Near-Real-Time Multi-Parametric Quantitative MRI using Parallel Non-Cartesian 6D Spatial-Temporal Dictionary Learning...

Multi-parametric quantitative MRI (qMRI) enables precise targeting during image-guided interventions such as deep brain stimulation. To address the demand for higher temporal resolution in multi-parametric qMRI of the brain, we propose an online pipeline for near-real-time quantitative MRI.

The acquisition utilizes an alternating dual-flip-angle blipped multi-gradient-echo sequence with a stack-of-stars (SoS) trajectory, enabling efficient sampling while facilitating T1 mapping. K-space data are streamed to the Gadgetron framework for preprocessing and parallel non-Cartesian reconstruction. We extend the original 4D spatial-temporal dictionary learning neural networks (stDLNN), which exploit correlations across 2D space, time, and real/imaginary channels in dynamic MRI, to a 6D formulation for dual-flip-angle multi-echo complex-valued 3D reconstruction. The proposed 6D-stDLNN jointly exploits correlations across 3D space, echo, flip angle, and real/imaginary channels during reconstruction. Subsequently, quantitative calculations are efficiently performed on the GPU.

Prospective results from volunteer studies show good consistency (intraclass correlation coefficients > 0.9) between 3D multi-parametric maps from accelerated sampling and those from fully sampled references. In the slab-selective scan, the temporal resolution achieved was 8.1 s per frame, and the total algorithm runtime was below 5 s per frame.

We present a near-real-time imaging pipeline that simultaneously provides 3D quantitative susceptibility mapping (QSM), R2* mapping, and T1 mapping, along with susceptibility-weighted imaging (SWI) and maximum intensity projection (MaxIP) images.

This work demonstrates the feasibility of immediate multi-parametric quantitative assessment in the brain, indicating potential utility in time-critical applications and future interventional studies.