Brain tumors are prevalent and aggressive disease, with a relatively short life expectancy in their most severe form. Thus, treatment planning is an important element in improving patient quality of life. In general, image techniques such as computed tomography (CT), magnetic resonance imaging (MRI), and ultrasound imaging are used to examine tumors in the brain, lung, liver, and breast. MRI scans are used in this study to diagnose brain tumors. As a result, a reliable and automated classification technique is required to prevent death. Automatic brain tumor detection using convolutional neural networks (CNN) classification is proposed in this chapter. Small kernels are used to conduct the deeper architectural design. In machine learning, brain tumor classification is done by using a binary classifier to detect brain tumors from MRI scan images. In this chapter, transfer learning is used to build the classifier, achieving a good accuracy and visualizing the model's overall performance.
TopIntroduction
A brain tumor is one of the essential organs in the human body, consisting of billions of cells. The aberrant collection of cells is created by the uncontrolled division of cells, which is also known as a tumor. Brain tumors are classified into two types: low grade (grade 1 and grade 2) tumors and high grade (grade 3 and grade 4) tumors (Auzias, 2015). A benign brain tumor is one with a low grade. Similarly, a high grade tumor is sometimes referred to as a malignant tumor. A benign tumor is one that is not cancerous. As a result, it does not spread to other areas of the brain. The malignant tumor, on the other hand, is a cancerous tumor. As a result, it spreads quickly with indeterminate borders to other parts of the body. It results in instant death. The brain MRI picture is primarily used to detect tumors and to simulate tumor progression. Early detection of brain tumors can alter the disease's course and save lives. Computers, on the other hand, can execute complicated tasks in a relatively short amount of time. Computers are utilized in a variety of fields, including health care, since they provide automated, quick, and precise results (Bauer,2015). As a result, computer-aided diagnostic (CAD) systems are now widely employed. The intricate relationships are analyzed using CAD systems. Early diagnosis and categorization of a brain tumor is critical for a patient's successful and timely treatment. The capacity of the human visual cortex to distinguish between distinct shades of gray is known to be limited, as evidenced by magnetic resonance imaging (MRI). This gives rise to computer-aided diagnosis (CAD) or brain tumor classification (BTC) techniques, which can help radiologists, visualize and classify tumor forms. For the categorization of a brain tumor, traditional BTC techniques rely on low-level characteristics and the use of statistical learning methodologies. This group of segmentation methods focuses on estimating the tumor's borders and localization, which includes certain pretreatment processes including contrast enhancement, picture sharpening, and edge detection/refinement. Image capture, preprocessing, ROI segmentation, feature extraction and selection, dimensionality reduction, classification, and performance evaluation are the steps in the fundamental workflow of classic BTC techniques (Fresilli,2020). Deep learning-based methods, in contrast to traditional approaches, rely primarily on training data and require far less preparation. The research on deep learning shows that, notably in the area of BTC the accuracy of a system is strongly reliant on the amount of data. Convolutional neural networks are used in most deep learning approaches in BTC (CNNs). Indeed, the growing use of CNNs for a variety of computer vision issues in many fields encourages their use in BTC, especially for smart health monitoring. These automated brain tumor identification, segmentation, and classification techniques benefit mankind by minimizing the need for surgery (biopsy). These approaches are always available to aid radiologists who are unsure about the type of a tumor or want to visually study it in greater detail. Scientists working in image processing and computer vision are interested in developing precise and efficient techniques for automatically detecting, classifying, and segmenting tumors. This data is mostly utilized in tumor diagnosis and therapy procedures. An MRI picture contains more information about a particular medical imaging than a CT or ultrasound image (Feng,2019). An MRI picture offers precise information on brain anatomy as well as the detection of anomalies in brain tissue. Scholars have really presented unique automated ways for identifying and categorizing brain cancers utilizing brain MRI pictures since the time when it was able to scan and freight medical images to the computer (Hiran, K. K., et al., 2013). In contrast, during the last several years, Neural Networks (NN) and Support Vector Machines (SVM) have been the most utilized approaches for their good implementation. Deep Learning (DL) models (Mahrishi, M., et al.,2020), on the other hand, have recently established a stirring trend in machine learning, as the subterranean architecture can efficiently represent complex relationships without requiring a large number of nodes, as do superficial architectures such as K-Nearest Neighbor (KNN) and Support Vector Machine (SVM). As a result, unlike health informatics, they soon expanded to become the cutting-edge (Hiran, K. K., et al.2021).