Combined Attention-Based Fusion of Multiscale MRI Medical Images for Improving Early Brain Tumor Detection

Abstract

The effective diagnosis of early-stage brain tumors relies heavily on the analysis of multimodal medical images. To address this need, we propose a novel multimodal medical image fusion approach that utilizes convolutional neural networks (CNNs) for enhanced feature extraction and representation. Unlike conventional CNN-based fusion methods that employ straightforward weighted averaging, our method incorporates a "Multiscale Attention Fusion Module" and a "Visual Relevance Fusion Strategy" to refine the fusion process. Our methodology aims to effectively combine multiple MRI modalities while emphasizing the most crucial diagnostic information, thereby mitigating the issue of non-essential information that often degrades the quality of fused images. By integrating these innovative components, our research contributes to improved early brain tumor detection, ultimately enhancing the quality and efficiency of medical diagnoses.