Semi-automatic Detection of Liver Lesions in Computed Tomography Images Using Digital Image Processing Techniques Nebras Ahmed Mohamed Musa B.Sc in Me

Semi-automatic Detection of Liver Lesions in Computed Tomography Images Using Digital Image Processing Techniques Nebras Ahmed Mohamed Musa B.Sc in Medical Instrumentation, University of Gezira (2016) A Dissertation Submitted to the University of Gezira in Partial Fulfillment of the Requirements for the Award of the Degree of Master of Science in Biomedical Engineering Department of Electronic Engineering Faculty of Engineering and Technology November /2018 1

Semi-automatic Detection of Liver Lesions in CT- Images using Digital Image Processing Techniques and Otsu s Thresholding Algorithm Nebras Ahmed Mohmed Musa Supervision Committee: Name Position Signature.. Main Supervisor..... Co-supervisor..... Co-supervisor... Date: October/2018 2

Semi-automatic Detection of Liver Lesions in CT- Images using Digital Image Processing Techniques and Otsu s Thresholding Algorithm Nebras Ahmed Mohmed Musa Examination Committee: Name Position Signature.. Chair Person..... External Examiner..... Internal Examiner... Date of Examination: day/november/2018 3

DECLARATION My name is nebras ahmed Mohamed Musa, I graduate from university of Gazira This dissertation is a presentation of my original research work. Wherever contributions of others are involved, every effort is made to indicate this clearly, with due reference to the literature, and acknowledgement of collaborative research and discussions. 4

DEDICATIONS إىلذؿنذجعلذاهللذاجلةذحتتذأؼداؿفنذ...ذووصىذبفنذدقدذاخلؾقذثالثةذدلؽافن.. إىلذؿنذأعجزنذاألدباءذواؾجباءذواؾشعراءذيفذوصػفنذوإعطائفنذوؾوذجزءذقدريذؿنذػقضذ حافن... أؿفاتاذثمذأؿفاتاذثمذأؿفاتا إىلذاؾبحورذاؾزاػرقنذ...ذواحلؿاةذاؾدافرقنذ...ذواألوارذاؾبافرةذ...ذواؾرجالذاؾطافرقنذ...ذ ؾؾصعابذدائؿا ذؼافرقنذ... إؾيذآبائا إىلذؿنذفمذؽؾاذوبعضاذ...ذدداذوعضداذ...ذقػرحونذؾػرحاذوحيزونذحلزاذ...ذوقؼضونذ أغراضاذ أذؼائاذوذؼقؼاتا إىلذؽلذؿنذصاغذحرػا ذ...ذوػؽرذػؽرا ذ...ذعؾؿا ذوأخالؼا ذوأدبا ذ...ذوعؿؾواذؾرػعةذاإلدانذواإلداقةذ...ذ وعؿارةذاألرضذ...ذصاعذاؾتارقخذذ..ذأداتذتاذاؾؽرام فديذفذاذاجلفدذادلتواضع إىلذمشسذاؾعؾمذاؾيتذالذتعرفذادلغقبذوؿارةذاؾعؾمذيفذارضذاؾورنذاحلبقبذجاؿعةذاؾصؿودذواؾتؿقزذ اؾعؾؿيذاؾرذقد جاؿعةذاجلزقرة. 5

AKNOWLEDGMENTS ؼالذاهللذتعاؾيذيفذحمؽمذتزقؾهذ: ذ ذ" ل ئ ن ش ك ر ت م ل أ ز يد ن ك م و ل ئ ن ك ف ر ت م إ ن ع ذ اب ي ل ش د يد " ذصدقذاهللذاؾعظقم إذاذحتؾىذادلدؾمذبصػةذاؾشؽرذؾربهذعؾىذعؿهذػذؾكذدقضؿنذؾهذدعةذيفذاؾرزقذواؾعمذ ؿنذؽلذجاب ذػادلدؾمذفوذاؾذيذقؼ د رذادلعروف ذوقشؽرذاهللذدبحاهذوتعاىلذعؾىذ عؿهذؽؾفا ذوقشؽرذاؾاسذعؾىذؿاذقؼدؿوهذؾهذؿنذخريذ. ذ قعتربذادلعؾمذفوذمبثابةذاألبذؾؾطاؾبذ ذألهذالذقعؾؿهذػؼطذاؾدروسذوذادلذاؽرةذبلذ قعؾؿه ذاألخالق ذو ذاؾصػات ذاحلدة ذػاؾرتبقة ذو ذاألخالق ذاحلؿقدة ذفي ذعوان ذؽلذ راؾبذجمتفدذوذشقطذوذاجحذيفذدرادته ذوذؿفؿاذؽتباذوذؼؾاذؾنذدتطقعذردذ مجقلذادلعؾؿنيذوذادلعؾؿاتذعؾقاذوذؾنذدتطقعذذؽرفمذؽؿاذجيبذوذؽؿاذقدتحؼونذ ػفم ذتعبوا ذؽثريا ذحتى ذدرس ذو ذتعؾم ذو ذصبح ذأذخاص ذممقزون ذوذاجحنيذيفذ حقاتاذ ذوذالذميؽاذإالذأنذشؽرفمذبجاحاذوذتػوؼاذوذالذضقعذجفودفمذفدرا ذ وذأنذحنرتؿفمذوذشؽرفمذببعضذاؾؽؾؿاتذاؾيتذالذتويفذحؼفمذوذجمفودفمذ. 6

ABSTRACT Semi-automatic Detection of Liver Lesions in CT- Images using Digital Image Processing Techniques and Otsu s Thresholding Algorithm Abstract The incidence of liver lesions has increased dramatically in recent years. Early diagnosis is the best solution for rescuing patients exposed to this type of cancer, which opens the possibility of treatment.there are a number of techniques that are used in the treatment of digital medical images, but segmentation is the best technique, the technique used in this research depends on the calculation of the value of intensities of anatomical structures of the liver in the so called histogram to determine the extent of intensities of the infected and non-infected regions this method called Otsu s method. Otsu's method uses an exhaustive search to evaluate the criterion for maximizing the between-class variance First all the noises had been removed using smoothing and shapening filters the use of two filters to optimize because smoothing out tissue variation would remove all image details that are needed for proper segmentation. Second the histogram had been calculated which represent the intensities of each elements in the image,then using mathematical equations the value of the main key had been obtained in the detection of the affected regions.after segmention done, morphological operations used to enhance the lesions boundaries.to simplify the understanding and use of this method, a user interface is designed for users who are not require to know technical details. This method is effective in detecting of liver lesions but it take time in the process of calculation, so Otsu work on to develop its technique to new one name it 'two dimensional Otsu s method' and characterized by effective and less calculation time, this is the next step for those who want development in the way of my research. 7

عنوان البحث: الكشف الشبه ألي عن سرطان الكبد في صور االشعه المقطعية باستخدام تقنيات معالجه الصور الرقمية و خوارزميه أوتس للعتبة. الباحث: اسم نب ارس أحمد دمحم موسى. ملخص الد ارسة يعذالخ اإلصاتح تآفاخ انكثذ ذرضا ذ تشكم كث ش ف انس االخ ش عذ انرشخ ص ان ثكش أفضم دم انز ساعذ ف إ قار ان شض ان عشض نالصات ت زا ان ع ي انسشطا اخ د ث فرخ ان جال أياو إيكا ح انعالج. ذ جذ عذد ي انرق اخ انر ذسرخذو ف يعانج انص س انشق ح انطث ح ذعذ طش ق انرقس ى ي أفضم ز انرق اخ ذق انرقس ى ان سرخذو ف زا انثذث ذعر ذ عه دساب ق كثافاخ انرشاك ة انرشش ذ ح نهكثذ رنك نرذذ ذ يذ كثافاخ ان اطق ان صاتح انغ ش يصات نهفصم ت ا ذس ز انطش قح تطش ق أ ذس. طش ق أ ذس ذسرخذو انثذث انشايم نرق ى يا س ت ع اس ذعظ ى انرثا ت انطثقاخ. أ ال ذى انرخهص ي كم أ اع انض ضاء تإسرخذاو فهرش انر ع ى فهرش انذذ ( جعم انذذ أكثش ض دا( انثذث ع األفضم ت ى ال ذ ع ى إخرالفاخ اال سجح س ض م ذفاص م انص س ان طه ت نهرجضئ انصذ ذ نكم ع صش ف انص س. ثا ا دساب انشسى انث ا انز ثم كثافاخ كم ع صش ف انص س. ثانثا تاسرخذاو يعادالخ س اض ذصم عه ق انعرثح انر ذ ثم ان فراح انشئ س ف ع ه انرقس ى. نرثس ظ ف ى اسرخذاو ز انرق ح ذى ذص ى اج ص س نه سرخذي انغ ش يهضي ت عشف انرفاص م انرق ح. ذعرثش طش ق أ ذس انرقه ذ ح فعان ف انكشف ع أفاخ انكثذ إال أ ا ذؤخز قد ط م ف ع ه اخ انذساب نزنك قاو انعانى أ ذس ترط ش طش قر قاو تطش ق جذ ذ أس ا ا طش ق أ ذس ث ائ انثعذ انر ذر ض تكفاءج عان صي دساب أقم. ذ 8

