا ناطرس ىضرلم تايفولا لدعم دادز ي في ماع 2014 دحأ وه ةئرلا ناطرس نأ

تاببسم ةيسيئرلا تولما

لماعلا في ىلع ثلاثلا زكرلما لتيح هنأ و

.يازيلام ىوتسم

برتعي صحفلا مثلا ظهبا ةئرلا ناطرسب صالخا يعاعشلا

ضرعتلا ىلع يوتيح هنإف كلذكو ن

الهوخد وأ ةعشلأل لل

.دسج لا ةيمزراوخ قيبطت حترقت ةساردلا هذه نإف ،كلذل ةيعانطصلاا ةينوبصعلا ةكبش

( Artificial Neural Network لخاد ةشخشلخاو ةيداعلا ةيسفنتلا تاوصلأا نم ٍلُك زيمتل )

هذه .ةئرلا ناطرس ىضرم ةقيرطلا

ةعضبا يرغو ةنمآ دعُت ب ةناعتسلاا تم .

23 صخش و ،ميلس 23 يرم ض

ناطرسب .ةساردلا هذه في ةئرلا

وتح قيرط نع تناايبلا جارختسا تم لي

تاجولما ( ةلصفنلما

Discrete

Wavelet Transform ينُب يذلاو )

ت وم ىلع يج

اهمو ينتفلتمخ ينتيسيئر ينت زيشتيبود

7

( Daubechies 7 ِسو )

تل 7 ( Symlet 7 )

ليوتح و يريروف

( عيرسلا Fast Fourier

Transform تا ِس ةعبس جارختسا تم .)

إ لأا عاستلاا ،نيابتلا ،ةيطسولا ةميق يهو ةيئاصح ،رغص

برُكلاو ىرغُصلا ةقاطلا ،بركلأا عاستلاا ةينوبصعلا ةكبشلا مادختسا تم .ةقاطلل ةيطسولا ةميقلاو ى

يسفنتلا توصلا تاراشإ فينصتل ةيعانطصلاا ينب ة

اوصلأا ت .ةشخشلخا تاراشإو ةيداعلا ترهظأ

جئاتنلا

نأ زيشتيبود 7

( Daubechies 7 )

و ِس تل 7 ( Symlet 7 )

ص فينصت ةبسن اققح غلبت ةحيح

.0 99

% امأ ليوتح يريروف ( عيرسلا Fast Fourier Transform

) ةحيحص فينصت ةبسن ققح

غلبت .0 85 تايصاخ مادختسا نكيم هنأ لىإ يرشي اذه .%

زيشتيبود 7

و ِس تل 7 عيرسلا ريروف ليوتحو

.ةئرلا ناطرس ىضرم في ةيسفنتلا توصلا تاراشإ فينصت في

iv

APPROVAL PAGE

I certify that I have supervised and read this study and that in my opinion, it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Master of Science (Computer and Information Engineering).

_____________________________

Noreha binti Abdul Malik Supervisor

_____________________________

Teddy Surya Gunawan Co-Supervisor

I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Master of Science (Computer and Information Engineering).

_____________________________

Malik Arman bin Morshidi Internal Examiner 1

_____________________________

Khairul Azami bin Sidek Internal Examiner 2

This dissertation was submitted to the Department of Electrical and Computer Engineering and is accepted as a fulfilment of the requirement for the degree of Master of Science (Computer and Information Engineering).

_____________________________

Mohamed Hadi Habaebi Head, Department of Electrical

and Computer Engineering

This dissertation was submitted to the Kuliyyah of Engineering and is accepted as a fulfilment of the requirement for the degree of Master of Science (Computer and Information Engineering).

_____________________________

Ahmad Faris Ismail

Dean, Kuliyyah of Engineering

v

DECLARATION

I hereby declare that this dissertation is the result of my own investigations, except where otherwise stated. I also declare that it has not been previously or concurrently submitted as a whole for any other degrees at IIUM or other institutions.

Nurfatihah binti Shafian

Signature ... Date ...

vi

INTERNATIONAL ISLAMIC UNIVERSITY MALAYSIA

DECLARATION OF COPYRIGHT AND AFFIRMATION OF FAIR USE OF UNPUBLISHED RESEARCH

IMPLEMENTATION OF ARTIFICIAL NEURAL NETWORK ALGORITHM FOR CLASSIFICATION OF RESPIRATORY

SOUNDS IN LUNG CANCER PATIENTS

I declare that the copyright holders of this dissertation are jointly owned by the student and IIUM.

