Triple-D network for efficient undersampled magnetic resonance images reconstruction.

Area of Science:

• Medical Imaging
• Computer Vision
• Signal Processing

Background:

• Compressed sensing (CS) accelerates magnetic resonance imaging (MRI) by acquiring partial k-space data.
• Conventional CS-MRI methods are computationally intensive and limited by fixed transforms or shallow dictionaries.
• Existing deep learning approaches primarily focus on image domain reconstruction, neglecting k-space modeling potential.

Purpose of the Study:

• To develop an advanced deep learning model for accelerated CS-MRI.
• To explore the synergistic potential of modeling in both k-space and image domains.
• To enhance the performance and applicability of CS-MRI reconstruction techniques.

Main Methods:

• Proposes the Dual Domain Dense network (Triple-D network), integrating k-space and image domain sub-networks.
• Employs dense connections for effective utilization of multi-level feature maps.
• Incorporates multi-supervision strategies and channel-wise attention (CA) layers to improve model capabilities.

Main Results:

• The Triple-D network demonstrates promising performance in compressed sensing MRI reconstruction.
• The model effectively utilizes features from both k-space and image domains through dense connections.
• Experimental results validate the network's robustness across different sampling trajectories and noisy conditions.

Conclusions:

• The Triple-D network offers a significant advancement in accelerated CS-MRI.
• Dual-domain processing and attention mechanisms enhance reconstruction quality and efficiency.
• The proposed method shows potential for broad application in various MRI scenarios.