Chapter 1. Introduction

Transcription

1 Chapter 1 Introduction 1.1 Introduction Medical imaging technologies are widely used in clinical diagnosis to guide therapeutic and surgical intervention and to monitor disease progression, recurrence and treatment response and to improve surgical navigation. A plethora of imaging modalities are available to the medical community to provide anatomical and functional information. Extensive research indicates that the application of computers to medical image analysis may lead to significant reduction in health care costs by increasing the speed of diagnosis, avoiding the need for expensive treatments and surgical procedures and reducing mortality rate through early screening programs. Mammography is a specific type of imaging that uses a low-dose X-ray system to examine breasts which is used to aid in the early detection and diagnosis of breast diseases in women. Among women around the globe, breast cancer is both the most common cancer and the leading cause of cancer-related deaths [1]. In India, breast cancer is the second most common cancer in females after cervical cancers [2] and [3]. Cancer is the result of normal cells developing mutations which make them grow too fast and uncontrollably. In the case of breast cancer, these cells are located in the breast tissue. Both the number of people getting breast cancer and the number of people killed by breast cancer are rising faster in the developing world than in developed nations [1]. While advances in recent years have helped in the battle against breast cancer, it is still a global epidemic. 1

2 With digital mammography, the image of the breast is acquired electronically and stored directly in a computer. The information technology revolution is now changing methods for the interpretation of digital mammogram images and these changes are embraced enthusiastically by both the medical profession and the public (patients). When viewing simple objects on a simple background of noise, doctors/radiologist perform nearly as well as ideal observers. However, they are limited in their ability to detect and diagnose disease during image interpretation due to their non-systematic search patterns. In addition, the vast amount of image data that is generated by some imaging devices makes the detection of potential disease a burdensome task and may cause oversight errors. Another problem is that the similar characteristics of some abnormal and normal tissues may cause interpretational errors. Developments in computer vision and artificial intelligence in medical image diagnosis have shown the potential for computers as providers of a second opinion in image interpretation [4], [5] and [6]. In future, it is likely that all medical images will have some form of computeraided detection/diagnosis (CAD) performed on them [7] and [8]. In this thesis, image processing techniques and pattern classification methods are employed to analyze digital mammograms to aid in the detection/diagnosis of cancer. 1.2 Digital Image Representation The term digital image refers to a two-dimensional picture or photograph made up of a set of picture elements called pixels. The digital image can be categorized into gray level image and color image. A gray level image can be defined as a two-dimensional function I(x, y), where x and y are the spatial or plane coordinates, and the amplitude of the image I at any pair of coordinates (x, y) is called the intensity or gray level of the image at that point. When (x, y) and the amplitude values of I are all finite, discrete quantities, then the image is a digital image. Each pixel in a gray level image is represented digitally in terms of 8 bits. The total number of levels are 256 [9]. 2

3 This initial chapter starts with a brief introduction to breast cancer, mammography and computer aided detection and diagnosis. In addition an overview of the various chapters of the thesis is provided. 1.3 Breast Cancer and its Types Breast cancer is an uncontrolled growth of breast cells. A tumor can be benign (not dangerous to health) or malignant (has the potential to be dangerous). Benign tumors are not considered cancerous: their cells are close to normal in appearance, they grow slowly, and they do not invade nearby tissues or spread to other parts of the body. Malignant tumors are cancerous. If left unchecked, malignant cells eventually spread beyond the original tumor to other parts of the body. The doctor reading the mammogram will look for different types of changes namely mass and microcalcifications. Masses: Breast cancer is characterized with the presence of a mass accompanied or not accompanied by calcifications [10]. The identical intensities of the masses and the normal tissue and similar morphology of the masses and regular breast textures make it a tedious task to detect masses in comparison with that of calcifications [11]. The location, size, shape, density, and margins of the mass are highly informative for the radiologist to evaluate the type of cancer. A majority of the benign masses are well circumscribed, compact, and roughly circular or elliptical whereas the malignant lesions are characterized by blurred boundaries, irregular appearances and are occasionally enclosed within a radiating pattern of linear spicules [12]. Microcalcifications: Tiny deposits of minerals (calcium) that appear like localized high-intensity regions (spots) in the mammogram are known as calcifications. Calcifications are one of the significant and widespread findings that are frequently apparent in a mammogram. Microcalcifications and macrocalcifications or coarse calcifications are the two common categories of calcifications. Macrocalcifications are coarse calcium deposits that are spread about the breast. Commonly such deposits are accompanied by benign conditions and hardly necessitate a biopsy. The benignity or malignancy of 3

