CHAPTER 1 INTRODUCTION

Transcription

1 CHAPTER 1 INTRODUCTION 1.1 BACKGROUND STUDY Cancer is the foremost reason of death in economically developed countries and the second most important reason of death in developing countries. Cancer recognized medically as a malignant neoplasm, is an extensive group of a mixture of diseases, all concerning unregulated cell growth. In cancer, cells break up and breed widely, forming malignant tumors, and raid nearby parts of the body. The cancer may also expand to more distant parts of the body through the lymphatic system or bloodstream. Not all tumors are cancerous. Benign tumors do not produce uncontrollably, do not attack neighboring tissues, and do not spread throughout the body. Determining what causes cancer is complex. Many things are known to increase the risk of cancer, including tobacco use, certain infections, radiation, lack of physical activity, poor diet and obesity, and environmental pollutants. These can directly damage genes or combine with existing genetic faults within cells to cause the disease. Approximately five to ten percent of cancers are entirely hereditary. Cancer can be detected in a number of ways, including the presence of certain signs and symptoms, screening tests, or medical imaging. Once a possible cancer is detected, then it is diagnosed by microscopic examination of a tissue sample. Cancer is usually treated with chemotherapy, radiation therapy and surgery. The chances of surviving the disease vary greatly by the type and location of the cancer and the extent of disease at the start of treatment. 1

2 While cancer can affect people of all ages, and a few types of cancer are more common in children, the risk of developing cancer generally increases with age. In 2008, cancer caused about 13% of all human deaths worldwide (7.9 million). Rates are rising as more people live to an old age and as mass lifestyle changes occur in the developing world. More than 70% of all cancer deaths occurred in low- and middle-income countries. Deaths from cancer worldwide are projected to continue rising, with an estimated 11.5 million deaths in Breast cancer is a very common disease. It is the most common cancer in females and the second most common cancer for males and females combined. The breasts are made up of fat, supportive (connective) tissue and glandular tissue that contains lobes. The lobes (milk glands) are where breast milk is produced. These are connected to the nipple by a network of milk ducts.the organs and tissues of the body are made up of tiny building blocks called cells. Cancer is a disease of these cells. Breast cancer is a cancer originating from breast tissue, most commonly from the inner lining of the milk ducts or the lobules that supply the ducts with milk. Cancers originating from ducts are known as ductal carcinomas; those originating from lobules are known as lobular carcinomas. Breast cancer is a disease of humans and other mammals; while the overwhelming majority of cases in humans are women, men can also develop breast cancer[1] Breast cancer lifetime Risk The most common breast cancer statistic is that 1 in 8 women will develop breast cancer in their lifetime. If everyone lived beyond the age of 70, 1 in 8 of those women would get or 2

3 have had breast cancer. This statistic is based on everyone in the population living beyond the age of 70. Since the breast cancer risk increases as per age as follows: Age 20-29: 1 in 2,000 Age 30-39: 1 in 229 Age 40-49: 1 in 68 Age 50-59: 1 in 37 Age 60-69: 1 in 26 Ever: 1 in 8 More than 1 million cases of breast cancer are now diagnosed worldwide each year. The lifetime risk is now 1 in 8 women which will rise to 1 in 7 women by the year This represents 10% of all new cancers and 23% of female cancer cases. Lowest rates are in rural Africa and Asia where women still have many children, early first pregnancies and breastfeed for a long period of time (all three factors lower breast cancer risk) [2]. Higher rates of breast cancer are in the USA. Breast cancer rates are increasing year after year in Western populations. Breast cancer rates are higher in more affluent (richer) social classes. Breast cancer rates increase when breast screening programmes are introduced. Traditionally lower breast cancer rates in Central/Eastern Europe and the Far East are rising rapidly. In China the increase has been caused by a drop in the birth rate and exposure of Chinese women to lifestyle risk factors including weight gain, alcohol and hormone replacement therapy [3-5]. 3

