i

INVESTIGATION OF KIDNEY ULTRASOUND IMAGE TEXTURE

FEATURES FOR HEALTHY SUBJECT

WAN NUR HAFSHA BT WAN KAIRUDDIN

A thesis submitted in

fulfillment of the requirement for the award of the

Degree of Master of Electrical Engineering

Faculty of Electric & Electronic Engineering

Universiti Tun Hussein Onn Malaysia

JAN 2016

iii

Dedicated, in thankful appreciation for support, encouragement and understandings

to my beloved family and friends.

iv

ACKNOWLEDGEMENT

All praise and thanks to Allah, for His blessing at every stage of this journey until I

completed my study successfully.

It is my genuine pleasure to express my deep sense of thanks and gratitude to

my supervisor, Dr. Wan Mahani Hafizah bt Wan Mahmud for her guidance during

the term of my candidature. This work will not be completed without her full

cooperation and assistance.

I am also indebted to the staff and students Department of Electronic

Engineering, Faculty of Electrical & Electronics Engineering, UTHM and

Biomedical Department, KKTM Ledang for the cooperation during ultrasound

measurement.

It is my privilege to thank my late mother, who passed away before I

completed this thesis, during my last semester. She is my great supporter which gives

me strength to complete my Master’s degree successfully. Not to forget, beloved

husband, son and the entire family members for their understanding and constant

encouragement.

Last but not least, many thanks to the Ministry of Higher Education Malaysia

for providing the funding of my Master’s degree and UTHM as the place I study and

work

v

ABSTRACT

Image feature extraction is a technique to identify the characteristic of the image. In

this thesis, the focus is on a healthy kidney ultrasound image. The main objective is

to select the features that best describe a tissue characteristic of a healthy kidney

from ultrasound image. Three ultrasound machines that have different specifications

are used in order to get a different quality (different resolution) of ultrasound image.

Initially, the acquired images are manually cropped to get the region of interest

(ROI) of kidney. Then the cropped images are undergoing filtering process to

remove the speckle noise that presence in the ultrasound images. Four filtering

techniques (Wiener filter, Median filter, Gaussian Low Pass filter and Histogram

Equalization) are tested to find the best filtering technique for all the three groups of

image. By calculating Mean Square Error (MSE) and Peak Signal to Noise Ratio

(PSNR), result shows that Gaussian Lowpass Filter having the highest PSNR, then is

choose as the filtering method for this thesis. These enhanced images then are

segment to create a foreground and background where the mask is created. In this

thesis, only statistical based texture features method is used which depends on the

spatial distribution of intensity values or gray levels in the kidney region. Three

statistical feature extractions techniques used are Intensity Histogram (IH), Gray-

Level Co-Occurance Matrix (GLCM) and Gray-level run-length matrix (GLRLM).

By using One-Way ANOVA in SPSS, the result shows that three features (Contrast,

Difference Variance and Inverse Difference Moment Normalized) from GLCM are

not statistically significant; this concludes that these three features describe healthy

kidney characteristics regardless of the ultrasound image quality.

vi

ABSTRAK

Pengekstrakan ciri bagi imej adalah satu teknik untuk mengenalpasti ciri-ciri bagi

imej. Objektif utama adalah untuk memilih ciri terbaik pada imej ultrasound buah

pinggang yang dapat menggambarkan ciri-ciri tisu bagi buah pinggang yang normal.

Tiga jenis mesin ultrasound yang mempunyai spesifikasi yang berbeza digunakan

bagi mendapatkan imej yang mempunyai kualiti imej yang berbeza. Teknik manual

cropping dilaksanakan bagi mendapatkan region of interest (ROI) bagi imej buah

pinggang. Kemudiannya imej tersebut akan melalui proses penapisan bagi

meminimakan speckle noise yang hadir dalam imej. Empat teknik penapisan (Wiener

filter, Median filter, Gaussian Low Pass filter dan Histogram Equalization)

digunakan bagi mengetahui teknik penapisan manakah yang paling berkesan untuk

menapis speckle noise bagi ketiga-tiga kumpulan imej. Dengan mengira Mean

Square Error (MSE) dan Peak Signal to Noise Ratio (PSNR), keputusan

menunjukkan Gaussian Lowpass Filter mempunyai nilai PSNR tertinggi, jadi ia

dipilih sebagai teknik penapisan untuk kesemua imej. Imej yang telah ditapis ini

kemudiannya melalui proses segmentasi di mana bahagian latar belakang imej

dihitamkan dan hanya bentuk buah pinggang yang disegmenkan. Bagi proses

pengekstrakan ciri imej, kaedah statistik berdasarkan cirri-ciri tekstur imej

digunakan. Kaedah ini adalah berdasarkan kepada taburan bagi nilai kecerahan imej

(gray levels) pada bahagian imej buah pinggang. Tiga teknik yang digunakan ialah

Intensity Histogram (IH), Gray-Level Co-Occurance Matrix (GLCM) and Gray-

Level Run-Length Matrix (GLRLM). Dengan menggunakan kaedah One-Way

ANOVA pada perisian SPSS, keputusan menunjukkan tiga ciri (Contrast, Difference

Variance dan Inverse Difference Moment Normalized) daripada GLCM adalah sama

secara statistik, maka dengan itu disimpulkan bahawa ketiga-tiga ciri tersebut

menggambarkan ciri-ciri tisu bagi buah pinggang yang normal daripada imej

ultrasound tanpa mengira kualiti bagi sesuatu imej ultrasound yang digunakan.

