Deep learning-optimized bilayer metasurface based on robust magnetic dipole coupling for MRI enhancement at 70mT

Metasurfaces, periodic arrangements of tiny structures, enable precise manipulation of electromagnetic fields and show great potential for enhancing B-fields in magnetic resonance imaging (MRI). However, magnetic field enhancement at very-low fields remains challenging due to the weak coupling between localized magnetic dipoles in single-layer metasurfaces, which prevents the formation of coherent, collective behavior essential for substantial magnetic field amplification. In this work, we propose a bilayer metasurface composed of split-ring resonators (SRRs) that couples two eigenmodes, inducing band degeneracy and interlayer magnetic dipole coupling, thus achieving robustness and uniform magnetic field enhancement. Guided by conformal optics theory, we uncover the physical mechanism behind the improved magnetic response. A deep learning neural network is employed to optimize structural parameters, enabling rapid identification of band degeneracy states while reducing computational cost. High-Q resonance is achieved through the excitation of bound states in the continuum (BIC), and an equivalent circuit model is established to stabilize the resonant frequency at 2.9 MHz, ensuring reliable electromagnetic wave manipulation. Compared to a single receiving coil, integration of the SRR with a coil increases the B-field intensity in human head MRI images by approximately threefold at 70 mT. These results open a new avenue toward reconfigurable MRI metasurfaces with customized electromagnetic properties.