4 1.1.2 Rising incidence of breast cancer in India Breast cancer is now the most common cancer in most cities in India, and 2nd most common in the rural areas. The average age of developing a breast cancer has undergone a significant shift over last few decades. The figure 1.1 shows the incidence of breast cancer in India. Figure 1.1: The incidence of breast cancer in India The horizontal line lower down represents the age groups: 20 to 30 years, 30 to 40 yrs and so on. And the vertical line represents the percentage of cases. The green colour represents the incidence 25 years back, and brown colour represents the situation today. 25 years back, out of every 100 breast cancer patients, 2% were in 20 to 30 years age group, 7% were in 30 to 40 and so on. 69% of the patients were above 50 years of age. Presently, 4% are between 20 to 30 4

5 yrs age group, 16% are between 30 to 40, 28% are between 40 to 50 age group. So, almost 48% patients are below 50. An increasing number of patients are in the 25 to 40 years of age, and this definitely is a very disturbing trend. The figure 1.2 shows the percentage distribution of top ten cancers in females in Mumbai. Figure 1.2: Top ten cancers in females in Mumbai The breast cancer accounts for 23% of all the female cancers followed by cervical cancer (17.5%) in major cities such as Mumbai, Calcutta and Bangalore. Although the incidence is lower in India than in the developed countries, the burden of diagnosing and treating of breast cancer in India is alarming. According to the International Agency for Research on Cancer, which is part of the World Health Organization (WHO), there were approximately 89,000 women per year affected by breast cancer in India in the year 2004 and 92,000 women in

6 A radiologist looks for certain signs and characteristics indicative of cancer when evaluating a mammogram. A microcalcification is a tiny calcium deposit accumulated in the tissue in the breast and it appears as a small bright spot on the mammogram[6] Breast Cancer Breast cancer cluster is typically defined to be at least 3 to 5 microcalcifications within a square centimeter region. Up to 50 percent of malignant masses demonstrate clustered microcalcifications and in a number of cases, the clusters are the only sign of malignancy. The calcifications vary in size from smaller than 0.1 millimeters to 5 millimeters in diameter and a radiologist must carefully examine the mammogram with a magnifier to locate calcifications, which may be embedded in dense parenchymal (connective) tissue. Size, shape and radiographic density are the most important factors when analyzing individual calcifications. Mammographybased screening programs are carried out in many countries, and their effectiveness has had a great impact on prognoses[7, 8]. Tumors in the breast can be benign (not cancer) or malignant (cancer). Benign tumors are not as harmful as malignant tumors: Benign tumors: It is rarely a threat to life It can be removed and usually don t grow back Don t invade the tissues around them Don t spread to other parts of the body 6

7 Malignant tumors: It may be a threat to life Often can be removed but sometimes grow back It can invade and damage nearby organs and tissues (such as the chest wall) It can spread to other parts of the body Types of breast cancer Breast cancers are classified as non-invasive or invasive. Most breast cancer occurs in the ducts of the breast the tubes that carry breast milk to the nipple. This breast cancer is called ductal carcinoma. A second, but least common form of breast cancer occurs in the lobules where breast milk is made. This breast cancer is called lobular carcinoma. Noninvasive: noninvasive breast cancer is an abnormal growth of cells still within the area in which it started. These cancer cells have not invaded into surrounding breast tissue. Ductal carcinoma in situ (DCIS) is a non-invasive breast cancer and referred to as stage 0. In situ means in place. Although DCIS and lobular carcinoma in situ (LCIS) sound similar, LCIS doesn't consider breast cancer. LCIS is a risk factor for breast cancer. Invasive cancer When breast cancer cells spread into surrounding breast tissue from the ducts or lobules, the cancer is called invasive. This increases the chance for cancer cells to spread to the lymph nodes. Inflammatory breast cancer (IBC) and Paget s disease of the nipple are two rare types of invasive breast cancer. Other less common forms of invasive breast cancer are medullary, mucinous, papillary and tubular carcinoma. Invasive breast cancer is not the same as metastasis. 7