TABLE OF CONTENTS DEDICATIONS ACKNOWLEDGEMENTS... ABSTRACT... يهخص انذساسح TABLE OF CONTENTS LIST OF FIGURES LIST OF ABBREVIATIONS CHAPTER ONE INTRODUCTION.. 1.1 Introduction. 1.2 Problem Identification. 1.3 Motivation... 1.4 Research Objectives.. 1.5 Research Methodology... 1.6 Dissertation layout CHAPTER TWO. BACKGROUND AND LITERATURE REVIEW. 2.1 Background.. 2.1.1 Liver... 2.1.1.1 Anatomy. 2.1.2 Liver Tumors. 2.1.3 Computed Tomography 9

2.1.3.1 Background.. 2.1.3.2 Basic Principle.. 2.1.3.3 CT Image Noise and Artifacts.. 2.1.3.4 Contrast Media. 2.1.3.5 CT-abdominal Image... 2.1.3.5.1 Small lesion detection 2.1.3.5.2 Hepatocellular carcinoma (HCC).. 2.1.3.5.3 Effect of some parameters on the density in CT-images 2.1.4 Segmentation 2.1.4.1 The segmentation challenge.. 2.1.5 Thresholding Algorithm 2.1.5.1 Global Thresholding.. 2.1.5.2 Local Thresholding 2.1.5.1.1 Otsu s Thresholding method. 2.1.6 Graphic User Interface. 2.1.6.1 Background 2.1.6.2 User interface and interaction design 2.1.6.3 GUI in Mat-lab Program 2.1.6.3.1 Why use a GUI in MATLAB?... 2.1.6.3.2 Main Steps of Designing a Simple GUI.. 2.2 Literature Review. 2.2.1 Previous Studies. 2.2.2 Summary of Literature Review. CHAPTER THREE RESEARCH METHODOLOGY... 3.1 Introduction.. 11

3.2 Research Stages... 3.2.1 CT Image Acquisition. 3.2.2 Image Pre-processing 3.2.3 Thresholding Algorithm 3.2.4 Lesions Segmentation... 3.2.5 Graphic User Interface. CHAPTER FOUR.. RESULTS AND DISCUSSIONS.. 4.1 Results.. 4.1.1 Abdominal CT-Images 4.1.2 Image Pre-processing.. 4.1.3 Liver Segmentation. 4.1.4 Thresholding Ahgorithm. 4.1.5 Lesions Segmentation. 4.1.6 Graphic User Interface (GUI). 4.2 Discussions.. CHAPTER FIVE... CONCLUSION AND RECOMMENDATIONS. 5.1 Conclusion. 5.2 Recommendations. REFERENCES 11

LIST OF FIGURES Figure 2.1: Figure 2.2: Figure 2.3: Figure 2.4: Figure 2.5: Figure 2.6: Figure 2.7: Figure 2.8: Figure 2.9: Figure 2.10: Figure 2.11: Figure 3.1: Figure 3.2: Figure 3.3: Figure 3.4: Figure 3.5: Figure 4.1: Figure 4.2: Figure 4.3: Figure 4.4: Figure 4.5: 12

Figure 4.6: Figure 4.7: Figure 4.8: Figure 4.9: Figure 4.10: Figure 4.11: Figure 4.12: 13

LIST OF ABBREVATIONS HCC Hepatocelluercarcinoma. WHO World Health Organization. CT Computed Tomography. MRI Magnetic Resonance Imaging. 2D Two Dimensions. 3D Three Dimensions. GUI Graphic User Interface. GLIs Graphic Line Interfaces. HUD Head Up Display. ATM Automatic Teller Machine. POS Point of Sale. RTOS Real Time Operating System. FCN Fully Connected Network. ANN Artificial Neural Network. CNN Convolutional Neural Network. CAD Computer Aided Diagnosis. F CM Fuzzy C-Mean. VOI Volume of Interest. VOE Volumetric Overlap Error. SVM Support Vector Machine. 14

MRF Markov Random Field. RF Random Forests. DSC Dice Similarity Coefficient. CLAHE Contrast Limited Adaptive Histogram Equalization. CVHE Constrained Variable Histogram Equalization. ROI Regions of Interest. 15

CHAPTER ONE INTRODUCTION 1.1 Introduction When healthy cells change and grow out of control, they form a mass called a tumor. A tumor can be malignant or benign. The cancerous tumor is malignant, meaning it can be grow and spread to other parts of the body. A benign tumor means the tumor can grow but will not spread (Schraml et al., 2015). Tumor or Cancer is the abnormal growth of cells or tissues in the organ. Liver Cancer is a common type of cancer that affects the largest organ in the abdomen, the liver. There are two types of liver cancer namely Primary Liver Cancer and Secondary Liver Cancer. Primary liver cancer is the cancer that originates from the liver itself. This type of liver cancer is also known as hepatocellualr carcinoma (HCC) or hepatoma. It is the fifth most frequent cancer form in the world and third leading cause of cancer death (Rafiei et al., 2018). HCC cases mostly occur in Asia and Africa, but its number is increasing rapidly in U.S. and other western countries. Secondary liver cancer is the cancer which originates in other organs but then spreads to liver. Secondary liver cancer is also known as metastatic liver cancer. As per data from the World Health Organization (WHO), less than 30 cases per 100,000 people worldwide die as a result of liver cancer, with high rates in parts of Africa and Eastern Asia. So, it is clear that the major cause of death in human is considered to be liver tumor. There are various risk factors that can lead to liver cancer such as Hepatitis B infection, Hepatitis C infection, excessive alcohol consumption, diabetes and obesity (Schraml et al., 2015). The diagnostic accuracy of ultrasound, computed tomography (CT), magnetic resonance imaging (MRI) and angiography is dependent on a number of variables: expertise of the operator (especially with ultrasound), sophistication of equipment and technique, presence of cirrhosis and, most importantly, experience of the interpreter. For small tumors (<2 cm), the diagnostic accuracy ranges from 60 80%. The diagnostic accuracy increases significantly with an increase in tumor 16

size, ultimately reaching 100% with very large tumors with all modalities (Rafiei et al., 2018). 1.2 Problem Identification Segmentation of liver lesion has to deal with structures of high variability in a noisy medium where the structure differ by complex textures changes. Many techniques had been developed such as region-growing methods, classifier methods but every technique has its limitations region-growing methods ignore global spatial structure and focus on local features, classifier methods ignore spatial representation of the segmented region. Fully automatic and manual segmentation methods represent two ends of a wide spectrum of segmentation algorithms. The failure of most of automatic techniques to segment complex structures (such as liver-lesion) on one hand, and the practical infeasibility of manual segmentation on the other hand, have motivated the invention of a semiautomatic tools between these two extremes. 1.3 Motivation To develop and implement a semi-automatic software system to detect liver lesions, which is characterized by feasibility, accuracy and the ease of use graphic interface ; to assess viability of high-quality semi-automatic segmentation of lesions and feasibility of introducing it into everyday practice, to estimate the volume of a lesions with high accuracy compared to the given ground-truth segmentation, and estimation in less than a minute for a radiologist familiar with the system. It should be an intuitive and robust procedure. 1.4 Study Objectives The aim of this study is to diagnosis early stage of liver lesions to increase the information of proper diagnose, by designing an effective and accurse method to detect the lesions. Also use the imaging of liver lesions to serve several roles included the evaluation of suspected lesions, the choice of a therapeutic strategy, the planning of treatment and post-treatment follow-up. 17

There are many specific objectives such as:- To filter acquired image to enhance its resolution and reduce the noise. To design simple method to compute histogram and probabilities of each intensity level to separate regions using the set threshold which applied for all pixels of the image. 1.5 Study Methodology The methodology included collection, acquisition of images from CT-scan machine, process the images to remove the noises then liver region segmented from full abdominal images after that lesions segmented from liver images using thresholding techniques called Otsu s method finally all last steps introduce simply in graphic user interface (GUI) platform. The methodology stages are executed using Matrix Laboratory (MAT-LAB) R2014b program. 1.6 Dissertation layout This dissertation is organized as follows: Chapter 1 includes the Introduction, Chapter 2 about Background and Literature Reviews, Chapter 3 summarize the Methodology, Chapter 4 presents the Results and Discussions, Chapter 5 include the Conclusion and Recommendation,finally used References had been mentioned. 18