vii

CONTENTS

TITLE PAGE i

DECLARATION ii

DEDICATION iii

ACKNOWLEDGEMENT iv

ABSTRACT v

ABSTRAK vi

LIST OF CONTENTS vii

LIST OF TABLES x

LIST OF FIGURES xi

LIST OF SYMBOLS AND

ABBREVIATIONS xii

LIST OF APPENDICES xiii

CHAPTER 1 INTRODUCTION

1.1 Background 1

1.2 Problem Statement 3

1.3 Objectives 3

1.4 Scope of the thesis 4

1.5 Significant of the thesis 4

1.6 Outline of the thesis 4

CHAPTER 2 LITERATURE REVIEW

2.1 Introduction 6

2.2 Anatomy of Kidney 6

2.3 Ultrasound 9

2.4 Kidney Measurement 10

2.5 Pre-processing Technique 12

2.5.1 Image Cropping 12

viii

2.5.2 Image Enhancement Technique 12

2.5.3 Segmentation Technique 15

2.6 Feature Extraction 16

2.7 Feature Selection 16

CHAPTER 3 METHODOLOGY

3.1 Introduction 18

3.2 Research Purpose 18

3.3 Methodology 19

3.3.1 Image Acquisition 19

3.3.2 Image pre-processing 22

3.3.2.1 Image Cropping 22

3.3.2.2 Image Enhancement 22

3.3.2.3 Image Segmentation 23

3.3.3 Feature Extraction 24

3.3.3.1 Intensity Histogram Features 24

3.3.3.2 GLCM Features 26

3.3.3.3 GLRLM Features 31

3.3.4 Feature Selection 34

CHAPTER 4 RESULT

4.1 Introduction 39

4.2 Image Acquisition 40

4.3 Pre-processing 41

4.3.1 Image Cropping 41

4.3.2 Image Enhancement 41

4.3.3 Image Segmentation 44

4.4 Feature Extraction 45

4.5 Feature Selection 45

4.5.1 Intensity Histogram (IH) 46

4.5.2 Gray-Level co-Occurrence Matrix (GLCM) 47

4.5.3 Gray-Level Run Length Matrix (GLRLM) 49

ix

CHAPTER 5 CONCLUSION AND FUTURE WORK

5.1 Conclusion 51

5.2 Future Work 52

REFERENCES 53

APPENDIX A 58

APPENDIX B 80

x

LIST OF TABLES

TABLE TITLE PAGE

3.1 Comparison of US machines specifications 20

4.1 Comparison of PSNR value for three US machines 47

4.2 Result for IH 49

4.3 Result for GLCM 50

4.4 Result for GLRLM 53

xi

LIST OF FIGURES

FIGURE TITLE PAGE

2.1 Illustration of the human kidney 7

2.2 B-Mode cross-section of the kidney 9

2.3 Scanning planes 10

2.4 Transducer is placed below the right ribs 11

2.5 Transducer is placed in coronal view 11

2.6 US Kidney image 11

2.7 Basic steps in Frequency Domain Filtering 13

3.1 Direction of GLCM generation 26

3.2 4 level of gray-level image 27

3.3 An image and its Gray- Run

Length- Level Matrix 32

3.4 Geometrical relationship of GLRLM 32

3.5 Box-plot for detecting outliers 36

3.6 Flow process of overall framework 38

4.1 Kidney US image from three US machine 43

4.2 Image cropping consisting ROI 44

4.3 US images before and after enhancement

(Machine 1) 45

4.4 US images before and after enhancement

(Machine 2) 46

4.5 US images before and after enhancement

(Machine 3) 46

4.6 Manual contouring of kidney image 48

4.7 Blackmasked image background 48

xii

LIST OF SYMBOLS AND ABBREVIATIONS

CAD Computer Aided Diagnosis

GLCM Gray-Level Co-Occurrence Matrix

GLRLM Gray-Level Run-Length Matrix

IH Intensity Histogram

MRI Magnetic Resonance Imaging

ROI Region of Interest

SPSS Statistical Package for the Social Sciences

US Ultrasound

xiii

LIST OF APPENDICES

APPENDIX TITLE PAGE

A DATABASE 58

B MATLAB CODE 80

CHAPTER 1

INTRODUCTION

1.1 Background

The increasing reliance of modern medicine on diagnostic techniques such as

computerized tomography, histopathology, magnetic resonance imaging, radiology

and ultrasound imaging shows the importance of medical images [1]. Ultrasound

(US) imaging is an imaging technique that is far the least expensive and most

portable comparing to other standard medical imaging modalities. US imaging is a

safe technique, easy to use, noninvasive nature and provides real time imaging, hence

it is used extensively. But on the downside, ultrasound imaging has a poor resolution

of image compared with other medical imaging instrument like Magnetic Resonance

Imaging (MRI). US has wide spread application as a primary diagnostic aid of

obstetrics and gynecology, due to the lack of ionizing radiation or strong magnetic

fields. General US imaging applications include soft tissue organ and carotid artery

[2].

The US image produced has poor quality image which are affected by

multiplicative speckle noise due to loss of proper contact or air gap between

transducer and the body part. It also may occur during beam forming process or

signal processing. Speckle has variation of gray level intensities, where ranging from

hyper-echoic to

2

hypo-echoic and the presence of this noise make an analysis of this US images are

more complex [3]. Speckle occurs mostly in images of underlying structures in the

body like kidney and liver because they are too small to be resolved by large

wavelength US [4]. Speckle noise will affect the tasks of human interpretation and

diagnosis. There are advantages for US imaging, but it needs to use properly to cope

with the disadvantages.

This study proposes an approach of feature extraction of healthy kidney US image.

The extensive use of computer technology to process such obtained image data, the

area of medical image analysis is taking advantage to extract important features to

implement in a computer aided diagnosis (CAD) system. In most of the hospitals

these medical images are stored to create a patient database for further reference and

it is retrieved by the patient’s name. The extracted feature parameters that describe

characteristics of the images are then can be efficiently used to retrieve images from

huge database [5, 6]. Here in this study the feature extraction of the healthy kidney

will be derived based on the intensity-level and regional gray level distribution in

kidney region. Before the feature extraction process, the measured US images need

to go pre-processing stage for preserving pixels of interest prior to feature extraction.

The pre-processing techniques include cropping to get Region of Interest (ROI),

speckle noise de-noising and image segmentation. Texture features are used in this

research. Texture is an image feature that provides important characteristics for

surface and object identification from the image. Feature extraction methodologies

analyze the pre-processed images to extract the most prominent features that

represent various sets of features based on their pixel intensity relationship. The

features will be extracted from three different classes of images measured from three

different specifications of US machines that have different quality of the image

produced. Three technologies have hit the mainstream market that have the most

impact on ultrasound image quality. These technologies are Tissue Harmonics

Imaging, Compound Imaging and Speckle Reduction Imaging. The comparison of

these three US machine is described in Chapter 3. In this study, a set of three

statistical texture features are used namely Intensity Histogram (IH), Grey-Level Co-

Occurrence matrix (GLCM) and Gray-Level Run-Length Matrix (GLRLM). These

features parameters are derived from the total of 150 healthy kidney images. The

images are divided into three classes of images (50 images each) with different

3

qualities from the three different specifications of US machines. Then, these three

classes of images will be compared statistically to choose the best features that

described a healthy kidney.

