HARNESSING EXPLAINABLE AI: HYBRID DEEP LEARNING AND MULTI-HEAD ATTENTION MECHANISMS FOR ENHANCED MRI ANALYSIS

Abstract

Brain tumors is a the most life-threatening neurological diseases, with malignant cases often resulting in low survival rates. Surgery, chemotherapy, and radiotherapy are common treatments, but their success strongly depends on accurate and suitable diagnosis. Magnetic Resonance Imaging (MRI) has emerged as the most promising modality for brain tumor analysis. However, analyzing MRI scans can be a tedious task, especially for large images, and is highly subjective, prone to errors, and challenging for radiologists.this research, we develop a novel deep hybrid learning model, Res-MobAT-BRTC, for early and precise classification of brain tumors using MRI images. The model integrates Mobile Net and ResNet-50 as backbones to effectively extract both low- and high-level features, while a multi-head attention mechanism enhances discriminative capability and provides explainable outputs for clinical experts.The study employs benchmark datasets such as BraTS and Figshare MRI collections, encompassing different tumor types as well as normal instances. Experimental results demonstrate that the proposed framework is superior in terms of accuracy, precision, recall, and F1-score to existing machine learning and deep learning models.Despite the high performance, there are still challenges, including imbalanced datasets, computational complexity, and generalizability.Future work will focus on growing the dataset, integrating explainable AI (XAI) mechanisms to improve clinician trust, and optimizing the framework for multi-modal imaging integration. By addressing diagnostic challenges through a robust hybrid deep learning-based model, this research contributes to advancing automated brain tumor analysis, facilitating earlier detection, reliable classification, and ultimately improving patient survival rates