CHAPTER TWO BACKGROUND AND LITERATURE REVIEW 2.1 Background This section introduces liver anatomy, liver tumor, computed tomography (CT), Segmentation, Thresholding algorithm and Graphic User Interface (GUI). 2.1.1 Liver The liver is the most voluminous solid organ inside human body (3 % of body mass). This organ lays on the upper right part of the abdominal cavity, below the lungs and the heart, and to the right of the stomach, intestine and spleen as shown in figure 2.1. It has many roles detoxifies various metabolites, synthesizes proteins, and produces bio-chemicals necessary for digestion. Its other roles in metabolism include the regulation of glycogen storage, decomposition of red blood cells and the production of hormones. Figure 2.1: Abdominal Structure (Uhlén et al., 2015). 19

2.1.1.1 Anatomy The liver is highly variable organ that contains many vascular networks. It also has a highly variable shape that consists of several anatomical segments. In the widely used Couinaud system, the functional lobes are further divided into a total of eight sub-segments based on a transverse plane through the bifurcation of the main portal vein. The caudate lobe is a separate structure which receives blood flow from both the right- and left-sided vascular branches. The Couinaud classification of liver anatomy divides the liver into eight functionally independent liver segments. Each segment has its own vascular inflow, outflow and biliary drainage. In the centre of each segment there is a branch of the portal vein, hepatic artery and bile duct. In the periphery of each segment there is vascular outflow through the hepatic veins as shown in figure 2.3. The classification system uses the vascular supply in the liver to separate the functional units (numbered 1 to 8), with unit 1, the caudate lobe, receiving its supply from both the right and the left branches of portal vein. It contains one or more hepatic veins which drain directly into the inferior vena cava. The remainders of the units (2 to 8) are numbered in a clockwise as fashion as shown in figure 2.2 below (Uhlén et al., 2015). Figure 2.2: Segmented Lobes of the Liver (Uhlén et al., 2015). As a soft organ, the liver shape indeed depends on the interactions with the other abdominal organs. Thus the organ variability is high. Moreover, this variability even increases with many types of pathologies. The liver contains many vascular networks partially because of its specific location 21

at the interface between the circulatory and the digestive systems. These numerous networks make liver a challenging organ for surgery and thus preoperative planning is required before any liver operation. The liver indeed contains five vessel networks, three of which are blood networks. First, the liver receives blood from two blood networks that are very close or even interwoven, the portal vein and the hepatic artery. The former supplies the liver with nutrients coming from the intestine, and the latter provides the liver with oxygen. Then, the blood is drained from the liver by the hepatic vein. One may note the tree-like shape of the vessels, and their interwoven organization. The vessels are indeed trees as there are no loops inside each single network. Moreover, these networks are also interwoven, in particular the hepatic artery and the portal vein. Finally, a lymphatic network and a biliary network are also present inside the liver. The former does not impact the study as the lymphatic vessels remain invisible on CT images. On the opposite the biliary ducts are sometimes visible as hypodense regions. In particular, a dilatation of the intrahepatic bile ducts happens for several pathologies. This dilatation appears on the CT images because ducts become bigger than usual and thus more visible (Wu et al., 2017). Figure 2.3: Anatomical Structure of the liver (Couinaud and Le,1957). 21

2.1.2 Liver Tumors Healthy cells reproduce and die in a stable and orderly manner, under some cases however, a group of cells start to grow uncontrollably and produce a tumor. Tumors can be benign, lacking the capacity to spread to other organs, or malignant. The malignant tumor display uncontrollable growth and may invade and destroy healthy surrounding tissue or even spread to other organs and continue the destruction there as shown in figure 2.4. Spread tumors are called metastases. Tumors in the liver may belong to the same categories (Rajput, 2017). The most common primary hepatic tumor is Hepatocellular carcinoma or HCC which accounts for 80-90% of the cases. The prevalence of HCC varies considerably across the globe. Incidence is highest in some Asian and sub-saharan countries (30-150/100.000). In Western Europe, the USA and Australia the incidence is mostly low (around 1-3/100.000), but it seems to be increasing globally. HCC is also much more common in men than in women (male to female ratio is approximately 4-8:1. Risk factors for HCC are chronic liver diseases such as cirrhosis, hepatitis B and C or exposure to chemical toxins such as aflatoxins. This is also the cause for the varying prevalence. In Europe and the USA the most common underlying cause is alcohol caused cirrhosis, while hepatitis B is the cause for the high prevalence in sub-saharan countries. The liver is one of the organs where tumor metastases most commonly appear (Wu et al., 2017). Figure 2.4: Light Portion of Hepatocelluercarcinoma(HCC) (Wu et al., 2017). 22

Liver metastases are 20 times more common than primary malignancies. It is therefore common to screen cancer patients for metastases in the liver as the presence of liver metastasis greatly affects patient survival. Metastasis can appear in all parts of the liver. Surgery to remove the diseased part of the liver is the main and most effective treatment for primary liver cancer in non-cirrhotic liver. The liver has a remarkable property to carry on as normal even if a large part of it is removed. In the case of a cirrhotic liver the best alternative is to transplant the whole liver. There are however many cases when surgery is not possible; when the tumor has grown too large or spread to other parts of the body, or when the liver is in a bad condition, as is often the case for HCC (depending on underlying cirrhosis or other chronic liver diseases). The early signs for tumors in the liver are often hard to detect, either there might not be any symptoms or the symptoms may be very diffuse. Hence, when the affected turns to the hospital with symptoms such as a painful and swollen upper abdomen, weight loss, jaundice and fewer, the tumor might be quite advanced and too large to remove. Neither chemotherapy nor radiotherapy are considered effective as treatments, but can in some cases alleviate the symptoms and even decrease the size of the tumor. The high doses of radiotherapy needed to kill the cancer cells would for example also damage the healthy liver tissue. Local tumor ablation (local destruction of tumor tissue) by heating, cooling or injecting alcohol in the liver can sometimes be effective for small tumors. The estimation of tumor growth is very important to judge how successful a treatment has been. The patient is often examined with a CT-scan, where the tumor appears as a lighter or darker shade in the liver. The tumor might also be heterogeneous, e.g. with a darker core with a surrounding lighter ring. The tumor size is usually approximated as the largest axial diameter of each lesion, a process which is done manually. It lies several different problems in this approach, first of all a 3D measurement is approximated as the 1D diameter, and clinical study has implied that a 3D measurement would give a much better representation of the tumor response. The simplification could be valid only if the tumor always appeared perfectly 23

spherical, which is not the case. Secondly, the process suffers from the same problems that all manual approaches do, such as always being subjective (there is always inter- and intra person variations) and timeconsuming (Li et al., 2015). 2.1.3 Computed Tomography 2.1.3.1 Background Medical imaging is a specific field of Computer Vision that deals with medical images. During the last decades many tools have been developed for the imaging of the body, first producing 2D representations of the body (X-Ray, Ultra-Sound) and more recently 3D volumes assets of 2D images of the body. These imaging modalities, such as Computed Tomography (CT) or Magnetic Resonance Imaging (MRI), allow taking 3D images of the body in a non invasive way. Thus, medical images became increasingly important for medical practice. These images indeed contribute too many medical tasks such as diagnosis, follow-up of patients and evaluation of treatments. However, the amount of data produced by imaging modalities cannot be handled with manual approaches, especially since the introduction of 3D images. The current CT machines indeed allow acquiring 3D images as sets of 2D slices of the body, where each slice can measure less than a millimeter. This resolution implies that medical images acquired with recent clinical equipments will count lots of slices; the typical abdominal images will count around 500 slices, and full body images can count up to 2000 slices. Thus, techniques coming from computer vision have been introduced to extract information from the medical images. While many techniques from Computer Vision are used in medical imaging, medical imaging is also a domain on its own. Indeed, the methods and the problems differ significantly from usual Computer Vision, because of the specificity of the images and the particular nature of living organisms. In particular the 3D nature of the images requires specific algorithms in order to deal with the problems induced by 3D (e.g. partial volumes) and for taking advantage of the information brought by 3D; working on a 3D image is different from working independently on 24

each slice that composes this volume. In this study only CT images will be considered, thus the CT modality will be briefly introduced with a focus on the image acquisition and the tomography reconstruction. CT scanners are relatively new machines. Their conception indeed began in the late 1960s before production in the 1970s. Then, scanners quickly showed their contribution in the medical domain (Wu et al., 2017). The technology for computed tomography (CT) is developed in the 1970s. It was the first technology that made it possible to generate tomography images of the human body without using invasive methods. The CT technology was initially exclusively used for brain imaging, as it gave the possibility to study the soft brain tissue inside the dense skull, but was soon developed for imaging of the abdomen and thorax, the main areas of interests in this dissertation. The greatest technological improvements since the 1970s have primarily been the increase of spatial resolution and a reduction of the scan times. In fact the improvement of the spatial resolution has been so successful that we today find ourselves close to the theoretical limits of the detectors. This means that it is difficult to improve the contrast resolution more without increasing the radiation dose received by the patient. Today CT-images can be used to see the body from several different angles, cross-sections and also in 3D. The standard CT technology of today is the spiral-ct, where the detectors and the X-ray tube that emits the x-ray radiation is rotated around the patient at a high speed at the same time as the patient is moved through the CT device. The detectors measure the amount of radiation that reaches them. The image is formed from the calculated attenuation of the radiation when it passes through the body and reaches the different detectors (Wu et al., 2017). 2.1.3.2 Basic Principle The process of CT image acquisition involves the measurement of X-ray transmission profiles through a patient for a large number of views. A profile from each view is achieved primarily by using a detector arc generally consisting of 800 900 detector elements, referred to as a detector row. By rotation of the X ray tube and detector row around the patient, a large number of views can be obtained. The use of tens or even 25

