SFC-Net: A spatial-frequency collaborative network for multi-modal medical image fusion

Multimodal medical image fusion aims to integrate complementary information across imaging modalities, enabling more comprehensive and accurate clinical diagnosis. Existing fusion algorithms typically focus on spatial domain features while neglecting to fully utilize frequency domain information, resulting in the loss of texture information. To address this issue, we propose a spatial-frequency collaborative network (SFC-Net) and apply it to the fusion of multimodal medical images. SFC-Net primarily consists of two key modules: a spatial domain feature fusion module and a frequency domain feature fusion module. The former utilizes a hybrid convolutional mechanism to extract both local and global features from the source images at different scales, thereby preserving spatial details such as texture and gradients. The latter incorporates a frequency domain attention block to enhance amplitude and phase information, effectively capturing high-frequency details. Finally, we enhance high-frequency details in the spatial domain features of the fused image by cascading frequency domain features and spatial domain features, thereby achieving synergy between spatial and frequency features. Experimental results show that, based on the average results across all datasets, SFC-Net improves upon the suboptimal values on M I , V I F , Q F M I , and Q P metrics by 22.6%, 15.3%, 6.4%, and 5.6%, respectively. This also confirms that SFC-Net can achieve excellent visual quality and high-frequency detail preservation while maintaining strong generalization ability. Moreover, downstream segmentation experiments further demonstrate that the fused images generated by SFC-Net can provide effective structural and semantic information for subsequent medical image analysis tasks.