1.2 Problem statement

1) The ultrasonography imaging in kidney diagnosis is limited by its

dependencies on the expertise of the sonographers to detect, measure,

segment and analyze structure of kidney.

2) An ultrasound image of kidney may be interpreted differently by

different operators and the result is relative to the sonographer’s

expertise, variations in human perceptions of the images, as well as

differences in features used in diagnosis.

3) The variation and limitation in the quality of ultrasound image itself

due to the speckle noise.

4) As imaging technologies advance, this would be a unique contribution

as computer aids are demanded and become indispensable in

physicians’ decision making in order to improve the existing

ultrasonography for detecting, and analyzing the kidney using

ultrasound machine.

1.3 Objectives

The works undertaken in this proposal are aiming on the following objectives:

1) To study pre-processing techniques and feature extraction method of

healthy kidney US image for different specification of US machine.

2) To extract features from healthy kidney to characterize the features

and characteristics of healthy kidney US image for different qualities

of US image using texture-based features.

3) To select the features after extraction to classify it as healthy kidney

features.

4

1.4 Scope of the project

The framework of this research is divided into four parts. Firstly, the measurement of

B-mode US image of healthy kidney in longitudinal direction using three different

specifications of ultrasound machines that produce different image quality. Next, the

images are pre-processes for the enhancement of the image to make it easier for the

extraction and analysis. The third part is feature extraction that is extracted from all

the US images that being taken. The textures are then going through feature

extraction selection by comparing these three groups of image from three different

US machines using statistical analysis.

1.5 Significant of the thesis

Kidney diagnosis using ultrasound may be improved by adding a CAD system to the

process, so that the dependencies to the sonographer or medical doctors will be

reduced. In order to do that, good features that can significantly characterize the

kidney need to be investigated, regardless of the ultrasound image quality. Different

types and specifications of US machine will produce different quality of images.

With this study, the best features that describe healthy kidney will be selected by

comparing the three groups of kidney US image measured from the different

specifications of US machine. Not only to analyze the current available techniques,

the development of new algorithm for feature extraction of kidney ultrasound images

study has potential to improve the existing ultrasonography for detecting, and

analyzing the kidney using US machine.

1.6 Outline of the thesis

This project consists of five chapters. In the first chapter, it discuss about the

background, problem statements, objective, scope and significant of this project.

Chapter 2 will discuss more on literature reviews and theories which are related to

this project. The third chapter is Methodology which discussed a description of the

framework, the parameters, and also the study done for this project. The result and

discussion will be presented in Chapter 4. Last but not least, Chapter 5 will

5

summarize the whole project and make a conclusion of this project and

recommendations for future work that can be done.

CHAPTER 2

LITERATURE REVIEW

2.1 Introduction

This chapter describes the basic concept and the reviews of anatomy of kidney and

ultrasound imaging techniques. The discussed previous works in this chapter are

mainly focused on the design technique (approach) in the enhancement techniques,

feature extractions of the US image and features selection techniques. Additional, the

proposed solution for this study is also discussed in this chapter.

2.2 Anatomy of Kidney

Kidneys are the organ that functions as a system of waste filtering and disposable in

the body. The kidneys are the excretory organs which help purify the blood and

remove toxins from the body. As much as 1/3 of all blood leaving the heart passes

into the kidneys to be filtered before flowing to the rest of the body’s tissues. Most of

the human have a pair of kidneys but a person can live with only one kidney or one

functioning kidney. The loss of both kidneys would lead to a raid accumulation of

wastes and death within a few days time. The kidneys are located on left and right

side of the spine as illustrated in Figure 2.1. Kidneys have a bean-like shape and are

approximately 11-14 cm in length, 6 cm wide and 4 cm thick. The weight for each

adult kidney is between 125 and 170 g in males and between 115 and 155 g in

females [8] . There is also the liver on the anatomically right side close to the kidney

and also slightly smaller [9]. The kidney is attached to the blood circuit via the renal

7

artery and the Renal vein and to the bladder via the ureter. The ureter can also be

easily recognized in Figure 2.1. With that, the main function of the kidneys is

obvious. They filter the blood and produce the urine that will be put into the bladder

using the ureter. That being the most obvious part, the kidneys are also responsible

for other tasks, for example regulation of electrolytes, the acid-base balance, and the

blood pressure.

Figure 2.1: Illustration of the human kidney

The kidney structures illustrated in Figure 2.1 have different functions to make sure

the kidney is working well. Short descriptions of kidney structures are described

below:

i. Cortex

Cortex is the outer part of the kidney and has reddish colour with smooth

texture.

ii. Medulla

Medulla is the inner part of the kidney and the word “medulla” means inner

portion. This area is striped red brown colour

8

iii. Renal pelvis

The renal pelvis is the funnel-shaped basin (cavity) that receives the urine

drained from the kidney nephrons (functional units of the kidneys) via the

collecting ducts and then the (larger) papillary ducts.

iv. Renal artery

The renal artery delivers oxygenated blood to the kidney. This main artery

divides into many smaller branches as it enters the kidney via the renal hilus.

These smaller arteries divide into vessels such as the segmental artery, the

interlobar artery, the arcuate artery and the interlobular artery. These

eventually separate into afferent arterioles, one of which serves

each nephron in the kidney

v. Renal vein

The renal vein receives deoxygenated blood from the peritubular veins within

the kidney. These merge into the interlobular, arcuate, interlobar and

segmental veins, which, in turn, deliver deoxygenated blood to the renal vein,

through which it is returned to the systemic blood circulation system

vi. Ureter

The ureter is the structure through which urine is conveyed from the kidney

to the urinary bladder

vii. Calyses

Calyces are parts of the kidney that collect urine before it passes further into

the urinary tract. The calyces are part of the renal pelvis, a convex system of

sinuses that connect the innermost part of the kidney to the ureters and, from

there, to the bladder.