4 the tumor is indicated by the number of calcifications that comprise a cluster. Microcalcifications are minute (less than 1/50 of an inch or of a millimeter) spots of calcium deposits that may exist in an area of rapidly dividing cells [13]. Fig. 1.1 shows mass and microcalcification. Fig. 1.1: Mass and microcalcifications. 1.4 Mammography Like all X-rays, mammograms use doses of ionizing radiation to create an image. Radiologists then analyze the image for any abnormal growth. At this time, mammography is the imaging modality for detecting and diagnosing breast cancer. It is the gold-standard with which other imaging tests are compared. Computed Tomography 4

5 has no real role in diagnosing breast cancer at present. Ultrasound, Ductography, and Magnetic Resonance are adjuncts to mammography. Fig. 1.2: Principle of mammography. Ultrasound is typically used for further evaluation of masses found on mammography or palpable masses not seen on mammograms. Magnetic Resonance Imaging (MRI) can be useful for further evaluation of questionable findings, or sometimes for pre-surgical evaluation to look for additional lesions. Mammography unit: A mammography unit is a rectangular box that houses the tube in which X-rays are produced. The unit is used exclusively for X-ray exams of the breast, with special accessories that allow only the breast to be exposed to the X-rays. Attached to the unit is a device that holds and compresses the breast and positions it so that images can be obtained at different angles. 5

6 Mammographic procedure: X-rays pass through most objects, including the body. Once it is carefully aimed at the part of the body being examined, an X-ray machine produces a small burst of radiation that passes through the body, recording an image on photographic film or a special digital image recording plate. During mammography, a specially qualified radiologic technologist will position the patient s breast in the mammography unit. The patient s breast is placed on a special platform and compressed with a paddle and X-rays is passed through the breast tissue. Fig. 1.2 shows the principle behind mammography. The technologist stands behind a glass shield during the X-ray exposure. The patient is asked to change positions between images. The process will be repeated for the other breast [14]. Screening mammograms: A screening mammogram is an X-ray exam of the breast on a woman who has no symptoms. Women in their fifties should receive screening mammograms every year. The goal of a screening mammogram is to find cancer when it is still too small to be felt by a woman or her doctor. Finding small breast cancers early by a screening mammogram greatly improves a woman s chance for successful treatment [15]. Screening mammography typically involves taking two views of the breast, from above (cranial-caudal view, CC) and from an oblique or angled view (mediolateral-oblique, MLO). Some patients may need to have more pictures to include as much breast tissue as possible. The mediolateral-oblique view is taken from the center of the chest outward. The cranio-caudal view is the view taken from above the head of the patient downwards. Figs. 1.3 and 1.4 show mammograms depicting these two views. Diagnostic mammograms: A diagnostic mammogram is an X-ray exam of the breast in a woman who either has a breast problem (for example, a breast lump, nipple discharge, etc.) or has something abnormal found during a screening mammogram. During a diagnostic mammogram, more pictures are taken to carefully study the area of concern. In most cases, pictures are enlarged to make a small area of suspicious breast tissue bigger and easier to evaluate. Many other types of X-ray pictures can be 6

7 Fig. 1.3: Mediolateral-oblique view (MLO). Fig. 1.4: Cranio-caudal view (CC). done, depending on the type of problem and where it is in the breast. For example, a diagnostic mammogram may show that an area that looks abnormal is actually normal when closely examined, and the woman can then return to routine yearly screening. Digital Mammogram: Both digital and film mammography use X-rays to produce an image of the breast. In film mammography, which has been used for over 35 years, the image is created directly on a film. Once a film mammogram is obtained, it cannot be significantly altered; if the film is underexposed, the contrast, which is the difference between the lightest and darkest areas on a display screen, is lost and cannot be regained. Fig. 1.5 shows the difference between the mammograms of the same nor- 7

