Abstract:
The review, analysis, and comparison of the performance of pre-trained models
for the classification of brain tumors based on magnetic resonance imaging is the
primary objective of this thesis (MRI). In the first part of the study, some background
information on artificial neural networks, such as perceptrons and multilayer
perceptrons, as well as the frameworks that are going to be used in the study, is
presented. The background of brain tumors, including their causes, symptoms, and
diagnostic approaches, is also covered in this article. The use of pre-trained models
in image classification, and more specifically in brain tumor MRI classification, is
analyzed in the section of the paper devoted to the literature review. Additionally, the
various classification algorithms that are put to use in this area of research are
discussed. The MRI brain tumor dataset and the pre-trained models (VGG16,
ResNet50, and InceptionV3) that will be compared are then described in detail in the
following section of the study. After that, feature extraction, fine tuning, transfer
learning, and a combination of these four methods are put into action and evaluated
as potential methods for classifying brain tumors using the previously trained
models. The outcomes of these methods are charted and compared using a variety of
metrics, and the models that perform the best are retained for further examination.
The purpose of this study is to provide insight into the most effective methods for
classifying brain tumors using pre-trained models and to suggest potential directions
for future research in this area. The models used in this study were obtained from the
Brain Tumor Classification Dataset available publicly on Kaggle.