A Multimodal Adaptive Inter-Region Attention-Guided Network for Brain Tumor Classification

Accurate brain tumor classification is critical for ensuring timely and effective medical interventions. In recent years, artificial intelligence (AI)-driven diagnostic systems have emerged as transformative tools that optimize the classification process and enable rapid, objective decision-making. However, existing methods often suffer from limitations such as the loss of high-frequency details during multimodal preprocessing, inadequate cross-modal feature alignment, and insufficient focus on shared tumor regions within 3D architectures. To address these challenges, this study introduces a novel AI-based framework for advanced brain tumor classification. Specifically, we propose a multimodal magnetic resonance imaging (MRI) architecture that integrates Diffusion-Weighted MRI (DW-MRI) and T2-weighted MRI (T2-MRI) modalities, uniquely combining them in a dual-branch 3D neural architecture with advanced preprocessing and attention mechanisms. The preprocessing pipeline employs a learnable High-Frequency Information Retention (HFIR) technique to resize T2-MRI images, maintaining consistent spatial dimensions across modalities while preserving essential image details. The architecture utilizes dual-branch 3D convolutional neural networks (CNN) for modality-specific feature extraction, enhanced by a novel Adaptive Region Attention (ARA) module that dynamically aligns and emphasizes highly informative regions shared across modalities, providing deeper and more consistent insights into tumor characteristics. Rigorous evaluation on a dataset of brain MRI scans including three tumor classes demonstrates that the proposed framework achieves overall accuracy, sensitivity, and specificity of 92.86%, 80.00%, and 94.12%, respectively. Statistical analyses using bootstrap-resampled F1-scores confirm significant outperformance over other state-of-the-art models, underscoring its robust and interpretable potential for precise brain tumor diagnosis.