The macroscopic part of the kidney can be seen on US-images, an example is

displayed in Figure 2.2. As for the healthy kidney image, it has a bright area around

it, made up of perinephric fat and Gerota’s fascia. The kidney periphery part appear

grainy gray, consists of renal cortex and pyramids while the central area of the

9

kidney, the renal sinus, will appear bright (echogenic), and consists of renal sinus fat,

calyces, as well as renal pelvis [15].

Figure 2.2: B-Mode cross-section of the kidney

2.3 Ultrasound

Ultrasound is a mechanical wave, with frequency for clinical use between 1MHz and

15MHz. Higher frequencies of ultrasound have shorter wavelengths and are

absorbed/attenuated more easily. For organs like liver and kidney lower frequencies

are used compared to organs like breasts and skin. The frequency used is normally

from 2.5 to 3.5 MHz. The range of wavelength in ultrasound tissue is ~0.1 and

1.5mm because the speed of sound in tissue is ~1540m/s. Ultrasound waves

produced by a transducer. As the ultrasound passes through tissue, a small fraction of

energy is reflected from the boundaries between tissues which have slightly different

acoustic and physical properties while the remaining energy of the beam is

transmitted through the boundary.

The reflected waves are detected by the transducer and the distance to each tissue

boundary is calculated from the time between pulse transmission and signal

reception.

There are several different modes of ultrasound imaging. Historically, the

first one was the A-mode. It only consists of a one-dimensional signal, showing the

backscattered signal over time for one line through the examined tissue. The

interpretation of those A-mode-signals is hard and a lot of experience is necessary, so

the B-mode was developed. Simplified, it consists of several parallel A-mode-signals

that are put together to build a two-dimensional image of the examined tissue. The

10

result is what most of the people think of when it comes to ultrasound. There are a lot

of other techniques, for example three- and four-dimensional ultrasound and

Doppler-US. In this work, only B-mode technique is used to get the image of kidney.

The basic principles of B-mode imaging are much the same today as they

were several decades ago. This involves transmitting small pulses of ultrasound echo

from a transducer into the body. As the ultrasound waves penetrate body tissues of

different acoustic impedances along the path of transmission, some are reflected back

to the transducer (echo signals) and some continue to penetrate deeper. The echo

signals returned from many sequential coplanar pulses are processed and combined

to generate an image.

2.4 Kidney Measurement

The three major planes for scanning are longitudinal, cross-section/transverse and

coronal. It can be referred in Figure 2.3

Figure 2.3: Scanning planes

To do the ultrasound, the subject needs to lie supine on bed, meaning that lay on

back. The transducer needs to be slide in longitudinal/cross-section plane below the

right ribs to see the right kidney. Refer Figure 2.4. For the right kidney, use the liver

as reference and aim the transducer lower it to find the kidney. Gently slide the

transducer up and down to scan the entire kidney. If needed, the subject can be

advised to inspire or exhale, which allows for subtle movement of the kidney.

11

Figure 2.4: Transducer is placed below the right ribs

If the image of kidney is not clear, try put the transducer at coronal view to get a

better image as in Figure 2.5. The B-mode US kidney image can be referred in

Figure 2.6.

Figure 2.5: Transducer is placed in coronal view

Figure 2.6: US Kidney image

Liver

12

The same procedures used to see the left kidney. The left kidney normally located

lower than right kidney. Take spleen as the reference to find the left kidney.

Normally the left kidney is hard to see compared to the right kidney.

A good image quality is fairly subjective. It’s also relative to the capabilities

of the machine. In this work, three different specifications of US machine will be

used to produce different ultrasound image quality.

2.5 Pre-Processing Techniques

Pre-processing techniques are divided into three sections, image cropping, image

enhancement and image segmentation. Overview of each technique is described in

the following sub topics.

2.5.1 Image cropping

The US image measured may contain unwanted information other than kidney, like

label or background noise. Other than that, some other organs also lie close to the

kidney which may give effect to the performance of US image processing, so that

finding the Region of Interest (ROI) which contains only kidney image are helpful to

speed up further processing of US image such as segmentation of the ROI which also

helps in increasing the accuracy. Most of the researchers developed the image

cropping by manually crop the US kidney image, which contains the ROI [12, 14,

17]. Wan M.Hafizah et al. [18] has proposed an automatic generation of ROI in

kidney US image using texture analysis.

2.5.2 Image Enhancement Technique

The quality of US image is affected by the presence of the speckle noise which leads

to the difficulties in interpretation of US image. Due to the presence of the Speckle

noise in US images, the enhancement of US image is extremely difficult in image of

liver and kidney whose underlying structures are too small to be resolved by large

wavelength [14].They also complicate further image processing, such as image

segmentation and edge detection .

13

Filtering techniques used for image enhancement can be classified as spatial

filtering and frequency domain filtering. Spatial filtering is defined by a

neighbourhood and an operation that is performed on the pixels inside the

neighbourhood or can be derived as an image operation where each pixel value I (u,

v) is changed by a function of the intensities of pixels in a neighbourhood of (u, v).

Two types of spatial filtering are linear spatial filtering and non linear spatial

filtering. A filtering method is linear when the output is a weighted sum of the input

pixels and methods that do not satisfy the linear property are called non-linear which

involving the neighbourhood encompassed by the filter. Frequency domain is a space

defined by values of the Fourier transform and its frequency variables (u, v). Relation

between Fourier Domain and image is u = v = corresponds to the gray-level average.

Low frequencies means image’s component with smooth gray-level variation.

Frequency domain filtering requires some steps to do the operation. The steps are

shown in Figure 2.7. The pre-processing stage might encompass procedures such as

determining image size, obtaining the padding parameters and generating the filter.

Post processing entails computing the real part of the result, cropping the image and

converting it to certain class for storage.

Figure 2.7: Basic steps in Frequency Domain Filtering

The intensity transformation function based on information extracted from image.

Intensity histogram plays a central role in image processing, in area such as

enhancement, compression, segmentation and description.

