An EfficientNet and Dual Path Network Approach for Enhanced Brain Tumor Classification

Abstract

Brain tumor classification is an essential step in medical image analysis, contributing to timely diagnosis and effective treatment planning. This study introduces a brain tumor classification model that integrates EfficientNet with Dual Path Networks (DPN) and a Multi-Head Self-Attention (MHSA) mechanism. The model is applied to classify three major types of brain tumors—glioma, meningioma, and pituitary—using MRI images. The integration of DPN allows the model to leverage both residual and dense connections for enhanced feature representation, while the MHSA module refines global and local contextual information. Experimental evaluation demonstrates that the proposed model achieves an overall accuracy of 97.82%, sensitivity of 97.83%, specificity of 98.41%, precision of 98.34%, and F-score of 98.08%. These results indicate competitive performance compared to widely used architectures such as CNN, ResNet, and DenseNet, suggesting that the combined use of EfficientNet, DPN, and MHSA can provide a robust approach for brain tumor classification.