hundreds of detector rows aligned along the axis of rotation allows even more rapid acquisition all as shown in figure2.5. Figure 2.5: CT Image Acquisition Showing the Transmission of X-rays Through the Patient by using a Detector Row (Wu et al., 2017). The acquired transmission profiles are used to reconstruct the CT image, composed of a matrix of picture elements (pixels) the values that are assigned to the pixels in a CT- image are associated with the attenuation of the corresponding tissue, or, more specifically, to their linear attenuation coefficient μ (m 1).The linear attenuation coefficient depends on the composition of the material, the density of the material and the photon energy, as seen in beer s law: I(x) I 0e x = μ (2.1) Where I(x) is the intensity of the attenuated X-ray beam, I0 the unattenuated X ray beam and x the thickness of the material. Note that beer s law only describes the attenuation of the primary beam and does not take into account the intensity of scattered radiation that is generated. For use in poly-energetic X-ray beams, beer s law should strictly be integrated over all photon energies in the X-ray spectrum. however, in the back projection methodologies developed for CT reconstruction algorithms, this is generally not implemented; instead, typically, a pragmatic solution is to assume where beer s law can be applied using one value representing the average photon energy of the X-ray spectrum. This assumption causes 26

inaccuracies in the reconstruction and leads to the beam hardening artifact. As an X-ray beam is transmitted through the patient, different tissues are encountered with different linear attenuation coefficients. If the pathway through the patient ranges from 0 to d, then the intensity of the attenuated X-ray beam, transmitted a distance d, can be expressed as: ( ) ( ) ( ) (2.2) Since a CT-image is composed of a matrix of pixels, the scanned patient can also be regarded as being made up of a matrix of different linear attenuation coefficient volume elements (voxel). Figure 2.6 shows a simplified 4 4 matrix representing the measurement of transmission along one line. Each element in the matrix can, in principle, have a different value of the associated linear attenuation coefficient.for such a discretization, the equation for the attenuation can be expressed as: ( ) ( ) (2.3) Figure 2.7: The Principle of Attenuation of an X-ray Beam in a Simplified 4 4 Matrix. (Wu et al., 2017). from the above, it can be seen that the basic data needed for CT-images are the intensities of the attenuated and un-attenuated X ray beams, respectively I(d) and I0, and that these can be measured. Image reconstruction techniques can then be applied to derive the matrix of linear attenuation coefficients, which is the basis of the CT-image. 27

2.1.3.3 CT Image Noise and Artifacts There are a number of image artifacts and imperfections that are associated with CT-images. Some are physics-related while others are caused by patient movements or organ and tissue dynamics, or systemrelated causes like detector insufficiencies. The images are affected by an unavoidable quantum noise from the statistical variations of the X-ray beam intensity. The noise levels are signal-dependent and can only be reduced by increasing the X-ray intensity or the acquisition time, and thereby increase the dose of radiation that the patient is submitted to. The radiologist may however use different kinds of smoothing filters to increase the low-level detail. Other physical dependent image artifacts are caused by X-ray scattering, beam hardening and partial volume effects. X- ray scattering is caused by the fact that some photons are reflected by the human body. Solutions to this problem might be to use detectors that can reduce the scatter effect or by using mathematical models that can correct for the effect. Beam hardening on the other hand is caused by the different attenuation of X-rays of different wavelength. Beam hardening is usually mathematically corrected for, but as the mathematical model assumes that the object in question only consists of one substance, usually water, the effect might occur for objects that differ largely from water attenuation (Li et al., 2015). 2.1.3.4 Contrast Media When using CT technology for liver tumor examinations the use of contrast media is important to get the best possible images. Medical contrast medium or contrast agent is used to increase the contrast between normal and abnormal tissues, get a better view of vascular anatomy (vascular = blood vessels) or improve the visualization in general for certain organs. The contrast agent causes an increase in X-ray absorption within the organ it passes through. Soft tissues such as organs and some diseased tissue, e.g. tumors, are often hard to distinguish from each other, but as different tissues absorb the contrast medium at different stages, the 28

tissue will be high-lighted in comparison to other organs with a smaller contrast uptake. 2.1.3.5 CT-Abdominal Image The advancements in the medical imaging technologies such as Computer Tomography (CT), Magnetic Resonance Imaging (MRI) and ultrasonography have dramatically increased the accuracy in diagnosing the abnormalities in human organs. An abdominal CT-scan helps the doctors see the organs, blood vessels and bones in the abdominal cavity. The multiple images provided give the doctors many different views of the body. The images acquired from such techniques provide vital information for radiologists to characterize the type and extent of abnormality. But most of the time, the radiographic view of the images does not completely provide diagnostic information for making adequate decision. This is due to the fact that the gray level differences in tissues may be small compared to the accuracy with which the measurements are carried out. Then it requires confirmation by biopsies. Though, biopsy is the most effective technique for characterizing the tissues as normal or abnormal, it is invasive. These limitations necessitate development of new analysis techniques that will improve diagnostic ability. 2.1.3.5.1 Small Lesion Detection Lesions in the liver may have very small difference in attenuation between the liver and the focal lesion. Contrast enhancement helps to increase this difference. With CT, reconstructions can be obtained without additional radiation. This can center a lesion and leads to increased lesion detection. 2.1.3.5.2 Hepatocellularcarcinoma (HCC) There are three CT patterns of hepatoma in the liver, a solitary mass, a dominant mass with smaller satellite lesions and diffuse involvement. Hepatomas are highly vascular lesions and may infrequently present with spontaneous hemorrhage. Most commonly lesions are well circumscribed with the appearance of a well-defined 'capsule'. Contrast enhancement is usually intense but transient, with internal areas of low attenuation and vascular channels. 29

2.1.3.5.3 Effect of Some Parameters on the Density in CT-images The radiographic density on films depends on both the thickness of a tissue and its atomic weight as shown in figure 2.7. Note that most of the soft tissues are clustered indistinguishably in the middle grays as shown in figure 2.7. Because of it mathematical accuracy and its digital underpinnings, computed tomography permits greater discrimination of individual soft tissues on an extended gray scale. 1. Tissue depth: X-ray absorption is proportional to the depth of the target tissue. 2. Atomic weight: The atomic weight of the tissue also plays a major role in determining image density. CT-abdominal image (film) contains imaging of gas pattern, calcification, soft tissue, bones and everything else. Usually calcification contains masses. Figure 2.7 : Effect of Tissues Types and Depth on the Density (Christ et al., 2016). When making an image, the body is scanned from a number of dots covering the entire part to be imaged as in figure 2.8. In this case we need to make an image of the liver and as we mentioned earlier it contains a number of sections appear in the CT image. 31

Figure 2.8: CT Scanning Process( Guite et al., 2013). After the completion of the scanning process, the doctor preview the image and search for any abnormal parts and based on the knowledge of anatomy of the normal parts and it locations the diseased part can be also located which usually appears different, for example in the case of tumor it's appears darker but in the case of liver tissue it appears lighter as shown in figure 2.9. Figure 2.9 : Liver Tumor with Dark Density and Liver Tissue with Light Density ( Guite et al., 2013). 31

