AliasNet: Alias artefact suppression network for accelerated phase-encode MRI.

Area of Science:

• Medical Imaging
• Computer Vision
• Signal Processing

Background:

• Compressed Sensing (CS) MRI reduces scan time but 1D undersampling creates structured aliasing artefacts.
• Existing 2D regularisation methods struggle with these direction-associated artefacts.
• Hardware constraints favor 1D Cartesian phase-encode undersampling in 2D CS-MRI.

Purpose of the Study:

• To develop novel techniques for improved sparse reconstruction in MRI.
• To address the limitations of 2D regularisation for 1D undersampling artefacts.
• To enhance image quality and reconstruction efficiency in accelerated MRI.

Main Methods:

• Developed two decoupling techniques for explicit 1D regularisation of phase-encode artefacts.
• Derived a combined 1D + 2D reconstruction technique leveraging spatial relationships.
• Integrated 1D AliasNet modules with existing 2D deep learning (DL) recovery methods.

Main Results:

• Demonstrated improved image quality on retrospectively undersampled brain and knee MRI data.
• AliasNet showed superior performance scaling compared to increasing 2D network size.
• The proposed 1D + 2D approach is compatible with existing 2D DL recovery techniques.

Conclusions:

• AliasNet effectively regularizes aliasing artefacts from phase-encode undersampling by tailoring network architecture.
• The 1D + 2D reconstruction approach enhances MRI image quality and reconstruction performance.
• This method offers a compatible and efficient solution for accelerated MRI acquisition.