Brain tumor is a common tumor and is damaging depending upon the type of tumor and the stage at which it is diagnosed. It is revealed by a doctor using magnetic resonance imaging of the brain. Analyzing these images is an exacting task, and human intervention might be a scope of error. Therefore, applying deep learning-based image classification systems can play a crucial role in classifying several tumors. This chapter aims to implement, analyze, and compare pre-trained convolutional neural network models and a proposed neural architecture to classify brain tumors. The dataset includes 7000 images classified into four classes of tumors: glioma, meningioma, no tumor, and pituitary. The proposed methodology involves cautious analysis of data and the development of a deep learning model. This has produced testing results with high accuracy of 99.0% and an error rate of 6.8%. According to the experimental findings, the proposed method for classifying brain tumors has a respectable level of accuracy and a low error rate, making it an appropriate tool for use in real-time applications.
TopIntroduction
Deep Learning is a blooming field of machine learning that is used almost everywhere, from speech recognition and facial recognition to self-driving cars. It aims at mimicking the functioning and decision-making ability of the human brain. One of its prominent applications is in healthcare. Its ability to solve highly exacting and extremely complex problems makes it a significant asset to the healthcare and medical industry. An example of such a complex and high accuracy demanding problem is the detection and classification of brain tumors. Because they can press on or spread into healthy areas of the brain, brain tumors are harmful. Additionally, some brain tumors have the potential to develop into cancer (Zülch, K. J., 2013).
They may be problematic if they obstruct fluid movement around the brain since this may raise the pressure inside the skull. Certain malignancies have the potential to travel through the spinal fluid to distant regions of the brain or spine. Any portion of the brain or skull can develop a brain tumor, including the brain's protective coating, the base of the head, the brainstem, the sinuses, the nasal cavity, and many other places. Depending on the tissue from which they originate, the brain can develop any one of more than 120 distinct tumor forms. Generally, tumors are of two types malignant (cancerous) and benign (non-cancerous) tumors, and for this chapter, we will discuss only the primary three brain tumors: glioma, meningioma, and pituitary. Inside the brain and certain areas of spinal cord, a peculiar kind of glioma is observed. Meningioma develops from the meninges, the membranes that cover the brain and spinal cord. It is included in this group even if it is not strictly a brain tumor since it might compress or pressure the nearby brain, nerves, or blood vessels (Lanktree, C., & Briere, J., 1991).
Diagnoses of the tumor generally consist of a neurological exam and image testing. A neurological exam is a basic test in which a doctor checks a patient's abilities like hearing, vision, coordination, and strength; any problem in these abilities gives the doctor a clue of which kind of tumor it may be (Alqudah, A. M., et al., 2020). Brain cancers are frequently diagnosed using brain images with magnetic resonance imaging (MRI). MRI gives a detailed and greyscale image of the internal organs of the human body. A doctor analyses the brain MRI scans and diagnoses whether a patient has a tumor or not. This procedure may seem easy theoretically, but it is a highly complex task when performed practically, as there is a vast scope of human errors. This pool of errors was considered and was attempted to eliminate with the help of machine learning (ML) and deep learning (DL) techniques. Deep learning is a sub-field of machine learning which works on algorithms and models that mimic the human brain. That can be achieved with the help of neural networks with three or more layers, which aim to solve extremely abstruse problems (Alzubaidi, L., et al., 2021).
Convolutional Neural Network (CNN) is a trendy, state-of-the-art neural network widely used in the medical industry to diagnose diseases using CT and MRI scans. It is a robust image recognition algorithm that can extract features from an image and convert them into low dimensions without losing their features. Its architecture generally consists of an input layer (input image), a convolutional layer responsible for extracting features from the picture, a pooling layer that aims to reduce computational costs, and a fully connected layer used to join neurons in two different layers (Aggarwal, G. et al, 2020). Machine learning approaches to classify brain tumors through image recognition are made possible due to advancements in technology and algorithms. The use of Support vector machines (SVM) and decision trees yielded results that were not promising enough. Applying deep learning and neural networks like CNN for brain tumor categorization produced fascinating results (Bharati, P., et al., 2020).