Unsupervised deep learning model for correcting Nyquist ghosts of single-shot spatiotemporal encoding.

Area of Science:

• Magnetic Resonance Imaging (MRI)
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Nyquist ghosts are artifacts in single-shot spatiotemporal encoding (SPEN) MRI.
• These artifacts degrade image quality and diagnostic accuracy.
• Existing correction methods may have limitations in complex scenarios.

Purpose of the Study:

• To develop an unsupervised deep learning model for Nyquist ghost correction in SPEN MRI.
• To evaluate the model's performance in simulated and real MRI data.
• To address the need for robust artifact removal in accelerated MRI.

Main Methods:

• An unsupervised Residual Encoder and Restricted Subspace Mapping network (RERSM-net) was designed.
• The network generates phase-difference maps from even and odd SPEN images.
• A spin physical forward model and cycle-consistency loss were utilized for training.

Main Results:

• The RERSM-net successfully generated smooth phase-difference maps.
• Effective correction of Nyquist ghosts in single-shot SPEN was demonstrated.
• The proposed method outperformed state-of-the-art techniques in simulations and in vivo experiments.

Conclusions:

• The developed unsupervised deep learning method effectively corrects Nyquist ghosts in single-shot SPEN MRI.
• The RERSM-net offers a promising solution for improving MRI image quality.
• The model shows significant advantages validated by ablation studies.