Cancer is the major cause of death after cardiovascular infections. In comparison to other sorts of cancer, brain cancer has the lowest survival rate. Brain tumors have many types depending on their shape and location. Diagnosis of the tumor class empowers the specialist to decide the optimal treatment and can help save lives. Over the past years, researchers started investigating deep learning for medical disease diagnosis. A few of them are concentrated on optimizing deep neural networks for enhancing the performance of conventional neural networks. This involves incorporating different network architectures which are obtained by arranging their hyperparameters. The proposed idea of this chapter is concerned in providing implementation details of solutions for the problem of classifying brain tumors using classical and hybrid approaches combining convolutional neural networks CNN with classical machine learning. The authors assessed the proposed models using MRI brain tumor data set of three types of brain tumors (meningiomas, gliomas, and pituitary tumors).
TopIntroduction
Brain tumors can emerge from anomalous growth of the cells interior the brain or can come from cells that have spread to the brain from a cancer somewhere else. There is a wide assortment of brain tumor sorts that are classified in accordance with their cell of root. The essential tumors are those begun inside the brain. The larger part of essential brain tumors starts from glial cells (named glioma) and are classified by their histopathological appearances utilizing the World Health Association (WHO) framework into low grade glioma (LGG) and high-grade glioma (HGG).
The World Health Association (WHO) has created an evaluating framework to show a tumor's danger or generosity based on its histological highlights beneath a magnifying lens as Most malignant Quick growth, aggressive Broadly infiltrative, Quick recurrence Corruption inclined.
At present, imaging innovation may be a must for understanding diagnosis. The different restorative pictures like attractive resonance imaging (MRI), ultrasound, computed tomography (CT), X-ray, etc. play a critical part in the method of malady, diagnosing and treating. The later insurgency in restorative imaging comes about from techniques such as CT and MRI, can give nitty gritty information almost malady, and can distinguish numerous pathologies.
Doctors are unable to accurately diagnose and predict patient survival. They can also decide the appropriate choice of treatment which can range from surgery, followed by chemotherapy and radiotherapy, to a “wait and see” approach which avoids invasive procedures. Hence, tumor grading is an important aspect of treatment planning and monitoring. Exact detection and precise detection of abnormal tissues are vital for diagnosis. This reality is completely upheld by the presence of successful approaches utilizing division or classification or their combination for brain characterization both quantitatively as well as subjectively. Based on human interactivity, MR pictures can be handled using manual, semi-automatic, and completely programmed procedures.
In medical picture processing, segmentation/classification ought to be accurate which is in this way commonly performed by specialists physically and consequently time devouring.
A correct diagnosis, on the other hand, helps patients begin the correct treatment sooner and live longer. Therefore, there is an urgent need in the field of artificial intelligence (AI) to develop and design new and innovative computer-aided diagnosis (CAD) systems. It is intended to reduce the burden of diagnosing and classifying tumors and to assist physicians and radiologists.
Convolutional Neural Networks (CNNs) are one of the most popular deep neural networks that rely on mathematical linear equations between matrices called convolutions. A CNN has multiple layers. There are classical layers, pooling layers, nonlinear layers, fully connected layers, etc. Nonlinear and pooling layers have no parameters, but conventional and fully connected layers have parameters. There are different types of deep CNNs, such as Visual Pure Mathematics Cluster Networks (VGG-Net), Residual Networks (ResNet), Inception, Inception Resent50, and Xception.
Many ML algorithms have been used to classify brain tumors. Classical methods such as SVM, KNN, NN were used. In addition, deep learning techniques have also been used with promising results. I will introduce it in this work. A hybrid approach that combines a deep learning CNN as a feature extractor. Conventional ML was then used for classification. We compared different deep learning architectures for feature extraction. We also compared the base model with a model built using data augmentation.
The organization of this chapter is as follows: Section II presents an introduction about the field and highlights the problem of the current available systems for brain tumor detection. Section III surveys current applications and research to clarify opportunities and new features of the proposed system. Sections (IV) present system architecture and software Implementation details and results are discussed in Section V. Finally, conclusion and future work are presented in Section VI.