**2. Machine learning**

**1. Introduction**

58 Breast Cancer and Surgery

Breast cancer is one of the most dangerous and common reproductive cancers that affect mostly women. The oldest documented cases of breast cancer were in Egypt in 3000 BC [1]. Breast tumor is an abnormal growth of tissues in the breast, and it may be felt as a lump or nipple discharge or change of skin texture around the nipple region. Cancers are abnormal cells that divide uncontrollably and are able to invade other tissues. Cancer cells have the ability to spread to other parts of the body through the blood and lymphatic systems [1]. It is the leading cause of death among middle aged and older women [1]. According to cancer statistics, breast cancer is the second most common and the leading cause of cancer deaths among women, second only to lung cancer [1]. Around 1 in 36 (3%) women dies due to breast cancer [2]. It has become a major health issue in the past 50 years, and its incidence has increased in recent years [1]; in Malaysia, breast cancer is the most frequent type of cancer among women. It has an incidence rate of about 26% (more than 4400 women) among cancer affecting women. Around 40% of the women who suffered from breast cancer in Malaysia have died (IARC). Hence, determining the right decision from a right diagnosis is crucial.

In today's world with the advent of personalized medicine, it increases the workload and complexity of the doctors in cancer diagnosis. Radiologic and pathology are the key players in making decision for cancer diagnosis. Based on the radiology diagnosis, the results will be submitted to pathology for further diagnosis. Pathology and radiology form the core of cancer diagnosis, yet based on our observation at our studied hospital and under current process of diagnostic medicine, the communication among them remained on papers. That paper contains their respective report of the case on the same patient. This scenario is in parallel with what James et al. [3] had highlighted in their paper. The working flows of both specialties remain ad hoc and occur in separate "silos," with no direct linkage between their case accessioning and/or reporting systems, even when both departments belong to the same host institution. Since both radiologists' and pathologists' data are essential to make correct diagnoses and appropriate patient management and treatment decisions, the isolation of radiology and pathology work flows can be detrimental to the quality and outcomes of patient care. These detrimental effects underscore the need for pathology and radiology work flow integration and for systems that facilitate the synthesis of all data produced by both specialties. With the enormous technological advances currently occurring in both fields, the opportunity has emerged to develop an integrated diagnostic reporting system that supports both specialties and, therefore, improves the overall quality of patient care. In this chapter, we are focusing on breast cancer diagnostic for data collected from UKMMC. Hence, breast radio-pathological correlation is essential. The covered topics would include radio-pathological correlation with recent imaging advances such as machine learning with use of technical methods such as mammography and histopathology. As a standard, the current diagnostic screening consists of a mammography to identify suspicious regions of the breast, followed by a biopsy of potentially cancerous areas. A breast biopsy is a diagnostic procedure that can determine if the suspicious area is malignant or benign [4–6]. Although criteria for diagnostic categories of radiologic and pathology are well established, manually detection and grading respectively is a tedious and subjective process and thus suffers from inter-observer and intra-observer variations. Early detection via mammography increases

ML comprises a broad class of statistical analysis algorithms that iteratively improve in response to training data to build models for autonomous predictions. In other words, computer program performance improves automatically with experience [9]. ML algorithm's aim is to develop a mathematical model that fits the data. It comprises of two types of learning which are supervised and unsupervised. Supervised learning algorithm required the data to be labeled for training purposes. For example, in training a set of medical images to identify a specific breast tumor type, the label would be tumor pathologic results or genomic information. These labels, also known as ground truth, can be as specific or general as needed to answer the question. The ML algorithm is exposed to enough of these labeled data to allow them to move into a model designed to answer the question of interest. Because of the large number of well-labeled images required to train models, curating these data sets is often laborious and expensive [10]. Unsupervised ML clusters the data that have similar characteristics, and the unlabeled data are exposed to the algorithm with the goal of generating labels that will meaningfully organize the data. This is typically done by identifying useful clusters of data based on one or more dimensions. Compared with supervised techniques, unsupervised learning sometimes requires much larger training data sets. Unsupervised learning is useful in identifying meaningful clustering labels that can then be used in supervised training to develop a useful ML algorithm. This blend of supervised and unsupervised learning is known as semi-supervised.

ML algorithms are to analyze any data set to extract data-driven model, prediction rule, or decision rule from the data set. Generally, in order to ensure the ML behave intelligently without human intervention, the system learns or extracts knowledge such as rules or patterns from a collection of input data or past experience. So the steps involved can be described as firstly, the system must acquire features from data. Elaboration of features is well explained in our previous work [11, 12]. Feature selection is very important as it contains information that can be used to train the system to identify specific patterns. The pixels are rich with qualitative abstractions or values of the input. Second step is analyzing all these features for detecting and classifying possible pattern or abnormality. Finally, the step is involving a ML algorithm to determine a best suitable model to represent the behavior or the pattern of the data [13].

that are not currently known or are beyond the limit of human detection [10]. **Figure 1** shows