14

Mean Squared Error (MSE) is the average squared difference between an

original image and a filtered image. It is computed pixel-by-pixel by adding up the

squared differences of all the pixels and dividing by the total pixel count. MSE of the

output image is defined as:

MSE = ∑ ∑ |x(i, j) − x �(i, j)� |�

�������

MN (2.1)

where � (�, �) is the original image, � � (�, �) is the output image and MN is the size of

the image.

Peak Signal-to-Noise Ratio (PSNR) is the ratio between the original image

and the filtered image, given in decibels (dB). The higher the PSNR, the closer the

filtered image is to the original image. In general, a higher PSNR value should

correlate to a higher quality image, but tests have shown that this isn't always the

case.

However, PSNR is a popular quality metric because it's easy and fast to

calculate while still giving good results. PSNR is defined as:

PSNR = 20 log�� �2� − 1

√MSE� dB (2.2)

where n is the number of bits used in representing the pixel of the image. For

grayscale image, n is 8.

There are many research has been done to compare the enhancement

technique for speckle noise in US image. Wan Mahani et al. [15] did a comparative

evaluation of US healthy kidney image and used spatial domain filter (Median

Filter), frequency domain filtering (Gaussian-low pass filter), morphological

processing and histogram equalization for image enhancement. The result shows that

morphological technique is the best technique in enhancing the ultrasound kidney

image by evaluating using MSE and PSNR. Prema T. Akkasaligar et al. [16] had

compared three type of filter to remove speckle noise from the US image. Gaussian

Low-Pass filter, Median filter and Wiener filter had been evaluated and observed that

15

optimal de-speckling is found in Gaussian Low-Pass filter. In [17], authors used a

combination of Gabor filter and histogram equalization for filtering, smoothing and

sharpening the US kidney image. Amira et.al [18] did a Comparative Study of

Different Denoising Filters for Speckle Noise Reduction in Ultrasonic B-Mode

Images. The result shows that Speckle Reducing Anisotropic Diffusion filter (SRAD)

is better than several commonly used filters including Gaussian, Gabor, Lee, Frost,

Kuan, Wiener, Median, Visushrink, Sureshrink and also the Homomorphic filter, but

on the other hand, the CPU for the SRAD is very high compared to the other filters.

The Log Gabor filter gives similar performance of SRAD in less CPU time.

2.5.3 Segmentation Technique

Segmentation a process to subdivides an image into its constituent regions or objects.

The segmentation process is stop when the ROI have been isolated. The main

purpose of the segmentation process is to get more information in the region of

interest in an image which helps in getting correct features of the image.

Segmentation process is one of the difficult parts to be achieved as segmentation

accuracy will determine either the analysis process and procedures success or not.

The segmentation will provide a boundary over a kidney image.

Most of the work in recent years on US kidney images deals with the

segmentation techniques to identify the boundary of kidney using various

methodologies. In [17], authors comparing the region based segmentation and cell

segmentation to segment the US kidney image which result the region based

segmentation gives better result. K. Bommanna Raja et al. [10] used a higher order

spline interpolated contour obtained with up-sampling of homogenously distributed

coordinates for segmentation of kidney region in different classes of ultrasound

kidney images. Wan Mahani Hafizah et.al [19] manually identified coordinates of

points along contour of and then interpolated using cubic spline interpolation

technique.

16

2.6 Feature Extraction

Texture is an image feature that provides important characteristic for surface and

object identification from an image. Texture is characterized by the spatial

distribution of gray levels in an image.

Feature extraction is a critical step for US kidney image processing. Feature

extraction will extract the most prominent features that represent various sets of

features based on their pixel intensity relationship and statistics.

According to the number of intensity pixels, statistics are classified into first -

order, second-order and higher-order statistics. The first-order features contain

conventional statistical measures like the mean, the standard deviation, and the

variance and parameters obtained by fitting various statistical distributions to the US

B-Mode data. The first-order statistics can be found in IH. The second-order texture

features were chosen to be extracted from GLCM and the GLRLM is the higher-

order statistics. The detailed features used in this work will be described in the

methodology in Chapter 3.

There are number of works that have been done by researchers regarding the

US image extraction. Raja et al. has extracted features in kidney US images based on

content descriptive multiple features [10], geometric moment features [20] and

regional gray distribution. Karthikeyini et al. used principal component analysis

(PCA) method and their analysis shows that there exists an appreciable measure of

relevance for weight vector in classifying kidney images [21]. Wan Mahani et al.

[19] used different classes of kidney US image for feature extraction that is based on

five intensity histogram features and nineteen gray level co-occurance matrix

(GLCM) features. The results show that feature extraction from kidney US images

based on those features is possible and they are highly effective in classifying the

kidney disease and disorders.

2.7 Feature Selection

Feature selection is the most important part in this work. Feature selection will

decide the features for a healthy kidney from US images. Feature selection

techniques are applied to choose as many image parameters as possible to identify

17

the image characteristic, e.g the kidney. A feature selection technique will select few

of those extracted features which are most significant and which describe the kidney

characteristic the best.

In the previous work done, Wan Mahani Hafizah et al. [19] the features

selection is done by finding the difference of features value between the group of

normal kidney, bacterial infection kidney, cystic disease (CD) and kidney stones.

The features with higher different value from different group of kidney are choosing

as the features to classify the different classes of kidney. K.Bommanna Raja et al.

[10, 11] used statistical analysis, student t- Test which measures the significance of

features values in distinguishing kidney disorders. Karthik Kalyan et al. [22]

performed the feature selections that have high significance using Waikato

Environment for Knowledge Analysis (WEKA) software that gives variety of feature

selection options.

CHAPTER 3

METHODOLOGY

3.1 Introduction

The study conducted for this project as well as the framework for whole research is

described in this chapter. This section will give details on the methodology of the

project which divided into four parts. The details of each part will be discussed in the

sub-section in this chapter. The methods used for this study is summarized by the

flow chart in Figure 3.6.

