FD-Net: An unsupervised deep forward-distortion model for susceptibility artifact correction in EPI

Area of Science:

• Medical Imaging
• Deep Learning
• Neuroimaging

Background:

• Susceptibility artifacts in echo planar imaging (EPI) degrade image quality.
• Existing correction methods often lack computational efficiency or consistency with acquired data.

Purpose of the Study:

• To develop an unsupervised deep learning method for fast and effective correction of susceptibility artifacts in reversed phase-encode (PE) EPI image pairs.
• Introduce the Forward-Distortion Network (FD-Net) for improved anatomical accuracy in EPI.

Main Methods:

• FD-Net predicts susceptibility-induced displacement fields and anatomically correct images.
• Enforces consistency between corrected image forward-distortions and acquired reversed-PE EPI data.
• Utilizes a multiresolution architecture for enhanced performance.

Main Results:

• FD-Net achieves competitive image quality compared to the gold-standard TOPUP method.
• Demonstrates significant improvements in computational efficiency over existing methods.
• Outperforms recent unsupervised methods in both image and field quality.

Conclusions:

• The unsupervised FD-Net method provides fast, high-fidelity correction of EPI susceptibility artifacts.
• Maintains consistency with measured data, holding promise for enhanced anatomical accuracy in EPI.
• Represents a significant advancement in artifact correction for EPI.