8 Metastatic Metastasis occurs when breast cancer cells break away from the breast tumor and spread to other organs of the body through either the bloodstream or the lymphatic system Stages of breast cancer Stage 0 is sometimes used to describe abnormal cells that are not invasive cancer. The cancerous cells are in their original location within normal breast tissue known as either ductoral carcinoma or lobular carcinoma, depending on the type of cells involved and the location. It is a pre-cancerous condition and only a small percentage of ductoral carcinoma tumors progress to become invasive cancers. Stage I is an early stage of invasive breast cancer. Cancer cells invade breast tissue beyond where the cancer started, but the cells have not spread beyond the breast. The tumor is no more than 2 centimeters (three-quarters of an inch) across. Stage II is one of the following: The tumor is no more than 2 centimeters (three quarters of an inch) across. The cancer has spread to the lymph nodes under the arm. The tumor is between 2 and 5 centimeters (three quarters of an inch to 2 inches). The cancer has not spread to the lymph nodes under the arm. The tumor is between 2 and 5 centimeters (three quarters of an inch to 2 inches). The cancer has spread to the lymph nodes under the arm. The tumor is larger than 5 centimeters (2 inches). The cancer has not spread to the lymph nodes under the arm. 8

9 Stage III is locally advanced cancer. It is divided into Stage IIIA, IIIB, and IIIC. Stage IIIA is one of the following: The tumor is no more than 5 centimeters (2 inches) across. The cancer has spread to underarm lymph nodes that are attached to each other or to other structures. Or the cancer may have spread to lymph nodes behind the breastbone. The tumor is more than 5 centimeters across. The cancer has spread to underarm lymph nodes that are either alone or attached to each other or to other structures. Or the cancer may have spread to lymph nodes behind the breastbone. Stage IIIB is a tumor of any size that has grown into the chest wall or the skin of the breast. It may be associated with swelling of the breast or with nodules (lumps) in the breast skin: The cancer may have spread to lymph nodes under the arm. The cancer may have spread to underarm lymph nodes that are attached to each other or other structures. Or the cancer may have spread to lymph nodes behind the breastbone. Inflammatory breast cancer is a rare type of breast cancer. The breast looks red and swollen because cancer cells block the lymph vessels in the skin of the breast. When a doctor diagnoses inflammatory breast cancer, it is at least Stage IIIB, but it could be more advanced. Stage IIIC is a tumor of any size. It has spread in one of the following ways: The cancer has spread to the lymph nodes behind the breastbone and under the arm. 9

10 The cancer has spread to the lymph nodes above or below the collarbone. Stage IV is distant metastatic cancer. The cancer has spread to other parts of the body, such as the bones or liver. Recurrent cancer is cancer that has come back after a period of time when it could not be detected. Even when the cancer seems to be completely destroyed, the disease sometimes returns because undetected cancer cells remained somewhere in your body after treatment. It may return in the breast or chest wall. Or it may return in any other part of the body, such as the bones, liver, lungs, or brain[9-11] Mammogram Detection of microcalcification requires high resolution mammogram. Mammogram has a multidimensional nature of the data provided from different sequential pulses and has a much greater range of available soft tissue contrast, depicts the anatomy in greater detail, and is more sensitive and specific in abnormalities within the mammogram itself. Digital mammography is a specialized form of mammography that uses digital receptors and computers instead of x-ray film to help examine breast tissue for breast cancer. The proposed detection of masses in a mammogram is designed to help radiologists in the diagnosis of cancer at an early stage[12-15] Computer-Aided Diagnosis System Computer-Aided Diagnosis (CAD) system has been developed for Automatic detection of breast cancer through a mammogram. In Computer Aided Diagnosis System for Computer- Aided Diagnosis (CAD) and automated prescreening by computer are two ways to potentially 10

