Structure-Preserved Graph Embedding for Improved MRF Reconstruction

We introduce a novel framework based on structure-preserved graph embedding for improved magnetic resonance finger-printing (MRF) reconstruction. Our work first uncovers that the parameters naturally form a low-dimensional manifold representation of the high-dimensional tissue fingerprints. We then model the acquired MRF data as graph data nodes and construct weighted Homogeneous graphs to preserve both the local and global intrinsic structure of the high-dimensional MRF data. Unlike data-prior-driven methods, our approach offers a fresh perspective on solving the under-determined MRF reconstruction problem. Importantly, many operations of our framework occur in low-dimensional space, substantially reducing computational complexity. Preliminary numerical experiments show that the proposed method can reconstruct high-quality MRF data and multiple parameter maps within significantly reduced computational time.