A noise robust image reconstruction using slice aware cycle interpolator network for parallel imaging in MRI.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Signal Processing

Background:

• Reducing Magnetic Resonance Imaging (MRI) scan time is crucial for clinical applications.
• Undersampling k-space data using multiple receiver coils accelerates MRI acquisition.
• This acceleration reduces the number of sampled k-space lines, but can introduce artifacts.

Purpose of the Study:

• To develop a deep learning method for parallel MRI reconstruction.
• The method aims to reduce the number of auto-calibration signal (ACS) lines required.
• The objective is to achieve robust reconstruction in noisy environments.

Main Methods:

• A cycle interpolator network was developed to estimate missing k-space lines for each coil.
• Estimated lines were used to re-estimate sampled k-space lines, improving reconstruction.
• A slice-aware reconstruction technique was incorporated for noise robustness with fewer ACS lines.

Main Results:

• The slice-aware cycle interpolator network successfully reconstructed enhanced parallel MRI images.
• The method outperformed existing techniques like RAKI and GRAPPA in eliminating aliasing artifacts.
• The network demonstrated effective brain image reconstruction even under severe noise and limited ACS lines.

Conclusions:

• The slice-aware cycle interpolator network shows potential for improving parallel MRI reconstruction accuracy.
• It offers a viable solution for scenarios requiring fewer ACS lines in noisy environments.
• This deep learning approach can enhance clinical MRI efficiency and image quality.