ABSTRACT
The brain is one of the most important organs of humans. Tumors in the brain occur with the abnormal development of cells in the brain tissue. Brain tumors seriously affect people’s lives and can cause death. Accurate and early detection of tumors in the brain is very important for treatment. Magnetic resonance imaging (MRI) method is frequently used today for the detection of brain tumors. Tumor regions can be distinguished by using magnetic resonance (MR) images such as texture, brightness, and contrast. In this study, it is aimed to classify glioma, meningioma, pituitary brain tumors, and healthy category by creating a hybrid model with deep learning (DL) and machine learning (ML) algorithms. In the model, feature maps are created from MR images using the VGG-16 model, which was previously trained for the ImageNet dataset, and classification is made with ML algorithms. As classification algorithms, linear regression (LR), support vector machines (SVM), k-nearest neighbor (KNN), decision trees (DT), random forest (RF), AdaBoost, naive Bayes (NB), and multilayer perceptron (MLP) are used. When their performances are compared, MLP gives the highest performance score with an accuracy of 0.973 and an F1-score of 0.971. Since DL methods work with the black-box approach, reliability problems occur. In order to express the transparency and intelligibility of the model, visualization is performed by applying the gradient weighted class activation map (Grad-CAM) 338algorithm to the VGG-16 last layer, one of the convolutional neural networks (CNNs) used for feature extraction.