2.1.4 Segmentation In image analysis and processing segmentation is an essential art. The ability to distinguish an object from the background is wanted in machine vision on robots, in industrial applications like quality control, in processing of seismic data and a number of other applications. Segmentation in a medical image which in many ways is different from other images is one of the most images analysis technologies (Christ et al., 2016). The segmentation process is a very essential step for the accurate detection of tumor cells. The liver segmentation process is to extract liver region from the abdominal CT-image which is a challenging task. 2.1.4.1 The Segmentation Challenge Image segmentation is the first step in image analysis. The main goal of image segmentation is to divide an image into several parts/segments having similar features or attributes. The main applications of image segmentation are: Medical imaging, Content-based image retrieval, and Automatic traffic control systems, Object detection and Recognition Tasks, etc. The image segmentation can be classified into two basic types: Local segmentation which concerned with specific part or region of image and Global segmentation which concerned with segmenting in whole image, consisting of large number of pixels (Saini and Singh, 2015). In order to assess the volume of a tumor, the tumor is to be segmented first. The segmentation of the tumor burden is a complex task for humans and thus also challenging for automatic segmentation algorithms. In addition to traditional problems of medical image segmentation, this task is challenging because of specificities in the appearance of liver tumors on the CT and other images. Low contrast, small size, non-uniformity of intensity, and various artifacts make this task difficult, sometimes even impossible with the state-of-the-art image processing techniques. First, non-uniformity of intensity inside a lesion makes inefficient most of the segmentation algorithms which are based on low variation of intensities within the segmented structure. Second, the wide variation of its locations 32

and mutual arrangements is a major source of difficulty for segmentation algorithms which involve global or local structure considerations (Christ et al.,2016). Successful medical image segmentation is difficult. This is so because of the complexity and diversity of anatomical structures on one hand, and particular properties, such as low contrast, noise, etc. on the other. No method has yet been proven to be fully effective for all types of images and anatomies. Yet, methods which a-priori ignores some of the aspects of the problem have even fewer chances to be successful. After the primary noise removal, the segmentation has to be carried out. The goal of the segmentation is to simplify and/or change the representation of an image into something that is more meaningful and easier to analyze. Image segmentation is typically used to locate objects and boundaries (lines, curves, etc.) in images. The segmentation of medical images of soft tissues into regions is a difficult problem because of the large variety of their characteristics. All connected components in the eroded image are labeled and number of connected components is computed (Vincey and Jeba,2013). 2.1.5 Thresholding Algorithm Digital Image processing is an emerging field in which doctors and surgeons are getting different easy pathways for the analysis of complex disease such as cancer, brain tumor, breast cancer, kidney stones, etc. The detection of liver disease is a very challenging task, in which special care is taken for image segmentation (Saini and Singh, 2015). Image segmentation is one of the most fundamental and difficult problems in image analysis. Image segmentation is an important part in image processing. In computer vision, image segmentation is the process of partitioning an image into meaningful regions or objects. Segmentation subdivides an image into its constituent region or object. Image segmentation methods are categorized on the basis of two properties discontinuity and similarity (Gonzalez and woods, 2007). Based on this property image segmentation is categorized as Edged based segmentation 33

and region based segmentation. The segmentation methods that are based on discontinuity property of pixels are considered as boundary or edges based techniques. Edge based segmentation method attempts to resolve image segmentation by detecting the edges or pixels between different regions that have rapid transition in intensity and are extracted and linked to form closed object boundaries. The result is a binary image. Based on theory there are two main edge based segmentation methods, gray histogram based and gradient based method (Kang et al., 2009). Region based segmentation partitions an image into regions that are similar according to a set of predefined criteria. The region based segmentation is partitioning of an image into similar areas of connected pixels. Each of the pixels in a region is similar with respect to some characteristic or computed property such as color, intensity and/or texture. There are different types of the Region based method like thresholding, region growing and region splitting and merging (Kang et al., 2009). Thresholding is an important technique in image segmentation applications. The basic idea of thresholding is to select an optimal graylevel threshold value for separating objects of interest in an image from the background based on their gray-level distribution. While humans can easily differentiable an object from complex background and image thresholding is a difficult task to separate them. The gray-level histogram of an image is usually considered as efficient tools for development of image thresholding algorithms. Thresholding creates binary images from grey-level ones by turning all pixels below some threshold to zero and all pixels about that threshold to one. If g(x, y) is a threshold version of f(x, y) at some global threshold T, it can be defined as : g(x, y) = 1 if f(x, y) T, Otherwise =0 (2.4) Thresholding operation is defined as: T = M [x, y, p(x, y), f (x, y)] (2.5) In this equation, T stands for the threshold; f (x, y) is the gray value of point (x, y) and p(x, y) denotes some local property of the point such as the average gray value of the neighborhood centered on point (x, y) Based 34

on this, there are two types of thresholding methods (Gonzalez and woods, 2007). 2.1.5.1 Global Thresholding When T depends only on f (x, y) (in other words, only on gray-level values) and the value of T solely relates to the character of pixels, this thresholding technique is called global thresholding. 2.1.5.2 Local Thresholding If threshold T depends on f (x, y) and p(x, y), this thresholding is called local thresholding. This method divides an original image into several sub regions, and chooses various thresholds T for each sub region reasonably (Kaur and Kaur, 2011). 2.1.5.1.1 Otsu s Thresholding method Otsu method is type of global thresholding in which it depend only gray value of the image. Otsu method was proposed by Scholar Otsu in 1979. Otsu method is global thresholding selection method, which is widely used because it is simple and effective (Qu and Hang, 2010). The Otsu method requires computing a gray level histogram before running. However, because of the one-dimensional which only consider the graylevel information, it does not give better segmentation result. So, for that two dimensional Otsu algorithms was proposed which works on both gray-level threshold of each pixel as well as its Spatial correlation information within the neighborhood. So Otsu algorithm can obtain satisfactory segmentation results when it is applied to the noisy images (Zhuang and Wen, 1993). Many techniques thus were proposed to reduce time spent on computation and still maintain reasonable thresholding results. Fast recursive technique that can efficiently reduce computational time proposed by (Gong et al.,1998). Otsu s method was one of the better threshold selection methods for general real world images with regard to uniformity and shape measures. However, Otsu s method uses an exhaustive search to evaluate the criterion for maximizing the betweenclass variance. As the number in classes of an image increases, Otsu s 35

method takes too much time to be practical for multilevel threshold selection (Sahoo et al., 1988). 2.1.6 Morphological Operation Morphology is a bread set of image processing operations that process images based on shapes. In morphology operation each pixel in the image is adjusted based on the value of other pixels in neighborhood. By choosing the size and shape of the neighborhood you can construct a morphology operation that is sensitive to specific shapes in the input image. The most basic morphological operations are dilation and erosion. Dilation adds pixels to the boundaries of objects in an image, while erosion removes pixels on objects boundaries. Dilation and erosion are often used in combination for specific image preprocessing applications such as filling holes or removing small objects. 2.1.7 Graphic User Interface 2.1.7.1 Background The graphical user interface (GUI), is a type of user interface that allows users to interact with electronic devices through graphical icons and visual indicators such as secondary notation, instead of text-based user interfaces, typed command labels or text navigation. GUIs were introduced in reaction to the perceived steep learning curve of commandline interfaces (CLIs), which require commands to be typed on a computer keyboard (Dayton and Tom, 2014). The actions in a GUI are usually performed through direct manipulation of the graphical elements. Beyond computers, GUIs are used in many handheld mobile devices such as MP3 players, portable media players, gaming devices, smart-phones and smaller household, office and industrial controls. The term GUI tends not to be applied to other lowerdisplay resolution types of interfaces, such as video games (where headup display (HUD) is preferred), or not including flat screens, like volumetric displays because the term is restricted to the scope of twodimensional display screens able to describe generic information, in the 36

tradition of the computer science study at the Xerox Palo Alto Research Center (Dayton and Tom, 2014). 2.1.7.2 User interface and interaction design The graphical user interface is presented (displayed) on the computer screen. It is the result of processed user input and usually the main interface for human-machine interaction. The touch user interfaces popular on small mobile devices are an overlay of the visual output to the visual input( Labarta et al., 2015). Designing the visual composition and temporal behavior of a GUI is an important part of software application programming in the area of human computer interaction. Its goal is to enhance the efficiency and ease of use for the underlying logical design of a stored program, a design discipline named usability. Methods of user-centered design are used to ensure that the visual language introduced in the design is well-tailored to the tasks ( Labarta et al., 2015). The visible graphical interface features of an application are sometimes referred to as chrome or GUI (pronounced gooey).typically, users interact with information by manipulating visual widgets that allow for interactions appropriate to the kind of data they hold. The widgets of a well-designed interface are selected to support the actions necessary to achieve the goals of users. A model view controller allows a flexible structure in which the interface is independent from and indirectly linked to application functions, so the GUI can be customized easily. This allows users to select or design a different skin at will, and eases the designer's work to change the interface as user needs evolve. Good user interface design relates to users more and to system architecture less. Large widgets, such as windows, usually provide a frame or container for the main presentation content such as a web page, email message or drawing. Smaller ones usually act as a user-input tool (Labarta et al.,2015). A GUI may be designed for the requirements of a vertical market as application-specific graphical user interfaces. Examples include automated teller machines (ATM), point of sale (POS) touch-screens at 37

restaurants, self-service checkouts used in a retail store, airline selfticketing and check-in, information kiosks in a public space, like a train station or a museum, and monitors or control screens in an embedded industrial application which employ a real-time operating system (RTOS). By the 1980s, cell phones and handheld game systems also employed application specific touch-screen GUIs. Newer automobiles use GUIs in their navigation systems and multimedia centers, or navigation multimedia center combinations (Dayton and Tom, 2014). 2.1.7.3 GUI in Mat-lab Program It is a graphical display in one or more windows containing controls, called components, which enable a user to perform interactive tasks. The user does not have to create a script or type commands at the command line to accomplish the tasks. Unlike coding programs to accomplish tasks, the user does not need to understand the details of how the tasks are performed. GUI components can include menus, toolbars, push buttons, radio buttons, list boxes, and sliders just to name a few (Dayton and Tom, 2014). 2.1.7.3.1 Why use a GUI in MATLAB? The main reason GUIs are used is because it makes things simple for the end-users of the program. If GUIs were not used, people would have to work from the command line interface, which can be extremely difficult and frustrating. 2.1.7.3.2 Main Steps of Designing a Simple GUI 1. Open the Mat-lab Program and wait for it to finish loading. 2. Click on "MATLAB" in the launch pad to expand the list and then double click on "GUIDE (GUI Builder)". If you cannot see the launch pad, click on view followed by launch pad. The GUI builder will appear. 3. Click on the "ok" button in the left hand side of the window. This will allow you to drag and drop a pushbutton. 4. Move the mouse to somewhere on the grey area in the center of the window. 38

