Autocalibrated loraks for fast constrained MRI reconstruction

Low-rank modeling of local k-space neighborhoods (LO-RAKS) is a recent novel framework for reconstructing MRI images from sparsely-sampled and/or noisy data. Previously-proposed LORAKS-based reconstruction approaches relied on low-rank matrix recovery methods, which were powerful but computationally expensive. In this work, we demonstrate that substantial computational accelerations can be achieved if the nullspaces associated with the low-rank LORAKS matrices are pre-estimated from autocalibration data. In addition to improving computation speed, we also show that auto-calibrated LORAKS can have substantial advantages over previous autocalibrated parallel imaging methods.