Simultaneous superresolution reconstruction and distortion correction for single-shot EPI DWI using deep learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Single-shot echo-planar imaging (SS-EPI) is a standard clinical technique for diffusion-weighted imaging (DWI).
• SS-EPI is prone to distortions and lower resolution, which can affect diagnostic accuracy.
• Point-spread-function (PSF)-encoded EPI offers high-resolution, distortion-free images but is less common clinically.

Purpose of the Study:

• To develop an end-to-end deep learning method for simultaneous resolution enhancement and distortion correction of SS-EPI DWI.
• To improve the quality and quantitative accuracy of DWI images derived from lower-resolution SS-EPI data.

Main Methods:

• A novel deep learning network architecture was designed, incorporating a dense net with gradient map guidance and a multilevel fusion block as the generator.
• A modified generative adversarial network (GAN) structure was employed to suppress oversmoothing effects.
• A fractional anisotropy (FA) loss function was introduced to leverage diffusion anisotropy information.

Main Results:

• The proposed method successfully generated high-resolution, distortion-corrected DWI images from low-resolution SS-EPI data.
• Restored structural details were observed in the enhanced images, benefiting from high-resolution anatomical references.
• The deep learning approach demonstrated improved quantitative accuracy of diffusion metrics compared to existing methods.

Conclusions:

• A deep learning-based method can effectively enhance SS-EPI DWI by simultaneously increasing resolution and correcting distortions.
• The developed technique shows potential for improving both the visual quality and the reliability of diffusion metrics in clinical DWI.
• This approach offers a promising solution for overcoming the limitations of SS-EPI in DWI applications.