5. Click once and hold down the button and drag your mouse until the square that this form is of the size you'd like. 6. Release the mouse button and you will see your push button appear. 7. Double click on the pushbutton you just created. A property manager will pop up. 8. Locate the "string field" and click the area to the right of it and type "Say Hello". Also Change the Tag to "button". 9. Find the button on the left labeled "txt" and follow the same step 8 again. 10. Now click on file then save to save your work. This will also pop up the code for your program. 11. Locate the line of code in the code editor that says function varargout = pushbutton1_callback(h, eventdata, handles, varargin). This is the callback function. Any code below this will be executed whenever the user pushes the button. Here we will make this change the text in the text box when the user clicks on the button. 12. Write set. 39

2.2 Literature Review 2.2.1 Previous Studies In (2018) Ben-Cohen, et al. in their paper ANATOMICAL DATA AUGMENTATION FOR CNN BASED PIXEL-WISE CLASSIFICATION proposed a method for anatomical data augmentation that was based on using slices of computed tomography (CT) examinations that were adjacent to labeled slices as another resource of labeled data for training the network. The extended labeled data was used to train a U-net network for a pixel-wise classification into different hepatic lesions and normal liver tissues. Their dataset contains CT examinations from 140 patients with 333 CT images annotated by an expert radiologist. They tested their approach and compared it to the conventional training process. Results indicated superiority of their method. Using the anatomical data augmentation they achieved an improvement of 3% in the success rate, 5% in the classification accuracy, and 4% in Dice (Ben-Cohen et al., 2018). In (2018) Rafiei.S et al in their paper Liver segmentation in CT images using three dimensional to two dimensional fully convolution network they proposed an efficient liver segmentation with their 3D to 2D fully connected network (3D-2D-FCN). The segmented mask was enhanced by means of conditional random field on the organ's border. Consequently, they segmented a target liver in less than a minute with Dice score of 93.52 (Rafiei. et al, 2018). In (2018) Huang et al in their paper automatic liver segmentation in CT images were using modified graph cuts and feature detection proposed a fully automatic procedure using modified graph cuts and feature detection for accurate and fast liver segmentation. The initial slice and seeds of graph cuts were automatically determined using an intensity-based method with prior position information. A contrast term based on the similarities and differences of local organs across multi-slices was proposed to enhance the weak boundaries of soft organs and to prevent over-segmentation. The term was then integrated into the graph cuts for automatic slice segmentation. Patient-specific intensity and shape 41

constraints of neighboring slices were also used to prevent leakage. Finally, a feature detection method based on vessel anatomical information was proposed to eliminate the adjacent inferior vena cava with similar intensities. Experiments were performed on 20 Sliver07, 20 3Dircadb datasets and local clinical datasets. The average volumetric overlap error, volume difference, symmetric surface distance and volume processing time were 5.3%, -0.6%, 1.0 mm, 17.8 s for the Sliver07 dataset and 8.6%, 0.7%, 1.6 mm, 12.7 s for the 3Dircadb dataset, respectively. The proposed method could effectively extract the liver from low contrast and complex backgrounds without training samples. Fully automatic, accurate and fast for liver segmentation in clinical settings was introduced (Huang et al., 2018). In (2018) Zhou.Z et al, in their paper Semi-automatic Liver Segmentation in CT Images through Intensity Separation and Region Growing proposed a novel liver segmentation method including intensity separation; region growing and morphological hole-filling was presented. Firstly, intensity separation was employed to increase the difference between the intensities of liver and its adjacent tissues. Then the following region growing algorithm was applied to segment the liver. And the morphological hole-filling was used at last to refine the segmentation results. The proposed method was evaluated with a patient dataset coming from Ningbo Li Hui-li hospital. The validation results and surface rendering showed that the method provided a reliable and robust way for liver segmentation. This method could provide a reference for clinical practice (Zhou et al., 2018). In (2017) Goel, et al. in their paper A Review of Feature Extraction Techniques for Image Analysis reviewed various feature extraction techniques. These methods were classified as low-level feature extraction and High-level feature extraction. Low-level feature extractions were based on finding the points, lines, edge, etc while high level feature extraction methods had used the low level feature to provided more significant information for further processing of Image analysis. Mostly high-level feature extraction method used the Artificial Neural Network (ANN) to extract the feature in multiple layers (Goel et al., 2017). 41

In (2017) Alakwaa,et al in their paper Lung Cancer Detection and Classification with 3D Convolution Neural Network (3D-CNN) proposed that demonstrates a computer-aided diagnosis (CAD) system for lung cancer classification of CT-scans with unmarked nodules, a dataset from the Kaggle Data Science Bowl. Thresholding was used as an initial segmentation approach to segment out lung tissue from the rest of the CT scan. Thresholding produced the next best lung segmentation. Neural Networks (CNNs) to ultimately classify the CT-scan as positive or negative for lung cancer. The 3D CNNs produced a test set Accuracy of 86.6% (Alakwaa et al., 2017). In (2017) Wu et al in their paper 3D Liver Tumor Segmentation in CT- Images Using Improved Fuzzy C-Means and Graph Cuts 2017, proposed an efficient semiautomatic method for liver tumor segmentation in CT volumes based on improved fuzzy C-means (FCM) and graph cuts. With a single seed point, the tumor volume of interest (VOI) was extracted using confidence connected region growing algorithm to reduce computational cost. Then, initial foreground/background regions were labeled automatically, and a kernelized FCM with spatial information was incorporated in graph cuts segmentation to increase segmentation accuracy. The proposed method was evaluated on the public clinical dataset (3Dircadb), which included 15 CT volumes consisting of various sizes of liver tumors. They achieved an average volumetric overlap error (VOE) of 29.04% and Dice similarity coefficient (DICE) of 0.83, with an average processing time of 45 s per tumor (Wu et al., 2017). In (2016) Sreeraj and Raju in their paper Automatic Detection of Liver Tumor in CT Image Using Region Growing and SVM Classifier proposed an approach to automatic detection of liver tumor in CT images by using region-growing and Support Vector Machine (SVM) which was successfully classifies the liver cancer types such as hepatoma, hemangioma and carcinoma. The method rectified the problem of manual segmentation and classification which was time consuming due to the variance in the characteristics of CT-images. Their proposed method has been tested on a group of CT images obtained from hospitals in Kerala with promising results both in liver and tumor segmentation. The average 42

error rate and accuracy rate obtained from their proposed method is 0.02 and 0.9 (Sreeraj and Raju, 2016). In (2016) Moghbel1 et al. in their paper AUTOMATIC LIVER TUMOR SEGMENTATION ON COMPUTED TOMOGRAPHY FOR PATIENT TREATMENT PLANNING AND MONITORING proposed a new segmentation method for liver tumors from contrast-enhanced CTimaging is proposed. As manual segmentation of tumors for liver treatment planning was both labors intensive and time-consuming, highly accurate automatic tumor segmentation was desired. The proposed framework is fully automatic requiring no user interaction. The proposed segmentation evaluated on real-world clinical data from patients was based on a hybrid method integrating cuckoo optimization and fuzzy c- means algorithm with random walker s algorithm. The proposed method was able to outperform most other tumor segmentation methods reported in the literature while representing an overlap error improvement of 6 % compared to one of the best performing automatic methods in the literature. The proposed framework was able to provide consistently accurate results considering the number of tumors and the variations in tumor contrast enhancements and tumor appearances while the tumor burden was estimated with a mean error of 0.84 % in 3Dircadb dataset (Moghbel1 et al., 2016). In (2016) Raj and Jayasree in their paper Automated Liver Tumor Detection Using Markov Random Field Segmentation proposed an automated computer aided diagnosis of liver tumors from CT images. Initially liver was segmented using Markov Random Field Segmentation (MRF) embedded level set method. It provided robustness to noise and fast segmentation. The shape ambiguities of the segmented liver were found out by shape an analysis method which uses training set for correction. From the corrected liver segmentation, hepatic tumors are detected by graph cut method and feature extraction was done to classify them using support vector machine (SVM) classifier (Amitha and Jayasree, 2016). In (2015) Ali and Hadi in their paper Diagnosis of Liver Tumor from CT- Images using Digital Image Processing proposed that the detection and 43

