Coil sketching for fast and efficient 4D lung MRI reconstruction

Motivation: Non-Cartesian 4D lung MRI is especially challenging due to large matrix sizes and cardio-respiratory motion, leading to high memory allocation and long image reconstruction times. To be clinically feasible, there is a need for faster and more computationally-efficient respiratory-resolved reconstructions. Goal(s): Extend highly-efficient coil sketching reconstruction to 4D MRI, and solve large respiratory-resolved low rank reconstruction problems within clinically feasible timeframes and within <24GB GPU memory. Approach: We developed and applied 4D coil sketching to low-rank and motion-compensated low-rank reconstruction for free-breathing Stack-of-Spirals pulmonary imaging. Results: Coil sketching provided >2-fold faster and 3-fold more memory-efficient reconstructions compared to conventional methods. Impact: 4D coil sketching enables <5 minute reconstructions for high-resolution respiratory-resolved lung imaging, without the need for server-grade GPUs. This approach is ideal for free-breathing lung MRI applications, like phase-resolved functional-lung (PREFUL) MRI, that demand extensive oversampling and high temporal resolution.