RESEARCH · RESEARCH · #2315
MoSAIK: Transformer-based motion estimation and motion-aware reconstruction from multi-coil TSE k-space
MoSAIK is a retrospective method that estimates in-plane rigid-body motion directly from standard multi-coil TSE k-space by detecting k-space self-consistency violations and using a convolutional embedding plus Transformer encoder to predict per-phase-encoding-line rotation and 2D translations; predicted motion parameters are fed into a model-based, motion-aware reconstruction rather than synthesizing images. Trained on motion-free clinical data with simulated motion and evaluated on held-out subjects, the method reports low estimation MSE (rotation <0.2 deg^2, translation <0.2 pixels^2), high correlations (~0.96 rotation, ~0.94 translation), and per-slice information gains in vivo (>75% of slices) with reduced visible artefacts.
KEY POINTS
- MoSAIK is a retrospective method that estimates in-plane rigid-body motion directly from standard multi-coil TSE k-space by detecting k-space self-consistency violations and using a convolutional embedding plus Transformer encoder to predict per-phase-encoding-line rotation and 2D translations; predicted motion parameters are fed into a model-based, motion-aware reconstruction rather than synthesizing images.
- Trained on motion-free clinical data with simulated motion and evaluated on held-out subjects, the method reports low estimation MSE (rotation <0.2 deg^2, translation <0.2 pixels^2), high correlations (~0.96 rotation, ~0.94 translation), and per-slice information gains in vivo (>75% of slices) with reduced visible artefacts.
- This enables fast retrospective motion correction for standard clinical TSE brain MRI without navigators or extra acquisitions, potentially reducing repeat scans and improving diagnostic confidence.
WHY IT MATTERS
This enables fast retrospective motion correction for standard clinical TSE brain MRI without navigators or extra acquisitions, potentially reducing repeat scans and improving diagnostic confidence.