Attention-Enhanced Mobile-Net for Multi-Class Brain Tumor Classification Using MRI Images

Abstract

Effective treatment planning for brain tumors relies on an early and accurate diagnosis. Manual MRI analysis involves a lot of work and is prone to inconsistent diagnoses. Using a Multi-Head Attention-enhanced MobileNet architecture, this paper presents a challenging in terms of technology automated classification system. The suggested model performs better in four classes: glioma, meningioma, pituitary, and non-tumor. It does this by utilizing depth-wise separable convolutions for computational efficiency and attention mechanisms for global feature dependency. A peak test accuracy of 96.3% can be seen in the experimental results, providing significant quantitative improvements over ordinary CNN and VGG19-based approaches.