3.2 Research Purpose

The purpose of the research carried out for this thesis is to investigate the texture

feature for a normal or healthy human kidney from ultrasound image. As already

mentioned in Chapter 1, the study is focusing on feature extraction and analysis of

different ultrasound image quality from three different US machine to choose the

best texture features that described the healthy kidney. With these features, the areas

containing kidney can be classified respectively. US kidney image had to be

acquired as well as a developed using MATLAB software along with the image

processing toolbox and SPSS software for statistical analysis.

19

3.3 Methodology

The flow process of the overall framework that will carry out for this study is

summarized in Figure 3.1. The details of each parts of the study are described as

follows:

3.3.1 Image Acquisition

For the measurements, three types of US machine is necessary that is used to obtain

human healthy kidney image with different image quality. Two types of US machine

used in this study (GE Healthcare (LOGIQ P5) & Philips (HDII XE)) belong to

Radiology Department Kolej Kemahiran Tinggi MARA (KKTM), Ledang while the

other machine (Toshiba Nemio XG (SSA-580A) is belong to Medical Electronics

Department, Electrical & Electronic Faculty, Universiti Tun Hussein Onn Malaysia

(UTHM). Later in this thesis, Toshiba Nemio XG (SSA-580A) is named as Machine

1, GE Healthcare (LOGIQ P5) is named as Machine 2 and Philips (HDII XE) is

named as Machine 3. The specifications and imaging technologies for these three US

machine are different. Three ultrasound imaging technologies have hit the

mainstream market that have the most impact on ultrasound image quality. These

technologies are Tissue Harmonics Imaging, Compound Imaging and Speckle

Reduction Imaging [33].

Tissue Harmonic Imaging is the common technology used in the advanced

US machine. The function of the harmonic imaging is to identify the body tissues

and will help to reduce artifacts or noise in the image. It works by sending and

receiving signals at two different ultrasound frequencies. This improves image

quality because body tissue reflects sound at twice the frequency that was initially

sent, which results in a cleaner image that better displays body tissue without extra

artifact. For example if the transducer emits the signal at 2MHz, it would ‘listen’ to a

4Mhz frequency, meaning double the frequency used.

Speckle Reduction Imaging uses an algorithm to identify strong and weak

ultrasound signals. Most manufacturers use terms as SRI, SRI HD, XRes, iClear,

Adaptive Speckle Reduction, MView, SCI, SonoHD, ApliPure+, TeraVision and

SRF for this technology. This technology works by evaluating the image on a pixel-

20

by-pixel basis, it attempts to identify tissue and eliminate the speckle noise that is

commonly degrade the US image .The weak signals will be eliminated while the

strong signals are enhanced/brightened that will result a smooth, clear and clean

image.

Compound imaging is a technique of combining three or more images from

different steering angles into a single image. The ultrasound will send signals at

multiple angles allowing it to see tissue from different angles as well as eliminating

the artifact. The traditional imaging technique is sending the ultrasound from the

transducer only in a single line of sight, meaning sending a sound signal

perpendicular to the probe head, then listens for each echo. Most manufacturers use

terms as CrossXBeam, CRI, SonoCT, iBeam, OmniBeam, XView, SonoMB,

ApliPure and Spatial Compounding for compound imaging technique.

The comparison of the three US machine that been used in this work is listed

in the Table 3.1.

Table 3.1: Comparison of US machines specifications

US MACHINE MODEL SPECIFICATIONS

Machine 1

Toshiba Nemio XG

(SSA-580A)

Machine 2

GE Healthcare

(LOGIQ P5)

Machine 3

Philips

(HD11 XE)

Compound Imaging

-not provide the compound

imaging technology

Compound Imaging

-provide a CrossXBeam

Imaging that helps

enhance tissue and border

differentiation

Compound Imaging

-provide SonoCT beam-

steered compound

imaging in both transmit

and receive modes. It

21

acquires multiple lines of

sight simultaneously,

compounds them in real

time and displays clear

images.

Tissue Harmonic Imaging

-provide tissue harmonic

imaging technology

without any advanced

technology

Tissue Harmonic

Imaging

-provide Phase Inversion

Harmonics for tissue

harmonic imaging

technology that helps

produce higher spatial

resolution and deeper

penetration

Tissue Harmonic

Imaging

-provide tissue harmonic

imaging with Pulse

Inversion technology for

producing pure,

broadband harmonic

signals for superb

grayscale image.

Speckle Reduction

Imaging

-not provide this

technology

Speckle Reduction

Imaging

-provide High Definition

Speckle Reduction

Imaging (SRI-HD)

that eliminate noise while

maintaining true tissue

architecture

Speckle Reduction

Imaging

-provide XRES adaptive

image processing

technology that

enhancing borders and

margins

Image of human healthy kidney is gathered from volunteer students and staff. All

subjects are considered as a healthy subject without anyone of them has been

diagnosed with any kidney abnormalities. For this study, only B-mode longitudinal

view of the kidney image will be taken. The total of 150 images will be measured

with 50 images for each type of US machine. The setting of all machine have to be

the same in order to get comparable data.

In this work, the setting for all the three machines is as follows:

Transducer: Convex probe with frequency 3.75MHz

US Machine Frequency: 6MHz

22

3.3.2 Image pre-processing

Image prepossessing techniques are used to select and enhance the region of interest

(ROI) and to eliminate erroneous data, which is of no interest from the acquired

images. The images are subjected to three steps of image pre-processing techniques

such as cropping, enhancement and segmentation.

3.3.2.1 Image Cropping

Cropping is an operation, which is performed on acquired images to accentuate the

ROI and to remove all the unwanted artifacts such as written labels and background

noise from them. Image cropping is needed to speed up further image processing. In

this study, manual cropping will be used where the image will cut in rectangular

shape which consist only the ROI.

3.3.2.2 Image Enhancement

Four filtering or enhancement techniques will be used in this study. The four

techniques are Wiener filter, Median filter, Gaussian Low Pass filter and Histogram

Equalization. These enhancement techniques are commonly used for enhance the US

image that is degraded by the speckle noise. These techniques have different

fundamental theories which include spatial filtering, frequency domain filtering and

histogram processing. The efficiency of all the filtering techniques will be evaluated

using MSE (Mean Square Error) and PSNR (Peak Signal to Noise Ratio) by

comparing the filtered image with the original image as reference. MSE and PSNR

have inverse relationship where higher value of PSNR will give a lower value of

MSE.

