Continuous-Domain Multimodal Medical Image Fusion With SNR-Guided High-Frequency Alignment

Medical image fusion integrates complementary information from multi-modal images into a single representation, enhancing structural clarity and supporting clinical diagnosis, surgical navigation, and intelligent analysis. However, conventional discrete processing often suffers from limited sampling resolution, leading to structural degradation and misalignment across modalities. To address these challenges, we propose a novel dual-branch implicit neural framework that leverages implicit representations for continuous-domain alignment and robust integration of heterogeneous features. The framework consists of a global branch for semantic context extraction and a high-frequency branch that preserves fine structural details, enhanced by an SNR-guided fusion module to adaptively emphasize diagnostically relevant regions while suppressing noise. An implicit alignment mechanism further ensures precise cross-modal consistency. Extensive experiments on CT–MRI, PET–MRI, and SPECT–MRI datasets demonstrate that our method consistently outperforms state-of-the-art approaches, achieving superior detail preservation, robustness to misalignment, and real-time efficiency. These advantages establish a solid foundation for practical deployment in downstream clinical tasks such as lesion detection and organ segmentation.