11 counter many of the problems that would result from screening a large number of women for breast cancer using mammography. Even if there is no large-scale screening program, computerized mammogram image analysis could perhaps be used to improve the quality of conventional mammography. Breast Cancer Detection using artificial bee colony optimization algorithm, the detection of microcalcification is performed in two phases: Preprocessing and Enhancement in the first phase and Segmentation and Classification in the second phase. Preprocessing and Enhancement techniques are used to improve the detection of the suspicious regions in the mammogram. In Preprocessing and Enhancement, initially the labels and film artifacts are removed from the obtained mammogram image and its visual appearance is improved. Bilateral subtraction describes the separation of the suspicious region from the background mammogram image. A Classification technique is used to find the cancer[16-20]. 1.2 MOTIVATION Early detection of breast cancer through the mammogram image analysis tools developed for the purpose of assisting radiologists. Digital Mammogram is increasingly being used in the clinical setting as an adjunct to x-ray mammography (which is, itself, the basis of breast cancer screening programs worldwide) and ultrasound. Of these imaging modalities, Digital mammogram has the highest sensitivity to invasive cancer and to multifocal disease. 11

12 Breast cancer analysis presently has two major shortcomings. First, although its sensitivity is high, its specificity is relatively poor; i.e. the method detects many false positives. Second, the method involves acquiring several high-resolution image volumes before, during and after the injection of a contrast agent. The large volume of data makes the task of interpretation by the radiologist both complex and time-consuming. These shortcomings have motivated the research and development of the computer-aided detection systems designed to improve the efficiency and accuracy of interpretation by the radiologist. 1.3 PROBLEM DEFINITION The medical image processing paves a means to detect the breast cancer. Most of the breast cancer detection methods provide compound information about the breast and they lack in providing an accurate result on existence of tumor. As a result, a formal consultation with a radiologist is mandatory, which becomes an unwanted expenditure in case of a non-cancer patient. Therefore, the problem here is to develop a system that provides an automatic result on the survival of microcalcification and if the cancer exists, the system elaborates the area being affected. 1.4 RESEARCH APPROACH Swarm Intelligence is a research field that studies the emergent collective intelligence of self-organized and decentralized simple agents. It is based on the social behavior that can be observed in nature, such as in flocks of birds, fish schools and bee hives, where a group of individuals with limited capabilities are able to emerge with intelligent solutions for complex problems. 12

13 Bees-inspired algorithms are more diverse, and some use pheromone and most do not. Almost all bee algorithms are inspired by the foraging behavior of honey bees in nature. Interesting characteristics such as waggle dance, polarization and nectar maximization are often used to simulate the allocation of the foraging bee along flower patches and thus different search regions in the search space. Honey bees live in a colony and they forage and store honey in their constructed colony. Honey bees can communicate by pheromone and waggle dance. For example, an alarming bee may release a chemical message (pheromone) to stimulate an attack response in other bees. Furthermore, when bees find a good food source and bring some nectar back to the hive, they will communicate the location of the food source by performing the so-called waggle dances as a signal system. Such signaling dances vary from species to species, however, they will try to recruit more bees by using directional dancing with varying strength so as to communicate the direction and distance of the found food resource. For multiple food sources such as flower patches, studies show that a bee colony seems to be able to allocate forager bees among different flower patches so as to maximize their total nectar intake. An optimization approach consists in exploring the search space to find the best solution according to an objective function and satisfying some constraints. It expresses a value of the cost to reduce or of benefit to maximize. The optimization framework is adapted to the problems 13

