Dual-Branch Deep Unfolding Network for Compressed Sensing MRI Reconstruction

Area of Science:

• Medical Imaging
• Computer Vision
• Signal Processing

Background:

• Compressed Sensing Magnetic Resonance Imaging (CS-MRI) utilizes Deep Unfolding Networks (DUNs) for improved performance and interpretability.
• Existing DUN methods often process MR image components uniformly, neglecting unique characteristics and leading to suboptimal detail capture.

Purpose of the Study:

• To develop an advanced DUN-based method for CS-MRI that addresses the limitations of uniform component processing.
• To improve the reconstruction accuracy and detail preservation in under-sampled MR images.

Main Methods:

• Proposing the Dual-BrancH Deep Unfolding Network (DBH-Net) with parallel under-complete (UC) and over-complete (OC) branches.
• The UC branch extracts low-frequency features by expanding the receptive field, while the OC branch focuses on high-frequency features by restricting the receptive field.
• Implementing an Auxiliary Information Fusion Block (AIFB) for multi-channel information transfer between stages to minimize information loss.

Main Results:

• DBH-Net demonstrated superior performance compared to state-of-the-art methods across three experimental datasets.
• The dual-branch architecture effectively captures distinct image components, enhancing overall reconstruction quality.
• The AIFB successfully reduced information loss during the reconstruction process.

Conclusions:

• DBH-Net offers a significant advancement in CS-MRI reconstruction by intelligently handling different image components.
• The proposed architecture provides a more effective approach to detail preservation and performance enhancement in CS-MRI.
• The method's effectiveness is validated by extensive experimental results on multiple datasets.