LCMF-Net: A lightweight collaborative multimodal fusion network for brain tumor segmentation.

Accurate brain tumor segmentation plays a pivotal role in clinical decision-making, providing essential support for disease screening, treatment planning, and surgical navigation. Traditional manual segmentation methods are time-consuming and labor-intensive. While deep learning has substantially advanced medical image analysis, current methods still suffer from two key limitations: the inadequate integration of complementary features across multimodal MRI sequences, and the fact that high-accuracy methods often come with substantial computational costs. To overcome these issues, this study proposes a Lightweight Collaborative Multimodal Fusion Network (LCMF-Net), which achieves high-precision and high-efficiency tumor segmentation through the synergistic optimization of Cross-Modality and Cross-Slice Attention (CMCSA) and a State Space Model-based Fusion module (SSM-Fusion). LCMF-Net adopts a multi-branch encoder architecture, in which the CMCSA module dynamically enhances and calibrates features both across different modalities (T1, T2, T1ce, and FLAIR) and between adjacent slices. To facilitate effective cross-modal integration, SSM-Fusion is introduced, enabling adaptive feature fusion while maintaining spatial continuity. Additionally, a spatial dimensionality reduction strategy, combined with an improved Residual Inception Block (RIB), allows for efficient multi-scale feature extraction and 3D contextual modeling under 2D computational constraints. Extensive experimental results demonstrate that LCMF-Net reduces computational cost by more than 50 %, while achieving a 1.6 % improvement in segmentation accuracy compared to state-of-the-art (SOTA) methods. LCMF-Net demonstrates strong segmentation performance while maintaining low computational complexity, making it a practical and reliable solution for real-world clinical applications. The code is available at: https://github.com/IAAI-SIT/LCMF-Net.