Deep learning corrects artifacts in RASER MRI profiles.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Quantum Optics

Background:

• Radiowave amplification by stimulated emission of radiation (RASER) MRI is a novel technique offering potential for higher resolution and background-free imaging.
• Nonlinear effects in RASER MRI can cause significant distortions, limiting its practical application.
• Current limitations hinder the widespread adoption of RASER MRI despite its theoretical advantages.

Purpose of the Study:

• To investigate the efficacy of deep learning (DL) in mitigating artifacts in RASER MRI.
• To develop and validate a DL pipeline for correcting nonlinear distortions in RASER images.
• To demonstrate the generalization capability of the DL model from synthetic to experimental data.

Main Methods:

• A two-step deep learning (DL) pipeline was developed and trained.
• The DL pipeline utilized purely synthetic data generated from a theoretical RASER MRI model.
• A convolutional neural network processed 1D RASER projections, and a U-net processed 2D random images.

Main Results:

• The DL pipeline successfully reduced heavy distortions caused by nonlinear effects in RASER MRI.
• The trained DL model demonstrated effective generalization from synthetic to experimental RASER MRI data.
• Artifact reduction significantly improved the usability of RASER MRI images.

Conclusions:

• Deep learning provides a powerful solution for correcting artifacts in RASER MRI.
• The developed DL pipeline enhances the practical utility of RASER MRI technology.
• This work paves the way for more reliable and higher-quality RASER MRI applications.