Accelerating Multi-parametric MRI Sequences Using Learned Acquisition and Reconstruction Optimization (LARO): A Retrospective Study

Motivation: Multi-parametric pulse sequences provide a comprehensive profile of MS lesions, but significantly prolonged scan time limits widespread use. Goal(s): Use simulations to demonstrate potential of accelerating multi-parametric sequences, while maintaining consistency of parameter maps and ROI segmentations. Approach: Learned Acquisition and Reconstruction Optimization (LARO) was applied retrospectively to a series of multi-contrast and quantitative MRI sequences with simulated fully sampled k-space data (gold standard) for acceleration. Quantitative multi-parametric values for relaxometry and susceptometry were compared before and after. Results: LARO accelerated sequences for R1, R2, R2*, QSM, and susceptibility source separation (acceleration factor=8) achieved high lesion-wise consistency (R-squared>0.94) compared to fully sampled maps. Impact: Our study demonstrates LARO's potential to accelerate time-consuming multi-parametric and multi-contrast MRI sequences, paving the way for comprehensive lesion profiling within practical scan times. LARO acceleration ensured consistent parametric maps (except for R2) and ROI segmentations.