Dual branch diffusion-Mamba hybrid network for multi-scale MRI modality translation.
Authors
Affiliations (6)
Affiliations (6)
- Institute of Electrical Engineering, Chinese Academy of Sciences, Beijing, China.
- University of Chinese Academy of Sciences, Beijing, China.
- Department of Radiology, Peking University Third Hospital, Beijing, China.
- College of Automation and Electrical Engineering, Shandong University of Aeronautics, Binzhou, China.
- Institute of Electrical Engineering, Chinese Academy of Sciences, Beijing, China. [email protected].
- University of Chinese Academy of Sciences, Beijing, China. [email protected].
Abstract
Objective MRI modality translation aims to synthesize missing modalities from acquired scans. However, existing methods often struggle to preserve local details while maintaining global structural consistency. We propose a dual-branch diffusion-Mamba hybrid network (DBDM-Net) to address this challenge. Materials and methods DBDM-Net consists of a diffusion-based local branch for fine-detail reconstruction and a Mamba-based global branch for structural coherence preservation. A local feature extraction encoder enhances local representations, while a modulation block and a cross-scale fusion block enable modality-specific adaptation and multi-scale feature integration. Results DBDM-Net achieves competitive PSNR, SSIM, and perceptual quality compared with representative baseline methods. Discussion The proposed dual-branch framework effectively combines local detail reconstruction with global structural modeling, supporting its utility for MRI modality translation.