diagnose of liver tumors from CT-images by using digital image processing, was a modern technique depends on using computer in addition to textural analysis to obtain an accurate liver diagnosis, despite the method's difficulty that came from liver's position in the abdomen among the other organs. This method made the surgeon able to detect the tumor and then easing treatment also it helped physicians and radiologists to identify the affected parts of the liver in order to protect the normal parts as much as possible from exposure to radiation. The study described a new 2D liver segmentation method for purpose of transplantation surgery as a treatment for liver tumors. Liver segmentation is not only the key process for volume computation but also fundamental for further processing to get more anatomy information for individual patient. Due to the low contrast, blurred edges, large variability in shape and complex context with clutter features surrounding the liver that characterize the CT-liver images. In this paper, the CT-images were taken, and then the segmentation processes were applied to the liver image which found, extracted the CT liver boundary and further classify liver diseases (Ali and Hadi, 2015). In (2015) Wen Li et al. in their paper Automatic Segmentation of Liver Tumor in CT Images with Deep Convolution Neural Networks proposed that an automatic method based on convolution neural networks (CNNs) was presented to segment lesions from CT images. The CNNs was one of deep learning models with some convolution filters which can learn hierarchical features from data. CNNs model to popular machine learning algorithms: AdaBoost, Random Forests (RF), and support vector machine (SVM) had been compared. These classifiers were trained by handcrafted features containing mean, variance, and contextual features. Experimental evaluation was performed on 30 portal phase enhanced CT images using leave-one-out cross validation. The average Dice Similarity Coefficient (DSC), precision, and recall achieved of 80.06% ± 1.63%, 82.67% ± 1.43% and 84.34% ± 1.61%, respectively. The results showed that the CNNs method has better performance than other methods and were promising in liver tumor segmentation (Wen et al., 2015). 44

In (2015) Krishan and Mittal in their paper Detection and Classification of Liver Cancer using CT Images, proposed that the enhancement of Computed tomography (CT) images using two different algorithms: Contrast Limited Adaptive Histogram Equalization (CLAHE) and Constrained Variable Histogram Equalization (CVHE). CLAHE enhanced the tumor region in a new look. CVHE enhanced with preserving the globalization of an image. The normal liver detection was done by the ox plot comparison in CLAHE. Primary liver cancer detection was done by CVHE. State vector machines (SVM) classifier works for the classification of CT-liver images (Krishan and Mittal, 2015). 2.2.2 Summary of Literature Review All studies related to segmentation of tumors or lesion contain main steps included image pre-processing which done after acquiesced the images.also it need to enhance the weak boundaries of soft organs and to prevent over-segmentation enhancement of Computed tomography (CT) images generally done using different algorithms for example: Adaptive Histogram Equalization it use to computes several histograms each corresponding to a distinct section of the image and uses them to redistribute the lightness values of the image but also it has disadvantage it over-amplify the noises so many sub-method found to solve this problem like Contrast Limited Adaptive Histogram Equalization (CLAHE) and Constrained Variable Histogram Equalization (CVHE). CLAHE limited the contrast amplification so it reduce the problem of noise amplification also it enhanced the tumor region in a new look. CVHE enhanced with preserving the globalization of an image. Effectively extract the liver from low contrast and complex backgrounds without training samples, extraction methods include low-level feature extraction and High-level feature extraction. Low-level feature extractions were based on finding the points, lines, edge, etc while high level feature extraction methods had used the low level feature to provided more significant information for further processing of Image analysis. Finally lesion segmentation done using many methods likes graphic cuts, intensity separation and region growing every method had are advantages. 45

The method rectified the problem of manual segmentation which was time consuming due to the variance in the characteristics of CT-images. The segmented images used in the classification process which had been done using different types of algorithms like State vector machines (SVM) classifier which has many advantages like less cost and better performance but it is kind of complex compare to other methods,artificial neural network (ANN) which is non-linear model that is easy to use and understand also it treat complicated problems in which too many variables to be simplified in a model, so it train a large amount of datasets in simple way but as other methods it has it disadvantages for example it doesn t have explanatory power i.e it train and extract the best signals to accurately classify and cluster data but it will not tell you why it reached a certain conclusion also it need a lot of distributed run-time to train on very large datasets and 3D Convolution Neural Network (3D-CNN) it is suitable for classification process because it needs little preprocessing but it primary downside is the increasing computational cost, This becomes especially challenging for 3D convolution where handling even the smallest instances require substantial resources. Several methods are available for segmentation, but this technique s doesn t give any accurate results. The above used methods are very complex, this shows the need of a better technique for liver lesion segmentation. In this study Otsu s thresholding is used to segment the liver and its lesions. 46

CHAPTER THREE RESEARCH METHODOLOGY 3.1 Introduction The implementation of the study objects accomplished using specific steps as shown in the following block diagram figure 3.1. Figure 3.1: Study Flow Diagram. 47

3.2 Study Stages 3.2.1 CT Image Acquisition Axial CT-scan had been used to develop the proposed algorithm. An axial CT scan involves an acquisition of transmission profiles with a rotating X ray tube and a static table. An axial acquisition is generally performed with one full 360º rotation of the X ray tube, but to enhance temporal resolution this may be reduced to a shorter 180º + fan angle acquisition. The rotation angle can be extended to, for example, a 720º acquisition to enhance low contrast resolution by allowing a higher tube charges (mas). A complete CT scan generally involves subsequent axial acquisitions in order to cover a clinically relevant volume. This is achieved by translation of the table ( step ) after each axial acquisition ( shoot ). This is referred to as a step and shoots acquisition. Usually, the table translation is equal to the slice thickness, so that subsequent axial acquisitions can be reconstructed as contiguous axial images. In order to reconstruct a CTimage, numerous measurements of the transmission of X-rays through the patient are acquired. This information is the basis for reconstruction of the ct image. This study contains data included 12 images. Ten of the CT Abdominal images which had been used in this study had been collected from wadmadani diagnostic center and two had been collected from the free online database from atlas liver database which is a free, web-based reference of liver CT, MR and ultrasound imaging for medical professionals that contains a comprehensive spectrum of liver diseases and their imaging features. Common and uncommon liver imaging cases can be found by the imaging feature, disease categories, diagnosis or key words. The atlas is designed to be used as a teaching repository by radiologist in training and their peers and as a tool for construction of image based differential diagnoses by radiologists in practice. 48

3.2.2 Image Pre-processing An average filter had been utilized for smoothing the images with mask [5x5] which had given better results than [7x7] and [9x9]. The main reason of using the Average filtering for the preprocessing step of this algorithm is because average filtering retains the edge information within the image where Mean filters and Gaussian filters tend to blur the edges in the image. This is because the average filter does not create new unrealistic pixel values in the case of the filtering window laying over an edge. The variations in the tissue which looks like random noise is mostly caused by natural variations in the tissue. Commonly reducing the noise can be achieved by smoothing the image with a mean filter. The drawback with filtering is that essential image details will disappear in the smoothing process. The variations in the liver are natural variations in the tissue. Smoothing out tissue variations would remove all image details that are needed for proper segmentation of the lesions. The segmentation method detects edges in the image. Edges vary between strong and weak depending on how abrupt the change in intensity is. Without smoothing most methods will find edges everywhere. This supports the concept of always using some smoothing of the image. To find the right balance between smoothing and preservation of details is essential to achieve valid results in medical image processing. Multiple segmentation methods worked with high degree of smoothing but details were lost. Another problem is broken edges when the image is smoothed too much, so another filter had been used it was sharp filter using a mask [9x9]. Sharpening filter used to highlight fine details in the image or in other word enhance detail that has been blurred. The efficiency of each filter based on the size of filter and the noise type. Many sizes had been tried but the most efficient size was [9X9] with standard deviation 11.In order to saw the efficiency of each filter compare to the another the intensities distributions of slide of the image had been calculated. Some statically properties (parameters) had been calculate for both of the filter like entropy (disorder) which is average number of bits per symbol required to encode the source information, mean which is pixel intensity for the 49

entire image, standard deviation which calculate how each pixel varies from the neighboring pixels and variance. After the results had been compared sharpen filter had gave better results than smooth filter. 3.2.3 Liver Segmentation The main objective of segmentation process is to transform an image into something that is more meaningful and softer to examine by extracting important features from the original image information. Therefore, this subdivision will depend on the problem being solved. As we know CTimages scan the entire abdominal region so liver region had to be segmented, manually segmentation using region of interest technique which also called mask creation had been used to extract the liver region. Extraction of the required region of interest (ROI) from within an image is very important. The effective solution is to use a combination of different image processing techniques to identify the region of interest. This process applies the ROI mask to the original image by comparison of the two images. The parts of the original image, which are identified by the ROI mask, will be transferred to the final image. On the other hand, the part of original image which is marked by the non-roi mask will be replaced by 0 in the final image (i.e. those areas not of interest). First input image had been resize as [1024x1024] then seed points had been selected which was 10 seeds use to determine the liver boundaries. After that both regions had been shown separately by labeling liver region with binary mask equal one (appear white) and rest of the abdominal part with pixels zeros (black), Finally liver after manual segmentation had been displayed also rest of the abdominal had been displayed in which liver region had been appear as black region. 3.2.4 Thresholding Algorithm In computer vision and image processing, Otsu's method, named after Nobuyuki Otsu, is used to automatically perform clustering-based image thresholding. The algorithm assumes that the image contains two classes of pixels following bi-modal histogram (foreground pixels and 51