8 Fig. 1.5: Difference between digital mammogram and film screen mammogram ( mal breast, taken using digital and film-based mammography. Digital mammography takes an electronic image of the breast and stores it directly in a computer. Digital mammography allows improvement in image storage and transmission because images can be stored and sent electronically. One of the principal advantages of any digital imaging system is the separation of image acquisition, process and display, allowing optimization of each of these steps. Additionally, advanced applications such as computer aided detection can be applied easily to the digital mammogram, assisting in image interpretation. Fig. 1.6 shows a digital mammogram unit. Larger clinical trials have been performed, demonstrating a statistically significant reduction in false positive results with digital mammography, compared with filmscreen examinations [16]. The study in [17] suggests that full-field digital mammography is superior to screen-film mammography for characterizing round and irregular masses and low and high density masses. Another study [18] shows that there is a significant increase in the number of breast cancers detected following the switch from film-screen to digital mammography. The number of cancers detected prior to 8

9 Fig. 1.6: Digital mammogram equipment ( the switch averaged between cancers per 1,000 women imaged. Following the switch, the cancer detection rate increased to 7.9 cancers per 1,000 women imaged and has remained high. Further the risk of misplacing or losing films/images has decreased. Radiologists say the images are much sharper and they are able to view calcifications better [19]. In general, regular screen-film mammograms and digital mammograms have similar accuracy. But digital mammograms have been shown to have some unique advantages. Some studies have found that women who have questionable areas on their mammogram have to return less often for extra imaging tests because with digital mammograms the original image can be magnified and looked at in many different ways on the computer screen. A recent large study also found that digital mammograms were more accurate in finding cancers in women younger than 50 years of age and in women with dense breast tissue [20]. 9

10 Computers have become an integral part of medical imaging systems and are used for everything from data acquisition and image generation to image display and analysis. As the scope and complexity of imaging technology steadily increase, more advanced techniques are required to solve the emerging challenges. The aim of medical image analysis is to identify and highlight hard-to-find features and anomalies on medical images that may indicate cancer and bring them to the attention of radiologists for further review, while also minimizing false-positive readings (i.e. identification of non-cancerous regions of interest as suspect) that burden physicians review process and which lead to higher recall rates. Despite the proved value of mammography in screening for breast cancer, its efficacy depends on radiologists interpretations. Medical imaging in diagnostic radiology has evolved as a result of the significant contributions of a number of different disciplines from basic sciences, engineering, and medicine. Medical image analysis demonstrates the benefits reaped from the application of digital image processing, computer vision, and pattern analysis techniques to biomedical images, such as adding objective strength and improving diagnostic confidence through quantitative analysis. This research work focuses on post-acquisition challenges such as image enhancement, detection of edges and objects, analysis of shape, quantification of texture and sharpness and pattern analysis, rather than on the imaging equipment and imaging techniques. Each chapter addresses several issues associated with mammogram image analysis, outlining the typical processes, then detailing more sophisticated algorithms directed to the specific problems of interest. Interpretation of breast images involves two major processes: Computer Aided Detection (CADe) and Computer Aided Diagnosis (CADi) collectively referred to as CAD. The image processing and pattern classification techniques have been used to help physicians to make important medical decisions through physician-computer interaction. Recently, intelligent or model based quantitative image analysis approaches have been explored for computer aided diagnosis to improve the sensitivity and specificity of radiological tests involving medical images [21], [22] and [23]. 10