The description of each filtering technique is described as followings:

i. Wiener Filter:

The Wiener filtering is a linear type of filter. It helps in inverting

the blur. Wiener filter removes additive noise in the image. It

optimally minimizes the overall mean square error in the process

of inverse filtering and noise smoothing.

23

ii. Median Filter:

The median filter is one of the nonlinear filter types. The filtering

is performed by replacing the median of the gray values of pixels

into its original gray level of a pixel in a specific neighborhood.

The speckle noise as well as salt and pepper noise can be reduced

by using the median filter. The neighborhoods’ spatial extent and

the number of pixels involved in the median calculation decide to

what extent the noise can be reduced.

iii. Gaussian Lowpass Filter:

Gaussian filtering is a frequency domain filtering. In Gaussian

filtering, the smoother cutoff process is used rather cutting the

frequency coefficients abruptly. It also takes advantage of the fact

that the discrete Fourier Transform (DFT) of a Gaussian function

is also a Gaussian function. The Gaussian low-pass filter varies

frequency components that are further away from the image

center.

iv. Histogram Equalization:

It is used for improving the contrast where it brightens the image

appearance and gives an improve quality of the image. The

histogram equalization is applied to identify the maximum of the

intensity value.

3.3.2.3 Image Segmentation

Segmentation is a method to segment the kidney region. Segmentation will subdivide

an image into its constituent regions or object. The segmentation is achieved when

the region ROI of object is been isolated. In this work, manual contouring is used to

segment the kidney edge. The image is segment into foreground and background

where the mask is created in order to erase pieces of a binary image that are not

24

attached to the object surrounded by the boundary. The complicated background that

is outside ROI will be masked. This process will eliminate unwanted intensity values

which are outside the contour (edge) of the kidney image. It is to avoid the

calculation of these unwanted intensities that will be incorporated during extraction

of feature parameters.

3.3.3 Feature Extraction

Three statistical feature extractions namely Intensity Histogram (IH), Gray-Level co-

Occurrence Matrix (GLCM) and Gray-Level Run Length Matrix (GLRLM) will be

used to extract the features of the kidney. Each type of features is described as

follows:

3.3.3.1 Intensity Histogram Features

The intensity-level histogram is a function showing the number of pixels in the

whole image, which have this intensity. The 8-bit gray scale image is having 256

possible intensity values. The parameters in the following statistical formulas are p

that represents the pixel intensity, p(i) represents the pixel intensity at i value and N

represents total number of pixels. Five individual features under this feature

extraction technique can be calculated utilizing the formulas in equation 3.2 to 3.6

[25].

p(i) = Number of pixels with grey level i

N (3.1)

Mean (µ)

A measure of the average intensity value of all pixels

Mean(μ) = � ip(i) (3.2)

���

���

53

REFERENCES

1. K.Bommanna Raja, M.Ramasubba Reddy, et al., Study on Ultrasound Kidney

Images for Implementing Content Based Image Retrieval System using

Regional-Grey Level Distribution, Proc. of International Conference on

Advances in Infrastructures for Electronic Business, Education, Science,

Machine and Mobile Technologies on the Internet, Vol.93, 2003.

2. Nadine Barrie Smith, Andrew Webb, Introduction to Medical Imaging:

Physics, Engineering and Clinical Applications, Cambridge University Press,

UK, 2011.

3. J.Ravell, M.Mirmehdi and D.McNally, Applied review of Ultrasound Image

Feature Extraction Method’, Proc.of 6th Medical Image Understanding and

Analysis Conference, pp.173-176, 2002

4. Carmen Mariana Nicolae and Luminita Moraru, Image Analysis and Kidney

using Wavelet Transform, Annals of University of Craiova, Mathematics and

Computer Science Series, Vol 38(1), pp. 27-34, ISSN:1223-6934, 2011

5. Prema T.Akkasaligar, Savitri S.Unnibhavi, Identification of Kidney in

Medical Ultrasound Images, Proceeding of 5th SARC-IRF Intenational

Conference, ISBN:978-93-84209-13-1, pp.85-90, 2014

6. Lilian H Y Tang, Rudolg Hanka and Horace H S LP, A review of intelligent

content-based Indexing asnd Browsing of Medical Images, Health

Informatics Journal, 5(1), pp. 40-49, 1999

54

7. Gabor.D, Theory of Communication, Journal of the Institute of of Electrical

Engineers, pp 429-457, 1993,

8. Glodny, B., Unterholzner, V., Taferner, B., Hofmann, K., Rehder, P., Strasak,

A. and Petersen, J. Normal kidney size and its influencing factors - a 64-slice,

MDCT study of 1.040 asymptomatic patients. BMC Urology, 2009.

9. Eko Supriyanto, Wan Mahani Hafizah Yeoh Jing Wui and Adeela

Arooj,Automatic Non-Invasive Kidney Valume Measurement based on

Ultrasound Image, IEEE Microwave and Wireless Components Letters, vol.

20, no. 9, Sept. 2010.

10. K.B.Raja, M.Madheswaran, and K.Thyagarajah, Analysis of Ultrasound

Kidney Image using Content Descriptive Multiple Features for Disorder

Identification and ANN based Computing, Proc.of the International

Conference on Computing:Theory and Application, 2007.

11. K. B. Raja, M. R. Reddy, S. Swaranamani, S. Suresh, M.Madheswaran, and

K. Thyagarajah, Study on Ultrasound Kidney Images for Implementing

Content Based Image retrieval System using Regional Gray-Level

Distribution, Proc. of International Conference on advances in

infrastructures for electronic business, education, science, medicine, and

mobile technologies on the internet, vol. 93, 2003.

12. Wan Mahani Hafizah, Eko Supriyanto, Comparative Evaluation of

Ultrasound Kidney Image Enhancement Techniques, InternationalJournal of

Computer Application, Vol 21, No.7, 2011.

13. Shrimali, V,Anand, R.S Kumar, V.2010. Comparing the performance of

ultrasonic liver image enhancement techniques, a presence studies, IETE

Journal of Research, Vol 56, Issue 1.

