<p>Multimodal medical image fusion aims to synthesize complementary information from multiple source images into a comprehensive composite. However, the absence of standardized ground-truth labels in clinical practice often limits the performance of existing methods. To overcome this challenge, we propose a novel two-stage knowledge distillation framework tailored for medical image fusion. First, we introduce a cross-model knowledge distillation module that transfers knowledge from a powerful teacher model to a lightweight student network, significantly enhancing its feature extraction capabilities. Second, we develop a self-knowledge distillation mechanism to capture critical cross-modal features based on component-specific attribute. Encouraging the network to concentrate on discriminative semantic and structural details. Extensive experiments demonstrate that the proposed method not only balances efficiency and performance but also outperforms several state-of-the-art approaches.</p>