Copyright © 2019 Nurfatihah binti Shafian and International Islamic University Malaysia. All rights reserved.

No part of this unpublished research may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise without prior written permission of the copyright holder except as provided below

1. Any material contained in or derived from this unpublished research may be used by others in their writing with due acknowledgment.

2. IIUM or its library will have the right to make and transmit copies (print or electronic) for institutional and academic purposes.

3. The IIUM library will have the right to make, store in a retrieved system and supply copies of this unpublished research if requested by other universities and research libraries.

By signing this form, I acknowledged that I have read and understood the IIUM Intellectual Property Right and Commercialization policy.

Affirmed by Nurfatihah binti Shafian

……..……….. ………..

Signature Date

vii

ACKNOWLEDGMENTS

In the name of Allah, the Most Merciful and Most Compassionate.

Praises to Allah S.W.T., the Exalted, for His blessings, guidance, mercy, and provisions.

Peace and blessings are showered upon our Holy Prophet Muhammad PBUH, his family, and his companions.

Firstly, I would like to express my thankfulness and gratitude to Allah S.W.T for His unlimited blessings and mercy, for allowing me to complete my master thesis.

A special thanks to Dr. Noreha binti Abdul Malik for her continuous support, encouragement, leadership, and advice, and for that, I will be forever grateful. I also would like to thank my co-supervisor, Assoc. Prof. Dr. Teddy Surya Gunawan for his supervision and assistance.

It is my utmost pleasure to dedicate this work to my dear parents, Encik Shafian bin Hassan and Puan Rokayah binti Abang Ahmad, my siblings and also my dearest friends who granted me the gift of their unwavering belief in my ability to accomplish this goal: thank you for your support and patience.

Finally, I wish to express my appreciation and thanks to those who provided their time, effort and support for this project. To the members of my dissertation committee, thank you for sticking with me.

