Attention-Enhanced EfficientNet Framework for Multi-Class Brain Tumor Classification: A Deep Learning Approach

Accurate brain tumor classification based on Magnetic Resonance Imaging (MRI) is of significant importance for the diagnosis and prognosis of brain tumors; however, it is a tedious and complex task due to the involvement of a large number of classes and the need for accurate and precise classification results, as existing deep learning models are only able to perform binary and three-class classification tasks, and are lacking in providing interpretable results. Therefore, in order to address the aforementioned challenges, a novel classification framework, i.e., attention-enhanced EfficientNetB4 with a Convolutional Block Attention Module, is proposed for brain tumor classification, where the proposed model is able to perform a 14-class classification task with improved accuracy and better interpretability of results, as it is based on the transfer learning of a pre-trained EfficientNetB4 model on ImageNet, and the use of a focal loss function for handling the problem of class imbalance, as well as a two-phase fine-tuning strategy for better adaptation of features. Grad-CAM is used for providing better interpretability of results, and the proposed model is validated through the calculation of accuracy, precision, recall, F1 score, and confusion matrix, where improved results are obtained compared to existing CNN models.