One-Shot Generative Prior in Hankel-k-Space for Parallel Imaging Reconstruction

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Signal Processing

Background:

• Magnetic resonance imaging (MRI) is crucial for diagnosis but limited by long acquisition times.
• Deep generative models offer potential for accelerating MRI reconstruction.
• Learning data distribution for image reconstruction from limited data remains a challenge.

Purpose of the Study:

• To propose a novel Hankel-k-space generative model (HKGM) for accelerated MRI.
• To enable image reconstruction from minimal k-space data.
• To improve the efficiency and quality of MRI scans.

Main Methods:

• Constructing a Hankel matrix from k-space data.
• Extracting structured k-space patches for prior learning.
• Utilizing an iterative reconstruction with a generative model, low-rank penalty, and data consistency.

Main Results:

• The HKGM can generate samples from a single k-space dataset.
• Internal statistics of k-space patches are sufficient for powerful generative models.
• State-of-the-art reconstruction results were achieved.

Conclusions:

• The proposed HKGM effectively learns from limited k-space data.
• This approach significantly accelerates MRI acquisition and reconstruction.
• HKGM demonstrates potential for advancing clinical MRI diagnostics.