To develop a dual-domain multipath self-supervised diffusion model for accelerated MRI reconstruction that improves accuracy and efficiency without relying on fully sampled data.
Proposed dual-domain multipath self-supervised diffusion model (DMSM) for MRI reconstruction.
Integrated lightweight hybrid attention network (LHAN) for better performance.
Evaluated on two human MRI datasets comparing to supervised and self-supervised models.
DMSM outperformed several baselines in preserving fine anatomical structures and suppressing artifacts.
Generated uncertainty maps that correlated well with reconstruction errors, aiding clinical interpretation.