Multimodal MRI image fusion network for brain tumors based on feature cross enhancement

Image fusion has been widely applied to medical image fusion tasks. Still, the existing methods are dedicated to improving the overall quality of the fused images without focusing on the more critical texture details among the lesion tissues. To address the above research challenges, this paper proposes a Multi-modal MRI Image Fusion Network for Brain Tumors Based on Feature cross-enhancement (BTMC-Fusion). This network acquires global-local multiscale features of different modal images by alternating CNN and Transformer feature encoding. Then, in this paper, the proposed feature cross-enhancement module (FCEM) is used to realize the global-local feature fusion for the same modality. Then, this paper proposes the modal cross-enhancement module (MCEM) to realize the efficient enhancement of local features of different modalities. Qualitative and quantitative analyses of the fusion results of the two publicly available datasets, BraTS 2019 and AANLIB, show that the BTMC-Fusion can extract and fuse the multi-order information of multi-source images, and it has a significant advantage compared with other commonly used fusion methods.