Single-spiral MRE image reconstruction using zero-shot self-supervised deep learning towards real-time stiffness mapping

Motivation: MR elastography provides important clinical information but is time consuming. Single-spiral acquisitions can speed up data acquisition but require advanced image reconstruction schemes. State-of-the-art deep learning methods require high-quality training data which is often not available for MRE. Goal(s): Developing zero-shot self-supervised deep learning approach towards real-time MRE from single-spiral acquisitions. Approach: We developed a PyPulseq spiral MRE sequence with a self-supervised deep learning reconstruction for single-subject data. This approach leverages inherent similarities and redundancies within MRE scans and incorporates a GPU-accelerated k-MDEV inversion. Results: Precise elastograms from single spiral arm MRE aquistions with a PyPulseq MRE sequence. Impact: The proposed self-supervised reconstruction ensures accurate elastogram estimation from highly undersampled data, reducing scan time. These developments promise to further enhance MRE's impact and efficiency in clinical practice.