Machine Learning Methods for Breast Cancer Diagnostic http://dx.doi.org/10.5772/intechopen.79446 61

In **Figure 1**, ML algorithm is implemented at the segmentation, feature extraction, and classification steps. One of the most popular and powerful ML algorithm for all the steps is support vector machine (SVM). SVMs are useful for taking a large number of features and discriminating inputs into one of two classes. SVMs, once trained, show the line or border that provides the greatest margin of separation. This concept can be extrapolated to a larger number of features (or dimensions), whereby the line of separation becomes an irregular plane known as a hyperplane. Because of the large number of features that can be combined mathematically, SVMs have been found useful for image processing. This chapter is focusing on SVMs for

Digital medical image recognition (DMIR) might give a promising solution. DMIR is considered as an essential aspect of artificial intelligence. DMIR techniques aim to extract specific information from medical images to assist doctors in diagnosing certain diseases and follow their progress. Many image processing techniques have been utilized in DMIR, such as segmentation, object detection, and classification. DMIR is concerned with numerous imaging modalities in the field of diagnosis including computed tomography (CT), digital mammography, magnetic resonance imaging (MRI), and microscopic histopathological images [16, 17]. Depending on the type of breast tissue, breast mass appears different in a mammogram. While it appears as solid block in dense breast, it appears as a roundish pie in a fatty breast. The mass may be alone or with microcalcifications [1]. In some cases, healthy breasts are also diagnosed as suspicious of cancer by the radiologist, and unfortunately, unnecessary biopsy is performed on them. Knowing that there are many possibilities of masses in breast cancer, detecting these features and localizing them are important. In general, localizing the mass is important in computer-aided detection, where it searches for the location in the mammogram images and segments it. Refs. [1, 18] examine the most important approaches used for mass segmentation in mammogram. In general, localizing the mass is important in computer-aided detection where it searches for the location in the mammogram images and do segmentation. Cheng et al. [18] examine the most important approaches used for mass segmentation in mammogram. Image segmentation using thresholding is the simplest way to isolate the object from its background when the image has a distinct gray level distribution. Segmentation separates the regions by assuming that the region that have gray levels below a specific value, called the threshold, as a background and the region with gray levels higher than the threshold as the object or vice versa. Identifying the threshold value is the key point in this algorithm. By selecting a representable threshold, object extraction will be more accurate. Mostly, image histogram is used to identify the threshold value. Mass localization method is discussed in this chapter. This section is based on our previous work on SVM rejection model for breast cancer. This method is a rejection model based on SVM algorithm used to reduce the FP of the output of the Chan-Vese segmentation algorithm that

the difference between CADe and CADx.

**3. Computer-aided detection**

was initialized by the MCWS algorithm.

both CADe for radiology and CADx for pathology diagnostics.

Various machine learning algorithms are now used to develop high-performance medical image processing systems such as computer-aided detection (CADe) system that detects clinically significant objects from medical images and computer-aided diagnosis (CADx) system that quantifies malignancy of manually or automatically detected clinical objects [14]. Therefore, CADe for mass in mammogram detects the suspicious region in the mammogram then tries to reduce the false positive and finally classifies this region to a mass or nonmass. In CADx for mass in a mammogram, most researchers use a region of interest (ROI) that contains the mass as an input to the CADx. Then, CADx tries to classify it into benign or malignant and gives the appropriate recommendation to do biopsy or follow-up screening [15]. Recent studies have shown that CAD systems, when used as an aid, have improved radiologists' accuracy of detection of breast cancer and also pathology decision [1, 7, 16]. It is worthwhile to distinguish ML from traditional computer-aided detection (CAD) algorithms. Traditional CAD algorithms are mathematical models that identify the presence or absence of image features known to be associated with a disease state. One of the examples is a microcalcification on a mammogram. Traditional CAD allows the developer to identify a feature explicitly and attempts to determine the presence or absence of that feature within a set of images. In contrast, ML techniques focus on a particular labeled outcome (ductal adenocarcinoma), and in the process of training, clusters of nodes evolve into algorithms for identifying features. The power and promise of the ML approach over traditional CAD is that useful features can exist

**Figure 1.** CADe vs. CADx. Source: Sampat et al. [7].

that are not currently known or are beyond the limit of human detection [10]. **Figure 1** shows the difference between CADe and CADx.

In **Figure 1**, ML algorithm is implemented at the segmentation, feature extraction, and classification steps. One of the most popular and powerful ML algorithm for all the steps is support vector machine (SVM). SVMs are useful for taking a large number of features and discriminating inputs into one of two classes. SVMs, once trained, show the line or border that provides the greatest margin of separation. This concept can be extrapolated to a larger number of features (or dimensions), whereby the line of separation becomes an irregular plane known as a hyperplane. Because of the large number of features that can be combined mathematically, SVMs have been found useful for image processing. This chapter is focusing on SVMs for both CADe for radiology and CADx for pathology diagnostics.