14 with multiple solutions. In traditional optimization problems, the objective function comprises a single criterion able to determine the optimum. Otherwise, the objective consists in finding a good compromise regarding multiple criteria. The problem is then a case of multicriteria optimization. In practice, optimization problems can reach a high complexity and require considerable computation times because of the number of potential solutions. The approximate methods are generally used to resolve this class of problems. These methods are based on an iterative exploration of the search space to find a solution of good quality in reasonable computation times. Among the most known approximate methods, it is mentioned the neighbourhood methods, such as Local Search, Simulated Annealing, Threshold Algorithms, Noising Method and Tabu Search. Tabu Search uses a local search procedure to iteratively move from a current solution x to a neighbor solution x' in the neighborhood of x, until some stopping criterion has been satisfied. The search process starts with an initial solution and moves from neighbor to neighbor as long as possible while improving the objective function value. Tabu Search improves the performance of the search process by using memory structures: the last explored configurations are kept in a short-term memory (Tabu list) in order to prohibit moves that lead to one of them. Due to simplicity and efficiency in navigating large search space for an optimal solution, ABCO is used in this research to develop efficient, robust and feasible algorithm to solve selected set of difficult problems in mammogram image analysis in the field of Medical image 14

15 processing. Out of these problems, ABCO and Tabu Search are elaborately tackled in this thesis, specifically image data. 1.5 METHODOLOGY The strategy of the asymmetry detection system can be described as follows: Given a pair of identical-view mammograms of the left and right breast, detecting all structural asymmetries between corresponding positions in the left and right breast. Significant asymmetries are taken for evidence in exploring possibilities of the cancer prevalence of a tumor. Figure 1.3 shows the block diagram for asymmetry approach. Figure 1.3: Block diagram for asymmetry approach. 15

16 The Mammography Image Analysis Society (MIAS), which is an organization of United Kingdom research groups interested in the understanding of mammograms, has produced a digital mammography database. The database which contains left and right breast images for 161 patients has been used. Its quantity consists of 322 images, which belong to three types such as normal, benign and malign. There are 208 normal images, 63 benign and 51 malign, which are considered abnormal. The benchmark database MIAS contains similar sizes of the left and right breast images. The preprocessing and enhancement activities such as removal of film artifacts and labels, filtering the image, normalization and removal of the pectoral muscle region. The enhancement method consists of four processing steps. In the first step, the given images are identified as left or right breast image and the film artifacts such as labels and X-ray marks are removed from the mammogram using the thresholding algorithm. In the second step, the high frequency components are removed using median filtering techniques. In the third step, the mammogram images are normalized to avoid the difference in brightness and contrast between the mammograms caused by the recording procedure. In the fourth step, the proposed modified tracking algorithm is applied to remove the pectoral muscle region from the breast region to increase the reliability of the segmentation of microcalcifications. The suspicious region or microcalcifications is segmented using bilateral subtraction for a pair of images. In bilateral subtraction, the asymmetries between corresponding 16

17 left and right breast images are considered for extracting the suspicious region from the background tissue. The breast border and the nipple position are used as reference points for alignment of mammograms. The novel method called ABCO is used to detect breast border and identify the nipple position. The spatial coordinates of the border points and nipple position are used to determine the angle of rotation to align the mammogram. After the right breast border is displaced, the correlation coefficient between the coordinates of the points on the left breast border and the coordinates of the points of the rotated right breast border are calculated for angles ranging between 5 and +5 with 1 step. The angle corresponding to the maximum value of the correlation coefficient is the angle of rotation of the right breast image, that is, the degrees that need to be rotated for the right breast border to match the left breast border. After the images are aligned, bilateral subtraction will be performed by subtracting the digital matrix of the left and right breast image. Microcalcifications in the right breast image have positive pixel values in the image obtained after subtraction, while microcalcifications in the left breast image have negative pixel values in the subtracted image. As a result, two new images are generated: one with positive values and the other with negative values. The most common gray value is zero, which indicates no difference between the left and the right images. The asymmetry images are used to find out suspicious regions through feature extraction, feature selection and finally classified as benign, malignant and normal. The textural features can be extracted from the asymmetry mammogram image. The textural analysis methods such as the Spatial Gray Level Dependency Matrix and the Gray 17

