Using SVM and CNN as Image Classifiers for Brain Tumor Dataset

Using SVM and CNN as Image Classifiers for Brain Tumor Dataset

Maryam Zia, Hiba Gohar
Copyright: © 2023 |Pages: 24
DOI: 10.4018/978-1-6684-8696-2.ch008
OnDemand:
(Individual Chapters)
Available
$37.50
Current Special Offers
No Current Special Offers
TOTAL SAVINGS: $37.50

Abstract

Brain tumors make up 85% to 90% of all primary central nervous system (CNS) malignancies. Over a thousand people are diagnosed with cancer each year, and brain tumors are one of those fatal illnesses. It is challenging to diagnose this because of the intricate anatomy of the brain. Medical image processing is expanding rapidly today as it aids in the diagnosis and treatment of illnesses. Initially, a limited dataset was utilized to develop a support vector machine (SVM) model for the classification of brain tumors. The tumors were classified as either present or absent. As the dataset was small, the SVM model achieved great accuracy. To increase the dataset's size, data augmentation, an image pre-processing technique was used. Due to the SVM's limitations in producing high accuracy over a large dataset, convolutional neural network (CNN) was used to produce a more accurate model. Using both SVM and CNN aided in drawing comparisons between deep learning techniques and conventional machine learning techniques. MRI scans were used for tumor classification using the mentioned models.
Chapter Preview
Top

Brain Tumor Classification Using Support Vector Machine

Research Background

Among health problems, brain tumors are by far the most frequent. Primary brain tumors are detected in about 4,170 people under the age of 15 per year. A brain tumor consists of a collection of aberrant brain cells. There are several subtypes of brain tumors, and both benign and malignant variants exist. When a brain tumor becomes large enough, it presses on neighboring brain tissue or spreads to other areas of the brain, putting the patient at risk for severe complications, including high intracranial pressure.

Magnetic Resonance Imaging

Magnetic resonance imaging (MRI) is a medical imaging technology that creates extremely clear pictures of the body's anatomical structures, such as the brain. These detailed pictures are produced using a large, powerful magnet, radio waves, and a computer. As opposed to computed tomography (CT) scans, which employ X-ray radiation that can be harmful to human tissue, MRI scans pose no such risk to the brain. Radiologists routinely use MRI to identify brain tumors because it provides a cross-sectional view of the brain and distinguishes between healthy and malignant tissue more precisely. MRI is also favored by patients as it is a painless, non-invasive, and non-harmful diagnostic procedure.

Support Vector Machines

Based on the principles of statistical learning, Support Vector Machine (SVM) is a supervised machine learning algorithm that can be applied for both classification and regression applications. However, its primary application is in classification tasks. SVM's central idea is to utilize hyperplanes to build decision boundaries for classifying data points. The hyperplane separates into two classes, each of which may be assigned to data points that lie on either side of it. The hyperplane's location and orientation can be affected by the data points called support vectors, which are those points that are nearest to the hyperplane. The margins of the classifiers are optimized using these support vectors. Also, the hyperplane's location will shift if the support vectors are removed. Some advantages of SVM are increased speed and compatibility with unstructured and semi-structured data types, including text, images, and trees. SVM also works effectively with small datasets (in the thousands), which is a constraint, as we will see later in this research.

Convolutional Neural Networks

Convolutional Neural Networks (CNNs) is a form of neural network that are commonly used to solve image processing problems. One key feature that sets CNN apart from other neural networks is its ability to recognize patterns using its hidden layers. Convolutional layers, pooling layers, and fully connected layers are the three primary CNN layer types. The foundation of a CNN is the convolutional layer, which is made up of multiple feature maps and may be customized through the application of different kernels. Its objective is to learn a set of features that characterize the inputs. Adding a pooling layer can have the same impact as a supplementary feature extractor, reducing the size of feature maps while simultaneously boosting their resilience. Layers of convolution and pooling are stacked on top of one another to extract high-level features from inputs, often positioned between two convolutional layers. CNN typically has one or more fully connected layers serving as the classifier. All the neurons in the preceding layer are copied into the current layer, and each neuron in the latter is linked to all the neurons in the former. An output layer comes after the final fully connected layer. SVM is common and may be used in conjunction with CNNs to accomplish a variety of classification jobs.

Complete Chapter List

Search this Book:
Reset