ACE-Net: AutofoCus-Enhanced Convolutional Network for Field Imperfection Estimation with application to high b-value...

Area of Science:

• Magnetic Resonance Imaging (MRI)
• Medical Physics
• Computational Imaging

Background:

• Spatiotemporal magnetic field variations (B0 inhomogeneity and eddy currents) cause artifacts in rapid MRI sequences like spiral and EPI.
• These artifacts degrade image quality, particularly in diffusion MRI at high b-values.

Purpose of the Study:

• To develop an automated, data-driven method for estimating and correcting magnetic field imperfections in MRI.
• To improve the accuracy of diffusion MRI reconstructions by addressing B0 inhomogeneity and eddy currents.

Main Methods:

• A novel approach combining autofocus metrics with deep learning.
• Utilizing a compact basis representation for expected field imperfections.
• Application to single-shot spiral diffusion MRI at high b-values.

Main Results:

• Accurate estimation of B0 inhomogeneity and eddy currents was achieved.
• High-quality image reconstruction was obtained for spiral diffusion MRI.
• The method eliminated the need for additional external calibrations.

Conclusions:

• The developed data-driven approach effectively corrects magnetic field imperfections in diffusion MRI.
• This technique enhances image reconstruction quality for rapid imaging sequences.
• It offers a calibration-free solution for improving diffusion MRI accuracy.