DEMFusion: A Diffusion-Enhanced Medical Image Fusion Network

Multimodal medical image fusion technology integrates complementary information from diverse medical imaging modalities (e.g., CT, MRI) to overcome the physical limitations of single-modality imaging, enabling precise correlation and fusion of structural features with functional metabolic information in clinical practice. Existing multimodal medical image fusion methods exhibit limitations in preserving the unique information inherent to each modality, resulting in suboptimal fusion accuracy. To address these challenges, we proposes a diffusion-enhanced medical image fusion network (DEMFusion). The proposed directly learns the mapping relationships between input and fused outcomes while leveraging modality-specific information guidance to significantly enhance fusion accuracy. Within the proposed, one diffusion-enhanced module (TDModule) is designed to generate corresponding pseudo-medical images from real medical modalities, from which modality-unique information are extracted using both pseudomedical images and real medical images. Future more, the proposed method integrates surface-layer constraints and deeplayer constraints to objectively define unique information. The chromatic constraint is further designed to enhance chrominance retention in fused outcomes. Qualitative and quantitative experiments demonstrate that the proposed method effectively preserves modality-specific information while achieving superior fusion performance on various datasets.