Fast multi-contrast MRI using joint multiscale energy model

Motivation: The clinical acceptance of high-resolution 3D radial multicontrast acquisition schemes is impeded by lengthy scan times and the high computational demands of current reconstruction algorithms. Goal(s): We develop a joint multiscale energy model for the fast and high-fidelity 3D multicontrast MRI reconstruction from highly undersampled data. Approach: The energy model learns the joint probability density, enabling exploitation of the inter-contrast dependencies. The explicit energy enables a majorize-minimize algorithm with rapid convergence. Results: The algorithm enables the recovery of four 3D contrasts with 1mm isotropic spatial resolution from a 2.25-minute scan with 12 minutes of computation time, with higher quality than independent methods. Impact: The algorithm offers a 4x reduction in scan time and is over 10 times faster than current gridding-based non-linear fitting methods. J-MuSE is a good fit for general multicontrast applications by exploiting the correlations between contrasts, while enabling fast recovery.