viii

TABLE OF CONTENTS

Abstract ... ii

Abstract in Arabic ………... ⅲ Approval Page ... iv

Declaration ... v

Copyright Page ... vi

Acknowledgments ... vii

Table of Contents ... viii

List of Tables ... x

List of Figures ... xi

List of Abbreviations ... xiii

List of Symbols ... xiv

CHAPTER ONE: INTRODUCTION ... 1

1.1 Background of the Study ... 1

1.2 Problem Statement ... 3

1.3 Research Objectives... 4

1.4 Research Methodology ... 4

1.5 Research Scope ... 5

1.6 Significance of the Study ... 5

1.7 Limitation of the Study ... 5

1.8 Thesis Organization ... 6

CHAPTER TWO: LITERATURE REVIEW ... 7

2.1 Introduction... 7

2.2 Respiratory System ... 7

2.3 Acquisition of Respiratory Sounds ... 9

2.4 Characteristics of Respiratory Sounds ... 10

2.4.1 Normal Respiratory Sounds ... 11

2.4.2 Adventitious Respiratory Sounds... 12

2.4.2.1 Wheeze ... 12

2.4.2.2 Rhonchi ... 12

2.4.2.3 Stridor ... 13

2.4.2.4 Squawk ... 13

2.4.2.5 Crackles ... 13

2.4.2.6 Pleural Rub ... 14

2.5 Feature Extraction Techniques ... 15

2.5.1 Fast Fourier Transform ... 15

2.5.2 Discrete Wavelet Transform ... 18

2.6 Classification of Respiratory Sounds... 22

2.7 Summary ... 25

CHAPTER THREE: METHODOLOGY ... 27

3.1 Introduction... 27

3.2 Method Stages ... 27

ix

3.3 Data Acquisition ... 28

3.4 Pre-processing... 30

3.4.1 Filtering ... 31

3.4.2 Framing and Windowing ... 31

3.5 Feature Extraction ... 32

3.5.1 Features Extraction using FFT ... 32

3.5.2 Features Extraction using DWT ... 33

3.6 Classification ... 35

3.7 Evaluation ... 38

3.8 Summary ... 40

CHAPTER FOUR: RESULTS AND DISCUSSION ... 41

4.1 Introduction... 41

4.2 Performance and Analysis of Classification ... 41

4.3 Evaluation on Respiratory Sounds Classification... 51

4.4 Summary ... 53

CHAPTER FIVE: CONCLUSION AND RECOMMENDATION ... 54

5.1 Conclusion ... 54

5.2 Recommendation ... 55

REFERENCES ... 56

APPENDIX A: MATLAB SOURCE CODE ... 60

x

LIST OF TABLES

Table 2.1 Type of Normal Breath Sounds (Bickley, 2013) 11

Table 2.2 Summary of Abnormal Breath Sounds 14

Table 2.3 Summary of Related Works on FFT Technique 17

Table 2.4 Summary of Related Works on DWT Technique 21

Table 2.5 Comparison of Related Works on Classification Technique 25 Table 3.1 Sub-bands Decomposition of DWT Implementation 34

Table 4.1 Performance of ANN Structures for db7 42

Table 4.2 Performance of ANN Structures for sym7 42

Table 4.3 Performance of ANN Structures for FFT 43

Table 4.4 Validation Performance based on No. of Epoch 47 Table 4.5 Percentage of Sensitivity, Specificity and Accuracy for db7 51 Table 4.6 Percentage of Sensitivity, Specificity and Accuracy for sym7 52 Table 4.7 Percentage of Sensitivity, Specificity and Accuracy for FFT 52

xi

LIST OF FIGURES

Figure 1.1 Estimated Number of Deaths for Both Sexes Worldwide (Stewart & Wild,

2014) 1

Figure 2.1 (A) Anterior View of Upper and Lower Respiratory Tracts. (B) Microscopic View of Alveoli and Pulmonary Capillaries. (Scanlon &

Sanders, 2007) 9

Figure 2.2 Structure of Discrete Wavelet Transform (Kandaswamy et al., 2004) 19 Figure 3.1 Implemented Stages of ANN Algorithm for Classification of RS in Lung

Cancer Patients 27

Figure 3.2 (Left) Anterior Lung-field Auscultation. (Right) Posterior Lung-field

Auscultation. (adapted from Bickley, 2013) 29

Figure 3.3 Thinklabs One Digital Stethoscope 30

Figure 3.4 Filtering Process using Thinklabs Phonocardiography 31

Figure 3.5 Example of Extracted Features in Excel file 35

Figure 3.6 Multilayer Perceptron Neural Network (MLPNN) 36

Figure 3.7 Confusion Matrix 39

Figure 4.1 Confusion Matrices of (a) Training, (b) Validation, (c) Test and (d) All

Data for db7 44

Figure 4.2 Confusion Matrices of (a) Training, (b) Validation, (c) Test and (d) All

Data for sym7 45

Figure 4.3 Confusion Matrices for (a) Training, (b) Validation, (c) Test and (d) All

Data for FFT 46

Figure 4.4 Best Validation Performance based on Cross-Entropy for db7 48 Figure 4.5 Best Validation Performance based on Cross-Entropy for sym7 48 Figure 4.6 Best Validation Performance based on Cross-Entropy for FFT 49 Figure 4.7 ROC Graph for (a) Training, (b) Validation, (c) Test and (d) All Data for

db7 50

Figure 4.8 ROC Graph for (a) Training, (b) Validation, (c) Test and (d) All Data for

sym7 50

xii

Figure 4.9 ROC Graph for (a) Training, (b) Validation, (c) Test and (d) All Data for

FFT 51

xiii

LIST OF ABBREVIATIONS

ANN DFT FFT FPR HPF LPF MLPNN NSCLC RS SCLC TPR

Artificial Neural Network Discrete Wavelet Transform Fast Fourier Transform False Positive Rate High Pass Filter Low Pass Filter

Multilayer Perceptron Neural Network Non-small Cell Lung Cancer

Respiratory Sounds Small Cell Lung Cancer True Positive Rate

xiv

LIST OF SYMBOLS

𝑡 𝑓 𝑥 𝑋 𝑥̅

𝑠² 𝑚^𝑒 𝑛 Φ 𝐸 𝜂

Time Frequency

Input signal in time domain Input signal in frequency domain Mean

Variance

Mean of the energy

Number of terms in the distribution Sigmoid activation function

Network error Learning rate

1

CHAPTER ONE INTRODUCTION

1.1 BACKGROUND OF THE STUDY

Lung cancer is the prominent reason of cancer-related death worldwide. As recorded by the International Agency for Research on Cancer (IARC) in 2014, the estimated number of deaths for both sexes worldwide is almost 1.6 million, as shown in Figure 1.1. It is rated as the highest death tolls among other cancer worldwide and ranked third in Malaysia as the most regular type of cancer followed by colorectal and breast cancers.

