Intelligent Design of Metamaterials for Magnetic Resonance Imaging: A Review

Electromagnetic metamaterials enable precise control of the magnetic resonance imaging (MRI) radiofrequency (RF) near-field via subwavelength architectures, thereby improving the receive RF-field (B₁^-) distribution and constraining the specific absorption rate (SAR). However, metamaterial microstructure design remains a high-dimensional, nonlinear, and computationally intensive inverse problem. As a result, conventional trial-and-error workflows based on electromagnetic simulations are often limited by prohibitive computational cost and low throughput. To address these challenges, this review systematically summarizes intelligent design paradigms for MRI metamaterials, covering resonant metallic metamaterials, high-permittivity materials, flexible conformal structures, and tunable or reconfigurable metasurfaces. The methods discussed include data-driven surrogate modeling, inverse and generative design, reinforcement-learning-based tuning, and physics-informed strategies that incorporate Maxwell-equation constraints to improve data efficiency and physical consistency, particularly in limited-data and strongly coupled loading scenarios. By connecting metamaterial physical responses, data-driven and physics-informed design strategies, and MRI-specific engineering constraints, this review clarifies how intelligent design can support the practical optimization of MRI metamaterials. Finally, engineering case studies, including signal-to-noise ratio (SNR) enhancement, RF-field homogenization, and conformal structure adaptation, are presented to illustrate the practical value of these paradigms. Future directions include the joint advancement of patient-specific high-fidelity physics modeling and transferable pretrained electromagnetic surrogate models that generalize across tasks.