background pixels), it then calculates the optimum threshold separating the two classes. In this study the algorithm had been generally performed in many steps first image had been formatted to uint8, then the histogram of the image had been done, after that the histogram had been normalize, finally initial and terminal values had been set up which contained the desired threshold. From the experimental results, the performance of global thresholding techniques including Otsu s method is shown to be limited by the small object size, the small mean difference, the large variances of the object and the background intensities, the large amount of noise added, and so on. A simple, effective good results method had been used depend on the concept of the Otsu thresholding but in other way avoid the limitation of using it. 3.2.5 Lesions Segmentation The main objective of segmentation process is to transform an image into something that is more meaningful and softer to examine by extracting important features from the original image information. The output images from liver segmentation stage which had been applied to the thresholding algorithm had been used to extract the lesions in the liver. As had been said before image segmentation plays an important role in image analysis and computer vision systems. Among all segmentation techniques the automatic thresholding methods are widely used because of their advantages of simple implement and time saving. Otsu s method is one of thresholding methods and frequently used in various fields. It is based on criteria that minimizes the within class variance. However, Otsu s method is an exhaustive algorithm of searching the global optimal threshold, also is needs to compute a gray level histogram firstly. The general syntax for the Otsu method is: T=otsuthresh(counts) T is the global thresholding which is normalized intensity value that lies in the rang [0, 1] that can be used with imbinarize to convert an intensity image to binary image. Otsu method chooses a threshold that minimizes the interclass variance of the threshold black and white pixels. 51

The algorithm assumes that the image contains two classes of pixels following bi-modal histogram (foreground pixels and background pixels), it then calculates the optimum threshold separating the two classes. In Otsu's method we exhaustively search for the threshold that minimizes the intra-class variance (the variance within the class), defined as a weighted sum of variances of the two classes: σ 2 w(t) = w 0 (t) σ 2 0 (t) + w 1 (t) σ 2 1 (t) (3.1) Weights w 0 and w 1 are the probabilities of the two classes separated by a threshold t,and σ 2 0 and σ 2 1 are variances of these two classes. The class probability w 0,1 (t) is computed from the L bins of the histogram: 0 (t) = ( ) (3.2) 1 (t) = ( ) (3.3) The weighted within-class variance is: σ 2 w (t ) = q 1 (t ) σ 2 1(t ) + q 2 (t ) σ 2 2 (t ) (3.4) Where the class probabilities are estimated as: q 1 (t) =Σ t i=1 P(i) (3.5) q 2 (t) =Σ I i=t+1 P(i) (3.6) And the class means are given by: μ 1 (t) = Σ t i=1 [ip(i)] /[q 1 (t)] (3.7) μ 2 (t) = Σ I i=t+1 [ip(i)] /[q 2 (t)] (3.8) Finally, the individual class variances are: σ 2 1(t) = Σ t i=1 [i μ 1 (t)] 2 [P(i)/ q 1 (t)] (3.9) Algorithm : σ 2 2(t) = Σ I i=t+1 [i μ 2 (t)] 2 [P(i)/ q 2 (t)] (3.10) 1. Compute histogram and probabilities of each intensity level. 52

2. Set up initial i (o) and µ i (0). 3. Step through all possible thresholds t= 1, maximum intensity Update i and µ i. Compute σ 2 b (t). 4. Desired threshold corresponds to the maximum σ 2 b (t). 3.2.5.1 Morphological Operations As mention before the main morphological operations are dilation and erosion there are also some operation called opening and closing are used. Before used any of the above ioeration we have to calculate what called structure element usually structure element is sized [3X3] and has it's origin at the center pixel. It shifted over the image and at each pixel of the image its elements are compared with the set of the underlyimg pixels. If the two sets of elememts match the condition defined by the set operator, the pixel underneath the origin of the structuring element is set to predefined value (0 or 1). A morphology operator is therfore defined by its structuring element and the applied set operator. 3.2.6 Graphic User Interface (GUI) GUI had been used because it makes things simple for the end-users of the program. If GUIs were not used, people would have to work from the command line interface, which can be extremely difficult, So GUI had been designed which were graphical display in one or more windows containing controls, called components, which enable a user to perform interactive tasks as shown in figure 3.2. The user does not have to create a script or type commands at the command line to accomplish the tasks. Unlike coding programs to accomplish tasks, the user does not need to understand the details of how the tasks are performed. User interface components can include menus, toolbars, push buttons, radio buttons, list boxes, and sliders just to name a few. Few steps had been used to design it which had included, first open the GUI main screen by writing the command guide in the command window, then the next figure had been appears 53

Figure 3.2 GUI Start Screen. Then needed tools had been selected from the next screen figure 3.3 Figure 3.3: Tools Selection Screen. Next the panel space where the tools of the program had been distributed were plan and the size had been determined as shown in figure 3.4. Figure 3.4 : Process of Select and Resize the Panel Space. 54

After that the tools had been selected and named as shown in the followed image Figure 3.5: The Final View of the Study Screen. 55

CHAPTER FOUR Results and Discussions 4.1 Abdominal CT image Abdominal CT image consist of liver and other organs. Segmentation of liver from abdominal region is based on their intensity value, so we have to separate liver from other organs. 4.2 Image Pre-processing Figure 4.1: Original CT-abdomen Image. The main function of pre-processing is to suppress the noise present in the image and to enhance quality of the image. In biomedical image processing the main difficulties are the structure is complex, unknown boundary between the organs, shape, structure are vary from patient to patient even if same imaging modalities are used. Two filter had been used smoothing and sharpening filter to evaluate each one of them two parameters had been used signal to noise ratio (SNR) and Mean Square Error (MSE) after filters had been implemented the smoothing filter show better results as shown in Table 4.1. Figure 4.2: CT-abdominal Image after using the Smoothing Filter. 56

Figure 4.3: Slide Crop Process. Figure 4.4: Intensities Distributions of The Slide Cropped Smoothed Image. Figure 4.5: Image after using the Sharpen Filter. Figure 4.6: Intensities Distributions of the Slide Cropped Sharping Image. Table 4.1: Performance Analysis of Smooth and Sharp Filters No. SNR(smooth) SNR(sharp) MSE(smooth) MSE(Sharp) 1 10.6940 14.0651 1.0714e-04 5.321e-04 2 11.3586 15.0931 6.5166e-05 1.4805e-04 3 10.7706 14.7214 1.8917e-04 3.7769e-04 57

4 11.5624 15.3975 3.6385e-04 0.0018 5 11.0864 15.3153 2.1081e-04 5.1548e-04 6 13.3079 15.1883 3.0353e-04 0.0014 7 13.3937 14.6785 5.5689e-05 2.8796e-04 4.3. Liver Segmentation Mask Creation is used to create the binary mask which has the value of 0 or 1 s. This operation is performed by Specify polygonal region of interest (ROI) in MATLAB software. This is used to draw around the edge of the liver manually, making it possible to discard all irrelevant information. Here we can see the result of our own encirclement of the liver. A binary image used as a mask. For masked filtering where the liver is giving the value 1 and the background is given the value 0, the final step involved applying the segmented mask to the original image, multiplying the final eroded image with the original image to give the masked image. Figure 4.7: Image Used to Determine the Region of Interest area (Liver Organ). Figure 4.8: Mask Creation Process. 58

Figure 4.9: Region Which Determine by the Mask in the Original Image. Figure 4.10: Rest of the Abdominal After Liver Region Segmentation. 4.4 Lesions Segmentation After segment the liver region from the abdominal image, remove the noise, applied thresholding algorithm lesion had been extract as shown in figure 4.11. Figure 4.11: Segmented Liver Lesions. 4.5 Graphic User Interface (GUI) All that can be simpler to users by using the graphic user interface, all what is need is open the GUI file then the follow screen appears 59

4.6 Performance Evaluation Figure 4.12: The Main GUI Screen. The image assesment plays vital role in validating the performance of segmentation which is important in image analysisi there are two kind of performance measurement subjective (human involvement) and objective (compare the method with golden standard based in image properties), the objective (empirical) also divide into two discrepancy (depend on ground truth) and goodness (don't need ground truth) which we use it here one of the goodness parameter is entropy which is evaluation measure used to compute the randomness or information contains in an image our method gave segmentation with entropy factor less than one as shown in Table 4.2. Table 4.2 : Performance Evaluation using Entropy Coefficient Image Entropy Coefficient 1 0.9706 2 0.9181 3 0.9474 4 0.9618 61