An Efficient Ensemble Deep Learning Comparative Study Based on Multi-Class Classifications of Brain Tumors Using Brain MRI Images

Brain tumors are a severe health problem that requires accurate diagnosis and classification for optimal treatment. Magnetic resonance imaging (MRI) is a widely used technique for detecting and analyzing brain tumors. However, manual classification of MRI images can be time-consuming and error prone. Therefore, computer-aided diagnosis systems (CAD) based on deep learning models are proposed to automate and improve this task. This paper presents an efficient deep-learning framework for the multi-class classification of brain tumors using MRI images. A large and customized dataset of 7023 MRI images, obtained from Kaggle and other sources, was used for training the models. The images were divided into four classes: pituitary tumors, meningiomas, gliomas, and non-tumors. The images were prepared for model training using a variety of data preparation methods. Nine different deep learning models, namely VGG16, VGG19, Xception, Inception V3, ResNet101, MobileNet, DensNet121, ResNet50, and InceptionResNetV2, were then used to extract complex visual features and perform classification tasks. Our framework's performance has been evaluated using metrics for recall, accuracy, and precision. Findings were highlighted and compared with those of other studies. The ensemble framework outperforms some of the individual models and the current techniques, achieving remarkably high accuracy values. Using MRI scans, the suggested deep learning method shows promise for multi-class categorization of brain cancers.