Lung cancer death tolls in Malaysia are 4134 where males contribute most among them which are 2783 (Cancer Today, 2012).

Figure 1.1 Estimated Number of Deaths for Both Sexes Worldwide (Stewart & Wild, 2014)

1.5898

0.745517 0.723027 0.693881

0.521817

0.400156

0.330372 0.307471 0.265653 0

0.5 1 1.5 2

Number of death (millions)

Type of cancer

Estimated number of deaths for both sexes worldwide

2

Lung cancer or also known as lung carcinoma is among the most fatal disease usually caused by cigarette smoke (Siang & John, 2016). Other plausible factors include exposure to cigar smoke or second-hand cigarette smoke, air pollution and vulnerability to carcinogens such as asbestos, nickel, radon, arsenic and others (Siang & John, 2016).

Lung cancer grows speedily from a single cell until it became cancerous cells, forming a growth or tumor. The growth of the cell is commonly increased up to 26% in diameter between 30 to 400 days, where the average growth of the cell is 240 days (Bhatt et al., 2012). It also can grow at other parts of the tissues or organs because the cancerous cell can penetrate through them.

Lung cancer originated from the lung is called primary lung cancer. Meanwhile, secondary lung cancer is referred to when cancer had been spread out to the lung from the other part of the body or also known as metastasis (Eldridge, 2016). There are two main types of lung cancer; small cell lung cancer (SCLC) and non-small cell lung cancer (NSCLC) (Eldridge, 2016). The most frequent type that affects patients is NSCLC, where the cell causes no symptoms to be disclosed until it had spread vastly. However, nowadays, various treatments and advanced procedures such as surgery, chemotherapy and radiation therapy can be done in order to decrease the cell cancer growth.

Among the symptoms and signs of early lung cancer are breathing difficulties, coughing, weight loss, fever, lethargy and reappear of infections (Bhatt et al., 2012;

Bickley, 2013). These symptoms are also very common in non-serious diseases, such as headache or migraine. Thus, most people always ignore the symptoms and do not even get medical attention. Nevertheless, for the serious signs of advanced stages of lung cancer, some of the patients may experience the presence of lumps in the neck or collar bone, coughing up with blood, bone pain, non-explained pneumonia, swelling of the face, arms or neck or unexplained huge amount of weight loss.

3

Most respiratory system related diseases like asthma, bronchitis and COPD can be detected by auscultation of respiratory sounds (RS). However, in this study, the process of auscultation of RS is being performed to detect crackles in lung cancer. The experimentation results will then be compared with previous researches to determine the accuracy of the proposed method. This proposed method is beneficial to screen lung cancer among a high-risk population.

1.2 PROBLEM STATEMENT

A screening test identifies an asymptomatic individual, who may have a certain disease.

While diagnostic test is used to determine the presence or absence of a disease when a subject shows any signs or symptoms of the disease. The diagnostic test is performed after a positive screening test to establish a definitive diagnosis. Among the current methods that have been used to screen lung cancer are chest X-ray, computerized tomography (CT) scan and positron emission tomography-computerized tomography (PET-CT) scan. However, some of these methods are invasive, not available at outpatients' clinics, expensive and might take a longer time to get the result. Moreover, these patients will be exposed to radiation during X-ray examination. In order to overcome the limitations, this study will propose a non-invasive, safe, inexpensive method through auscultation of RS to screen lung cancer.

There is a large volume of published studies on classifying respiratory sounds to detect disease such as asthma (Badnjević et al., 2016; Dokur & Ölmez, 2003; Göğüş, Karlık, & Harman, 2015; Güçlü, Halil, & Karl, 2017; Sezgin et al., 2001; Sovijri et al., 2000), pulmonary emphysema (Yamashita et al., 2011) and bronchitis (Lei, Rahman, &

Song, 2014). However, to the best knowledge of the author, there is only one study by Malik et al., (2018), on the classification of normal and crackles based on respiratory

4

sound signals. Thus, this study focuses on classifying between normal and crackles respiratory sounds for the purpose of lung cancer screening by extracting seven different features based on different mother-wavelet and fourier transform.

1.3 RESEARCH OBJECTIVES

The aim of this study is to detect crackles in lung cancer from respiratory sound signals.

