Local Versus Global Low-Rank Promotion in Dynamic MRI Series Reconstruction

Joshua Trzasko¹, Armando Manduca¹

Recent works have suggested that dynamic MRI series reconstructions can be significantly improved by promoting low-rank structure in the estimated image series. However, when there exists a significant discrepancy between the spatial and temporal dimensions of the image series, low-rank approximations begin to lose their efficacy, resulting in either inadequate noise removal or temporal blurring. In this work, we present a generalization of the low-rank recovery paradigm, which we call Locally Low Rank (LLR) image recovery, that promotes low-rank structure locally rather than globally. As demonstrated, this migration can improve both the efficacy of noise removal and temporal signal fidelity.