An optimal control framework for joint-channel parallel MRI reconstruction without coil sensitivities.

Area of Science:

• Medical Imaging
• Magnetic Resonance Imaging
• Computational Imaging

Background:

• Parallel MRI (pMRI) accelerates image acquisition but requires sophisticated reconstruction techniques.
• Existing pMRI methods often struggle with calibration and effective utilization of multi-coil data.
• Reconstruction accuracy is crucial for diagnostic quality in MRI.

Purpose of the Study:

• To develop a novel, calibration-free parallel MRI reconstruction method.
• To leverage a discrete-time optimal control framework for improved regularization and feature extraction.
• To recover both magnitude and phase information efficiently.

Main Methods:

• A variational model with a learnable objective function was developed.
• An adaptive multi-coil image combination operator was integrated.
• The reconstruction network was formulated as a structured discrete-time optimal control system, using Lagrangian methods equivalent to back-propagation for training.

Main Results:

• The proposed method demonstrated promising performance in numerical experiments on real pMRI datasets.
• Comparisons with state-of-the-art pMRI reconstruction networks showed significant improvements.
• The method effectively learned regularization by sharing information across channels.

Conclusions:

• The developed method offers a general framework for deep network design and training in pMRI.
• It enables efficient joint-channel reconstruction by integrating learned operators and dual-domain regularization.
• The approach achieves highly efficient image reconstruction for pMRI.