11 1.5 Computer Aided Detection and Diagnosis in Mammograms Fig. 1.7: Computer analysis of digital images. Medical image analysis and image processing are often applied as an interpretive aid during the mammography image review. Techniques developed in computer vision and automated pattern recognition can be applied to assist radiologists in reading mammograms. Medical imaging centers, radiology departments and hospitals are embracing the benefits of computerized second read in mammogram images to detect breast cancers earlier. Radiologists report that on many occasions CAD systems help them to confirm that a suspicious area requires further investigation [24], [25] and [26]. The results of the study in [27] show that the computerized analysis methods for the diagnosis of breast mass lesions on digital mammograms are promising, and can potentially be used to aid clinicians in the diagnostic interpretation of digital mammograms. In [24] various tests were used to evaluate the statistical significance 11

12 of the differences in the performance measures without and with CAD and concluded that the performance improved with CAD. Fig. 1.7 shows the basic difference between digital mammogram and digital mammogram with CAD CAD Stages CAD encompasses various fields such as computer vision, image processing and artificial intelligence and can be classified into two categories according to their objectives, (a) those that are used to detect regions of interest (cancer) and (b) and those that are used to classify the detected regions of interest (cancer) based on the features extracted from those regions. The different stages of CAD can be classified as preprocessing, image enhancement, segmentation, feature extraction and classification to characterize abnormalities based on the features extracted. A description of each of these stages is given below. Preprocessing In order to limit the search for abnormalities by computer aided diagnosis systems to the region of the breast without undue influence from the background of the mammogram, removal of artifacts, detection of the breast contour and elimination of pectoral muscle is necessary. Breast contour detection and pectoral muscle elimination help to confine further analysis to the breast region alone which otherwise could bias the detection procedures in consequent stages. This work addresses two different techniques to remove artifacts: one based on morphological reconstruction and other by simple global thresholding. Breast contour is found by scanning the image for intensity differences after binarizing the image. Also two techniques are used to eliminate pectoral muscle: one using connected component labeling and the other using simple thresholding and raster scan technique. Image enhancement In this stage regions of interest are enhanced and the unwanted regions of the image are deemphasized. The enhancement procedure results in a better description of the objects of interest, thus improving the sensitivity of the detection system and leading 12

13 to better classification of the abnormalities in the case of diagnosis. The enhancement of the contrast of the regions of interest and the suppression of noise is performed in this stage. Median filtering and gray level transformations are done in this work to enhance the regions of interest. Segmentation Image segmentation is typically used to locate objects and boundaries (lines, curves, etc.) in images. More precisely, image segmentation is the process of assigning a label to every pixel in an image such that pixels with the same label share certain visual characteristics. Segmentation is a stage where a significant effort is made to delineate regions of interest and discriminating them from background tissue. In many cases the segmentation approach dictates the outcome of the entire analysis, since feature extraction and further classification of the abnormality depend on the accuracy of the segmented regions. Segmentation algorithms operate on the intensity or texture variations of the image using techniques that include threshold techniques, edge-based methods, region-based techniques, connectivity-preserving relaxation methods, deformable templates and pattern recognition techniques such as neural networks and fuzzy clustering. In this work, segmentation is carried out using thresholding, connected component labeling and the ground truth provided in the database helped in arriving at some criteria for the segmentation process. Feature Extraction Quantification algorithms are applied to the segmented regions to extract the essential diagnostic information such as shape, size, texture etc. Since the types of measurements and tissue vary considerably, a number of techniques that address specific applications are needed. In this work, statistical features are extracted to classify breast mass and the type of breast tissue. Also an effort has been made to distinguish normal and cancerous breasts using wavelet coefficients. 13

