Accelerating MR Imaging via Deep Chambolle-Pock Network^.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Signal Processing

Background:

• Compressed sensing (CS) accelerates Magnetic Resonance Imaging (MRI) data acquisition.
• Current CS-MRI methods face challenges with image detail loss and complex parameter tuning, especially at high acceleration factors.

Purpose of the Study:

• To develop a novel model-driven approach for Magnetic Resonance Imaging (MRI) reconstruction using deep learning.
• To address the limitations of existing CS-MRI techniques by improving image quality and simplifying parameter selection.

Main Methods:

• A deep neural network, CP-net, was developed, inspired by the Chambolle-Pock algorithm.
• CP-net learns the proximal operator and parameters within the Chambolle-Pock framework.
• The network reconstructs in vivo human brain MR images from highly undersampled k-space data.

Main Results:

• CP-net demonstrated superior accuracy in MR image reconstruction compared to state-of-the-art methods.
• The proposed method achieved better performance across various quantitative metrics.
• Effective reconstruction was achieved even with highly undersampled complex k-space data from diverse MRI scanners.

Conclusions:

• The proposed CP-net offers a significant advancement in CS-MRI reconstruction.
• This deep learning model effectively overcomes the detail loss and parameter selection issues in traditional CS-MRI.
• CP-net provides accurate and robust in vivo brain MR image reconstruction.