Dual-MFNet: AI-Driven Dual-Scale Multimodal Fusion With State Space Networks for Personalized MRI Synthesis

Personalized healthcare increasingly relies on AI-driven multimodal fusion to enhance diagnostic precision and treatment planning. However, long MRI acquisition times, imaging artifacts, and missing modalities often lead to incomplete critical imaging information, limiting the application of multimodal MRI in personalized diagnostics. To address this challenge, we propose Dual-Scale Multimodal Fusion Network (Dual-MFNet), a novel AI-driven approach to personalized MRI synthesis for reconstructing missing modalities with high anatomical fidelity. Our method leverages state-space models to capture long-range contextual dependencies while preserving local structural integrity, ensuring accurate cross-modal synthesis. The Dual-Scale Feature Fuser (Dual-Fuser) balances global coherence with fine-grained detail preservation, while the Twin-Stream Fusion module (TSF) dynamically enhances critical cross-modal information. In addition, the Feature Aggregation (FA) module consolidates multimodal input into a cohesive representation, producing high-fidelity synthesized MRI customized to individual patient needs. To assess clinical relevance, we conducted extensive quantitative evaluations and a radiological reader study with five experienced radiologists. The results demonstrate that Dual-MFNet outperforms state-of-the-art methods, particularly in preserving tumor boundaries, fine tissue textures, and anatomical clarity, making it a valuable tool to advance personalized MRI-based diagnostics.