Answers are grounded in this paper and cite their sources.
BACKGROUND: Compressed sensing magnetic resonance imaging (CS-MRI) aims to recover high-quality MRI images from undersampled k-space data. Deep unfolding networks (DUNs) have shown strong potential in CS-MRI due to their interpretability and reconstruction capability. However, existing methods still face two limitations: insufficient modeling of the complex-valued structure of MRI data and inadequate utilization of complementary k-space and image-domain information. These limitations may cause detail loss or noise artifacts in reconstructed images. PURPOSE: This study aims to address these two challenges by developing a dual-domain unfolding framework that effectively integrates complementary information from the k-space and image domains for CS-MRI reconstruction. METHODS: This paper proposes a dual-domain fusion unfolding network (DDFU-Net) to address the aforementioned challenges by effectively integrating complementary information from the k-space and image domains. DDFU-Net introduces two core modules: the k-space reconstruction module (KRM) and the fusion attention module (FAM). The k-space reconstruction module processes complex-valued k-space data through real-imaginary decomposition and amplitude-phase representation modeling, with real-valued convolutional operations used for feature extraction. This design enables the network to exploit complementary information contained in the complex-valued k-space representation and improves detail recovery under high acceleration. The FAM promotes cross-domain interaction between the k-space and image-domain branches, enabling complementary feature fusion and improving detail restoration. A data consistency module is further incorporated to preserve the acquired k-space measurements during reconstruction. RESULTS: On the Brain and fastMRI datasets, DDFU-Net achieves competitive or superior reconstruction performance across the evaluated undersampling settings, with clear improvements in most comparisons. Ablation studies further validate the contribution of the major components. CONCLUSIONS: The proposed DDFU-Net achieves strong reconstruction performance under different undersampling settings, demonstrating the effectiveness of the coordinated dual-domain reconstruction framework.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: