Learning MRI Artifact Removal With Unpaired Data
This addresses a practical limitation in medical imaging by enabling artifact correction without paired data, though it is incremental as it builds on existing neural network approaches for artifact removal.
The paper tackled the problem of MRI artifact removal without requiring paired artifact-free and artifact-corrupted images, which are often scarce or unavailable, by developing a neural network that learns from unpaired data, resulting in effective artifact removal and anatomical detail retention across different image contrasts.
Retrospective artifact correction (RAC) improves image quality post acquisition and enhances image usability. Recent machine learning driven techniques for RAC are predominantly based on supervised learning and therefore practical utility can be limited as data with paired artifact-free and artifact-corrupted images are typically insufficient or even non-existent. Here we show that unwanted image artifacts can be disentangled and removed from an image via an RAC neural network learned with unpaired data. This implies that our method does not require matching artifact-corrupted data to be either collected via acquisition or generated via simulation. Experimental results demonstrate that our method is remarkably effective in removing artifacts and retaining anatomical details in images with different contrasts.