Fig 1. Overall pipeline.

The goal is to generate an undistorted b0 from a single blip (distorted) b0 and an anatomical T1 image through a deep learning approach. The undistorted image can then be concatenated with the distorted b0 and run through FSL’s topup using a simulated infinite PE-bandwidth. This final correction can be used with FSL’s eddy (or another eddy current modeling tool) to provide a full correction for diffusion data given only a single phase encoding. Note that the proposed algorithm does not seek to model/correct eddy current effects.