Towards online 3D MR-STAT reconstructions

Motivation: MR-STAT is a transient-state qMRI technique that can deliver T1, T2 and proton density maps from a single short scan, even when Cartesian trajectories are used. The non-linear model-based MR-STAT reconstruction procedure is challenging and current state-of-the-art reconstructions take multiple hours for a 3D dataset. Goal(s): In this work we aim to accelerate the MR-STAT reconstructions to make the technique more valuable for clinical practice. Approach: A highly-optimized CUDA EPG simulator was developed and an optimization for signal computations for Cartesian trajectories has been discovered. Results: High-resolution 3D knee and brain MR-STAT datasets can now be reconstructed in 12.5 and 25 minutes, respectively. Impact: The tools are part of the COMPAS Toolkit that is available online. These tools are relevant to other researchers or educators that require high-performance Bloch/MRI simulations.