Multimodal Perception Algorithm Based on Spatial Attention for Brain Tumor Segmentation

In brain tumor segmentation, the integration of multi-modal MRI images, such as T1, T1ce, T2, and FLAIR, is crucial for capturing comprehensive information from different tumor components. However, traditional methods often fail to address the global correlations between modalities and the local features within each modality, resulting in suboptimal segmentation accuracy. To tackle these issues, we propose a multi-modal perception segmentation algorithm based on a spatial attention mechanism. This algorithm incorporates two key modules: Modal Correlation Modeling (MCM) for capturing global modality relations and Spatial Feature Enhancement (SFE) for enhancing local feature details. By dynamically adjusting modality weights and emphasizing tumor regions through attention mechanisms, our method improves segmentation accuracy, particularly in complex tumor areas. Experiments on the BraTS dataset demonstrate that our approach outperforms existing methods in Dice score, IoU, and 95HD, highlighting its superior ability to handle the multi-modal nature of MRI data and provide precise tumor delineation.