EffiNet-4HA: A Hybrid EfficientNetV2 And Multi-Head Attention Model For Accurate Brain Tumor Classification From MRI

Abstract

We propose EffiNet-4HA, a novel hybrid architecture for brain tumor classification from MRI images, which integrates the hierarchical feature extraction capability of EfficientNetV2-S with a Multi-Head Self-Attention (MHSA) mechanism to overcome the limitations of conventional convolutional networks in capturing long-range dependencies. The EfficientNetV2 backbone utilizes compound scaling along with Fused-MBConv and MBConv blocks to efficiently extract both shallow and deep features. To enhance feature representation, a four-head selfattention module is incorporated, enabling the model to focus on spatially and semantically important regions across the image. This integration allows effective modeling of both local features and global contextual information, improving the discrimination between visually similar tumor classes. The refined features are passed through a lightweight classification head consisting of pooling, dropout, and fully connected layers, followed by a softmax function for tumor classification. The model is trained using optimized preprocessing, augmentation, and regularization strategies to ensure robust performance. Experimental results on multiple benchmark datasets (BTMRI, BR35H, and EBTMRI) demonstrate that EffiNet-4HA achieves state-of-the-art performance, reaching up to 98.38% mean accuracy, outperforming several established deep learning models. The proposed approach provides an effective balance between accuracy and computational efficiency, making it suitable for real-world clinical applications. Overall, EffiNet-4HA offers a scalable and reliable framework for automated brain tumor classification, contributing to improved diagnostic support in medical imaging.

Ähnliche Arbeiten