Accurate classification of brain tumors using a hybrid approach of EfficientNetB۴ and Vision Transformer

Abstract

In this study, a hybrid model named FusionNet was proposed for the classification of brain tumors from MRI images. FusionNet consists of EfficientNetB4 and a Vision Transformer, which simultaneously extracts local features and long-range dependencies from the images. Experiments on a validated dataset demonstrated that this model, with an accuracy of 99.26%, performs better than baseline architectures and identifies three common types of brain tumors with high accuracy and a high F1-score. Analysis of the results indicates that combining the local features of EfficientNetB4 with the Transformer's ability to understand global image relationships plays a key role in improving performance. These findings suggest that FusionNet can serve as a high-efficiency automated brain tumor diagnosis tool suitable for application in clinical environments.