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.