Learning MRI artefact removal with unpaired data

Retrospective artifact correction (RAC) improves image quality post acquisition and enhances usability. Recent machine learning driven techniques for RAC are predominantly based on supervised therefore practical utility can be limited as data with paired artifact-free artifact-corrupted images typically insufficient or even non-existent. Here we show that unwanted artifacts disentangled removed from an via neural network learned unpaired data. This implies our method does not require matching to either collected generated simulation. Experimental results demonstrate is remarkably effective in removing retaining anatomical details different contrasts.