Synthesized b0 for diffusion distortion correction (Synb0-DisCo)

Summary

Synb0-DisCo uses a structural MRI scan to synthesize an undistorted diffusion reference image. This enables susceptibility-distortion correction for datasets lacking reverse phase-encoded scans, improving anatomical alignment and reducing variation in diffusion modeling in the evaluated data.

Study overview

Many historical or abbreviated diffusion MRI protocols lack the reversed phase-encoding images needed for standard susceptibility correction. Synb0-DisCo addresses this gap by using deep learning to synthesize a non-diffusion-weighted image from a structural scan, providing an undistorted anatomical target for TOPUP-based processing. The evaluated pipeline improved agreement with structural geometry and reduced variation in diffusion-derived measurements, with performance comparable to paired phase-encoding references in the study. The method makes additional datasets usable, but the synthesized reference is a model prediction rather than an acquired opposite-encoding image. Its reliability therefore depends on synthesis quality and the similarity of new data to the evaluated conditions.

Main findings

• A structural image was used to synthesize an undistorted b0-like reference.
• Correction improved geometric correspondence with anatomical images and reduced diffusion-model variation.
• The approach enabled TOPUP-like correction without an acquired reversed phase-encoding scan.