14. Yu, Y.Acton, S.T. 2002. Speckle Reducing Anistrophic Diffusion, IEEE

Trans on Imag Process, Vol 11, pp 1260-1270.

15 Wan M.hafizah and Eko Supriyanto, Automatic Generation of Region of

Interest for Kidney Ultrasound Images using Texture Analysis, International

Journal of Biology and Biomedical Engineering, Issue 1, Vol.6, pp. 26-34,

2012

55

16. Prema T.Akkasaligar and Sunanda Biradar, Classification of Medical

Ultrasound Image of Kidney, International Journal of Computer

Applications,pp 24-28, 2014

17. Tanzila Rahman and Mohammad Shorif Uddin, Speckle Noise Reduction and

Segmentation of Kidney Regions from Ultrasound Image, International

Conference on Informatics, Electronics & Vision (ICIEV), 2013

18. Amira A. Mahmoud, S. EL Rabaie, T. E. Taha, O. Zahran, F. E. Abd El-

Samie, Comparative Study of Different Denoising Filters for Speckle Noise

Reduction in Ultrasonic B-Mode Images, I.J. Image, Graphics and Signal

Processing, Vol. 2, pp.1-8, 2013

19 Wan Mahani Hafizah, Eko Supriyanto,Jasmy Yunus, Feature Extraction of

Kidney Ultrasound Images Based on Intensity Histogram and Gray Level Co-

occurance Matrix,Sixth Asia Modelling Symposium,2012

20. K. B. Raja, M. R. Reddy, S. Swaranamani, and S. Suresh,Analysis of Kidney

Disorders using Ultrasound Images by Geometric Moments, Biomedical

Engineering: Recent Developments, pp. 153 -154, 2002.

21. C. Karthikeyini, K. B. Raja, and M. Madheswaran, Study on Ultrasound

Kidney Images using Principal Component Analysis: A Preliminary Result,

Proc. Of Fourth ICVGIP, pp. 190 – 195

22. Karthik Kalyan, Binal Jakhia et al., Artificial Neural Network Application in

the Diagnosis of Disease Conditions with Liver Ultrasound Images, Advances

in Bioinformatics, Hindawi Publishing Corporation, Vol.2014, Article ID

708279, pp.1-13, 2014.

23. Haryalli Dhillon, Gagan Deep Jindal, Akshay Girdhar, A Novel Threshold

Technique for Eliminating Speckle Noise in Ultrasound Images, 2011

International Conference on Modelling, Simulation and Control, Vol.10,

pp.128-136, 2011

24 M. Vasantha, V. S. Bharati, R. Dhamodharan, Medical Image Feature,

Extraction, Selection And Classification, International Journal of

Engineering Science and Technology, vol. 2, pp. 2071-2076, 2010

25. H.S.Sheshadri, A.Kandaswamy, Experimental investigation on breast tissue

classification based on statistical feature extraction of mammograms,

Computerized Medical Imaging and Graphics, No.31, pp.46-48, 2007.

56

26. R.M.Haralick, Stastistical and Structural Approaches to Texture, 1979. In

Proc. of IEEE, Vol. 67, No.5: 786 –804

27. L. Soh and C. Tsatsoulis, Texture Analysis of SAR Sea Ice Imagery Using

Gray Level Co-Occurrence Matrices, IEEE Transactions on Geoscience and

Remote Sensing, vol. 37, no. 2, March 1999.

28. D A. Clausi, An analysis of co-occurrence texture statistics as a function of

grey level quantization, Can. J. Remote Sensing, vol. 28, no.1, pp. 45-62,

2002

29. A.K Mohanty, S. Beberta, S.K Lenka, Classifying Benign and Malignant

Mass using GLCM and GLRLM based Texture Features from Mammogram,

International Journal of

Engineering Research and Applications (IJERA), Vol. 1, Issue 3, pp.687-693

30. RM.M Galloway, “Texture Analysis Using Grey-Level-Run-Length”,

Computer Graphics and Image Processing, Vol.4, pp.172-179,197

31. Fritz Albregtsen. Statistical Texture Measures Computed from Gray Level

Run Length Matrices. Image Processing Laboratory, Department of

Informatics, University of Oslo, Nov. 1995

32. Gupta S.P, Statistical Methods, Sultan Chand and Sons, 1991

33. Available: http://www.providianmedical.com/ultrasound-imaging-guide/

34. Gundivala V N and Raghavan V V,"Content-based image retreival systems",

IEEE Computer, 28 (9), pp 18-22, 1995

35. N. Zulpe, V. Pawar, GLCM Textural Features for Brain Tumor

Classification, IJCSI International Journal of Computer Science Issues, Vol.

9, Issue 3, No 3, May 2012

36. M.Harsha, S.Visweswara, GLCM Architecture for Image Extraction,

International Journal of Advanced Research in Electronics and

Communication Engineering (IJARECE) Volume 3, Issue 1, January 2014

37. A. Gebejes, R. Huertas, Texture Characterization based on Grey-Level Co-

occurrence Matrix, Conference of Informatics and Management Sciences

March, 25. - 29. 2013

38. D. p Tian, A Review on Image Feature Extraction and Representation

Techniques, International Journal of Multimedia and Ubiquitous

Engineering Vol. 8, No. 4, July, 2013

57

39. R.Rosa, F.C Monteiro, Computational Vision and Image Processing IV,

Taylor & Francis Group, London, ISBN-978-1-138-00081-0, 2014

40. J. Alison Noble, Ultrasound Image Segmentation: A Survey, IEEE

Transactions on Medical Imaging, Vol. 25, No. 8, Aug 2006

41. A. M. Khan, Ravi. S, Image Segmentation Methods: A Comparative Study,

International Journal of Soft Computing and Engineering (IJSCE) ISSN:

2231-2307, Volume-3, Issue-4, Sept 2013

42. R.Nithya, B.Santhi, Comparative Study on Feature Extraction Method for

Breast Cancer Classification, Journal of Theoretical and Applied Information

Technology, Vol. 33, pp. 220-226, 2011

43. Namita Aggarwal, R. K. Agrawal, First and Second Order Statistics Features

for Classification of Magnetic Resonance Brain Images, Journal of Signal

and Information Processing, 2012, 3, 146-153, 2012