In order to achieve the aim, the objectives are:

1. To analyze features of normal and crackles respiratory sound using Discrete Wavelet Transform (DWT) and Fast Fourier Transform (FFT).

2. To implement the Artificial Neural Network (ANN) algorithm to classify between normal and crackles sound.

3. To evaluate the algorithm in terms of accuracy, sensitivity and specificity.

1.4 RESEARCH METHODOLOGY

In this study, five basic steps were involved namely signal acquisition, pre-processing, feature extraction, classification and evaluation. Firstly, the respiratory sound signals were collected from 23 healthy subjects and 23 lung cancer patients by using an electronic stethoscope. Then, these data will undergo through the pre-processing stage where it is the stage for removing noise that was present in these signals. Afterward, the data will be segmented and windowed into several frames before being further processed in the next stage which is feature extraction. These signals will be classified by the ANN algorithm by mixing both normal and crackles samples in order to evaluate the performance of the system.

5 1.5 RESEARCH SCOPE

In order to detect crackles in lung cancer, this study focuses on only one adventitious sound which is crackle sound. This study also targeted to extract seven distinguish features which are mean, maximum and minimum amplitude, minimum, maximum and mean of energy and variance from DWT and FFT. Moreover, this study will determine the accuracy, specificity and sensitivity of the classification technique achieve 80%

successful rate using the ANN algorithm.

1.6 SIGNIFICANCE OF THE STUDY

The proposed method is suggested to minimize the mortality death of lung cancer patient by detecting crackles during the screening test. In current technology, patients must undergo several tests in order to detect cancer. However, this method can save a lung cancer patient’s life by detecting it at the very early stage through the screening process.

Recent method for screening test of lung cancer involving CT-scan, chest X-ray and PET-CT scan whereby these methods are costly, consume a lot of time and being exposed to the radiation. This study has been motivated to introduce a new method for a safe, non-invasive and time minimization of lung cancer screening test by analyzing the respiratory sounds using computer-based and machine learning technique.

1.7 LIMITATION OF THE STUDY

Among several limitations encountered in this study are the data collected from patients who suffered from lung cancer are in stage 3 and 4. Besides that, this study only focuses on a single adventitious lung sound which is crackle. This might affect the overall

6

results since crackle sound can also be found in other diseases such as heart congestion failure, pulmonary fibrosis or bronchiectasis.

During the data acquisition stage, there were a few data that must be recollected from the patients in order to gain more accurate information. However, there were a few patients who died before the appointment, or some of them were very worst in condition which makes it difficult for them to cooperate in the data collection stage. Therefore, the process of collecting data exceeded the actual proposed schedule.

Moreover, during the pre-processing stage, the noise in a respiratory sound signal might mimic the crackles sound. Meanwhile, during the windowing process, the data did not overlap between each framed signal. Thus, several information on respiratory sound might be lost.

1.8 THESIS ORGANIZATION

Chapter 1 presents the introductory of lung cancer. This chapter provides the definition of types and symptoms of lung cancer. Meanwhile, in Chapter 2, the study of the respiratory system, respiratory sound, the differentiation between several techniques on feature extraction and the classification method, based on past studies are described. It highlights their achievement and weakness. The methodology implemented and the process flow of this study are explained in Chapter 3. While Chapter 4 focuses on the results and clarification of the results. Finally, Chapter 5 concludes this research which includes a brief discussion on the future works that can be implemented and recommended.

7

CHAPTER TWO LITERATURE REVIEW

2.1 INTRODUCTION

This chapter provides the understanding of the respiratory system and the types of respiratory sounds (RS) which are normal and adventitious sounds. The techniques used in acquiring RS, the extraction of features from sound signals and classification method of sound signals were discussed and compared with existing studies and related works that have been done by previous researchers.

2.2 RESPIRATORY SYSTEM

The functions of the respiratory system are to provide oxygen to the blood and remove carbon dioxide from it (Scanlon & Sanders, 2007). The exchange of both gases will help balance the pH of the body fluids thus maintaining the health of the body. It also provides the function for a sense of smell, speech and other vocalization parts such as laughing or crying. The respiratory system contains upper- and lower-part respiratory tract (Saladin, 2008). The upper part of the respiratory tract is outside the chest cavity, while the lower part is found within the chest cavity. The air passage movement in the respiratory system is from the nose or mouth, then the air travels to pharynx (throat), goes down to larynx (voice box), enters the trachea, bronchi, then bronchioles and finally, the exchange of gases between oxygen and carbon dioxide occurs at the alveoli (Scanlon & Sanders, 2007).