18 Level Run-Length Matrix are used to extract the fourteen Haralick features from the segmented image and their performance is studied. The features are selected from the extracted set of features using ABCO. The selected textural features are given as input to a three-layer Back Propagation Neural Network classifier, to classify the microcalcifications into benign, malignant or normal. The BPN optimizes the net for correct responses to the training input data set. Initially the features extracted from the textural analysis method are normalized between zero and one. The network is trained to produce the output value 0.9 for malignant images, 0.5 for benign images and 0.1 for normal images. The BPN classifier is validated using Tenfold Validation techniques. A Receiver Operating Characteristics analysis (ROC) is performed to evaluate the classification performances of the proposed approaches. The area under the ROC curve Az is used as a measure of the classification performance. A higher Az indicates better classification performance. The effectiveness of the proposed technique is determined by extracting the suspicious region from the mammogram image using bilateral subtraction. The true positive detection rate and the number of false positive detection rate at various thresholds of the asymmetry images are used to measure the algorithm s performance. These rates are represented using ROC. True positive (TP) and false positive (FP) rates are calculated and selected on the segmented image to generate the ROC curve. The ROC curve is a plot of the classifier s true positive detection rate versus its false positive rate. The FP rate is the probability of incorrectly classifying a non-target object (e.g. 18

19 normal tissue region) as a target object (e.g. suspicious tumor region). Similarly, the TP detection rate is the probability of correctly classifying a target object as being a target object. The ROC is the best suited to analyze the performances of segmentation. The area under the ROC curve (Az value) is an important criterion for evaluating diagnostic performance. The ROC curve is bound in the range between zero and one. The value of Az is 1.0 when the diagnostic detection has perfect performance, which means that the TP rate is 100% and FP rate is 0%. The Az value is computed by the trapezoidal rule. The Az value of the proposed algorithm is ORGANIZATION OF THE THESIS In this study, a novel method is proposed for image segmentation using soft computing. The thesis is organized into nine chapters. Chapter 1: INTRODUCTION The first chapter is introductory in nature and the subsequent chapters discuss the proposed technique in detail. The gist of each chapter is provided below. Chapter 2: LITERATURE REVIEW In this Chapter, the systematic overviews of the existing techniques for identifying the microcalcifications in digitized mammograms are summarized. In particular, the preprocessing, enhancement, bilateral subtraction techniques, ROC curve analysis and their performance are also studied and compared. 19

20 Chapter 3: IMAGE ACQUISITION, PRE-PROCESSING AND ENHANCEMENT In this Chapter presents the 161 pairs of mammograms which are obtained from the MIAS database to analyze the proposed methods and the mammogram images enhanced using filtering techniques. They are normalized and the pectoral muscle region is removed from the breast region. Chapter 4: BILATERAL SUBTRACTION FOR MAMMOGRAM IMAGES The Chapter 4 presents bilateral subtraction for mammogram images to extract the suspicious regions. The bilateral subtraction technique is used to extract the suspicious region from the digital mammograms based on identity asymmetries between left and right breast image. Chapter 5: FEATURE EXTRACTION AND FEATURE SELECTION In this chapter 5 textural analysis methods such as Spatial Gray Level Dependency Matrix (SGLDM) and Gray Level Run-Length Matrix (GLRLM) is used to extract the fourteen Haralick features from the segmented images. The features are selected using ABCO algorithms from the features extracted using textural analysis methods. Chapter 6: CLASSIFICATION The features selected in chapter 6 are given as input to the three-layer BPN to classify the microcalcifications into benign, malignant and normal ones. 20

21 Chapter 7: PERFORMANCE EVALUATION In this chapter, ROC analysis is presented to evaluate the classification performance of the textural features extracted by texture analysis method. The area under the ROC curve Az is used as a measure of the classification performance. Chapter 8: CONCLUSION In this concluding chapter, result obtained through this research work and contributions made are presented. Chapter 9: FUTURE WORK Scope for future research is discussed. 21