Motivation: Current sampling-optimization methods, such as AutoSamp, jointly train sampling-patterns with deep-learning(DL) models for accelerated MRI reconstructions, but suffer from complexity and lengthy protocol-specific optimization. Goal(s): Design simpler sampling patterns that deliver comparable performance that can be created in real-time on a scan-by-scan basis. Approach: By heuristically extracting insights from extensive AutoSamp results, we parameterized Voronoi-cell-area as a function of kr, and construct a variable-density radial hexagonal sampling parameterized for acceleration rates and coil configurations. Results: Our method achieves reconstruction quality approaching AutoSamp, without case-specific training. We also derive heuristic relationships between sampling parameters, acceleration rates, and coil configurations. Impact: Our method simplifies the complexity and drastically improve the speed of sampling pattern generation using only a few parameters. This eliminates extensive retraining, offering a practical alternative for optimizing MRI acquisition across different configurations and use cases.
Bae et al. (Tue,) studied this question.