The fundamental part of inspiration is nose; whereby it is made up of bone and cartilage wrapped up with skin (Scanlon & Sanders, 2007). Anatomy of the nose starts

8

from the anterior opening of the nose called the nostrils and extends to a pair of posterior openings which is called choanae. Then, it enters the nasal cavity where it is divided into right and left halves called nasal fossae by a wall of bone and hyaline cartilage which is referred to as the nasal septum. The nose contains three folds of tissue; the superior, middle and inferior nasal conchae that will enable the nose to clean, warm and moist inhaled air and remove impurities besides its main function which is to smell or to detect any scents (Saladin, 2008). The pharynx is a muscular funnel prolong from choanae to the larynx. It is divided into three regions; nasopharynx, oropharynx and laryngopharynx (Saladin, 2008). The larynx is the airway passage between the pharynx and the trachea. It is made up of nine pieces of cartilage joined together by ligaments in order to retain the air passages open at all the time (Scanlon & Sanders, 2007). The trachea extends from the larynx to the primary bronchi and the length is about 10 to 13 cm long (Scanlon & Sanders, 2007). There are about 16-20 C-shaped rings of hyaline cartilage which support the wall of the trachea.

The inhaled air then enters the branches of the primary bronchi which is the secondary bronchi that eventually leads to the lobes of each lung; two on the left and three on the right. The branches of the bronchial tubes which referred to as bronchial tree will extend further and become smaller and smaller. These smaller branches are known as bronchioles. Each bronchiole divides into 50-80 terminal bronchioles which are the final branches of the conducting division. There are approximately 65000 of these in each lung where the diameter measurement is around 0.5 mm or less (Saladin, 2008). At each and every terminal bronchiole, there are alveoli; air sacs of the lungs. It is surrounded by blood capillaries where the exchange of gases between oxygen and carbon dioxide take place in the human body. Figure 2.1 showed the anterior view of

9

upper- and lower-part respiratory tracts along with a microscopic view of alveoli and pulmonary capillaries.

Figure 2.1 (A) Anterior View of Upper and Lower Respiratory Tracts. (B) Microscopic View of Alveoli and Pulmonary Capillaries. (Scanlon & Sanders, 2007)

2.3 ACQUISITION OF RESPIRATORY SOUNDS

Traditionally, the acquisition of RS has been done by auscultation. It is a process of hearing a breathing sound and any abnormal sounds produced from inside the patients’

body (Bickley, 2013). It can be done via bare ears or with the help of modern equipment such as the stethoscope. From time to time, the stethoscope had evolved around the

10

technology not only in terms of its design but the efficiency as well. Fundamentally, there are two types of stethoscope; acoustic or electronic, sometimes referred to as stethophone. Acoustic is the most recognizable and familiar stethoscope which being used commonly and widely over the world. Meanwhile, the electronic stethoscope is the type that can amplify the sound levels accordingly and minimize the surrounding noise (Hoffmann et al., 2013).

There is no standard procedure to determine the point of auscultation. Still, the normal procedure of physical examination of RS that has been conducted usually takes place at both the anterior and posterior part of the body. This includes 10 points at the anterior and 10 points at the posterior.

2.4 CHARACTERISTICS OF RESPIRATORY SOUNDS

There are three attributes that belong to RS; frequency, intensity and quality (Sarkar et al., 2015). These features will categorize the RS because various sounds belong to different categories. Usually, the normal classification of RS in bands of frequency ranges between the lowest which is 100-300 Hz, medium (between 300-600 Hz) until highest range; up to 600-1200 Hz (Pasterkamp, Kraman, & Wodicka, 1997). The heart sounds dominant frequency range is less than 100 Hz (Charbonneau et al., 1983).

Hence, any signals that fall below 100 Hz must be filtered out for the evaluation of RS.

Intensity is the characteristics to measure the strength of the RS. Meanwhile, quality is the influential characteristics of a sound that can differentiate between two identical sounds which contain the same loudness and pitch. RS is normally being sampled at 4- 32 kHz which indicate only important information that being embedded in the RS signal (Lei, Song, & Rahman, 2012). RS can be categorized either normal or abnormal, where sometimes abnormal is being referred to as adventitious sounds.