14 Classification A mathematical or statistical model called the classifier is used to classify the regions of interest (pathology) into different classes. Classifiers can be used to automatically derive knowledge from features extracted from the images and use this knowledge to recall previously seen patterns and classify new patterns with high accuracy. This knowledge may assist physicians in making the diagnostic process more objective and more reliable. In this work, statistical features such as mean, skewness, kurtosis etc., are extracted and these features are used to classify masses as malignant or benign. The breast tissue is classified as fatty, glandular and dense based on the statistical features extracted from the whole breast region excluding pectoral muscle. Further, wavelet coefficients are extracted and used to classify between normal and cancerous breasts. 1.6 Data Source This work is done using digital mammogram images obtained from the Mammography Image Analysis Society (MIAS) database and histopathological data obtained from Wisconsin Diagnostic Breast Cancer (WDBC) and Wisconsin Prognosis Breast Cancer (WPBC) database. MIAS is an organization of research groups interested in the understanding of mammograms, and has produced a digital mammography database. The X-ray films in the database have been carefully selected from the United Kingdom National Breast Screening Program and digitized with a Joyce - Lobel scanning micro densitometer to a resolution of 50 µm 50 µm and 8 bits represent each pixel. It has been reduced to 200 micron pixel edge and clipped/padded so that every image is pixels [28]. This database contains left and right breast images for a total of 161 patients. The images also include the locations of any abnormalities that may be present. The existing data in the collection consist of the location of the abnormality (like the center of a circle surrounding the tumor), its radius, breast position (left or right), types of breast tissue (fatty, fatty-glandular and dense) and tumor type if it exists (benign or malign). The radiologistś opinion and this ground truth provided in the database are used to validate the experimental results. 14

15 Wisconsin diagnostic breast cancer and Wisconsin prognosis breast cancer database was created by Dr. Wolberg to accurately diagnose breast masses based solely on a Fine Needle Aspiration (FNA) [29]. He identified nine visually assessed characteristics of an FNA sample which he considered relevant to diagnosis. Features are computed from a digitized image of a fine needle aspirate of a breast mass. They describe characteristics of the cell nuclei present in the image. An FNA is taken from the breast mass. This material is then mounted on a microscope slide and stained to highlight the cellular nuclei. A portion of the slide in which the cells are well-differentiated is then scanned using a digital camera and a frame-grabber board. Once all (or most) of the nuclei have been isolated a program computes values for each of ten characteristics of each nuclei, measuring size, shape and texture. The mean, standard error and extreme values of these features are computed, resulting in a total of 30 nuclear features for each sample. In this work, diagnosis and prognosis of breast cancer is done using these databases. 1.7 Objectives of the Thesis The objectives of this research work is to analyze mammograms to automatically detect and diagnose breast cancer using image processing and pattern classification techniques. The following issues are addressed: Preprocessing of mammograms which include enhancing mammograms for further analysis, artifact removal, breast contour detection and pectoral muscle elimination Breast tissue density analysis Mass detection, feature extraction and classification Categorizing breasts as normal or cancerous using wavelets and adaptive neurofuzzy inference system Diagnosis and prognosis of breast cancer using histopathological data 15

16 1.8 Organization of the Thesis The focus of the research work presented in this thesis is to automatically detect and diagnose breast cancers using image processing techniques and pattern classification methods. Barring Chapter 2, which deals with literature review, the remaining contents of the thesis are organized according to the above mentioned objectives. In Chapter 2, Section 2.2 reviews methods related to preprocessing of mammograms. Here the literature related to artifact removal, image enhancement and pectoral muscle elimination are covered. Section 2.3 discusses literature related to breast profile and breast mass segmentation. The different features used in the literature for the classification stage are discussed in Section 2.4. An extensive survey of pattern classification methods is done in Section 2.5. In Chapter 3, methods for artifact removal and breast boundary detection using mathematical morphology, connected component labeling, global thresholding and raster scan techniques are proposed. This chapter also discusses two methods for pectoral muscle removal, one using connected component labeling and the other using simple thresholding and raster scan technique. Chapter 4 describes breast density classification using support vector machines (SVM) and backpropagation neural networks (BPNN). Statistical features extracted from the breast region are used for creating classification models. Chapter 5 describes three different methods for mass segmentation. SVM and BPNN are used to classify breast mass using statistical features extracted from the region of interest and the performance of the classifiers is given in this chapter. Chapter 6 describes wavelets and presents a method for classification of breast into cancerous and non-cancerous using wavelet coefficients and adaptive neuro-fuzzy inference system. In Chapter 7, diagnosis and prognosis of breast cancer using histopathological data is presented. Chapter 8 summarizes the work presented in this thesis. This Chapter also highlights the contributions of this research and the directions for future research. 16