Alignment-Guided Forward-Distortion Model for Deep Unsupervised Correction of Susceptibility Artifacts in EPI

Area of Science:

• Medical Imaging
• Deep Learning
• Neuroimaging

Background:

• Susceptibility-induced distortions in echo planar imaging (EPI) are a major challenge in magnetic resonance imaging (MRI).
• Traditional artifact correction methods are computationally intensive and not suitable for clinical use.
• Deep learning offers potential for efficient EPI artifact correction.

Purpose of the Study:

• To develop a deep learning model for rapid and accurate EPI susceptibility artifact correction.
• To address the challenge of subject motion during artifact correction.
• To improve the clinical feasibility of EPI artifact correction.

Main Methods:

• Proposed an alignment-guided forward distortion network (agFD-Net) for unsupervised, physics-driven training.
• Integrated a pre-trained alignment network (AlignNet) to handle subject motion.
• Evaluated agFD-Net on an experimental NIH dataset with realistic motion levels.

Main Results:

• agFD-Net achieved rapid and high-fidelity correction of susceptibility artifacts.
• The model successfully accounted for subject motion between reversed-phase encoding acquisitions.
• Demonstrated over two orders of magnitude speed-up in computational efficiency compared to classical methods.

Conclusions:

• agFD-Net offers a significant advancement in EPI artifact correction efficiency and accuracy.
• The model's ability to handle subject motion makes it highly promising for clinical MRI applications.
• This deep learning approach overcomes the limitations of traditional methods, enabling practical clinical use.