<p>Multimodal medical image fusion integrates information from different imaging modalities, offering more comprehensive visual support for clinical diagnosis. However, existing methods often struggle to balance global structures and local details, leading to structural distortion, blurred textures, or redundant information. To address these issues, this paper proposes a Global–Local Collaborative Medical Image Fusion Network (GLCMNet) based on sparse attention and dynamic weighting. GLCMNet adopts a multi-scale N-shaped architecture for hierarchical feature extraction and incorporates a Sparse Feature Extraction Module (SFEM) that combines Top-K selection with an enhanced feed-forward network (ConvFFN) to adaptively highlight key regions and enhance local structural detail. Additionally, a Feature-Guided Fusion (FGF) module generates spatial importance maps (SIM) for each channel and applies dynamic weighting, enabling collaborative optimization of global context and local precision. To improve fusion quality, we introduce a global–local collaborative loss that integrates Structural Similarity Index (SSIM), Mean Squared Error (MSE), and Regional Mutual Information (RMI), ensuring both structural consistency and detail preservation. Experimental results show that GLCMNet outperforms existing methods across various medical image fusion tasks and demonstrates strong generalization and robustness in capturing both global structures and local details.</p>