Accelerated Electromagnetic Simulation of MRI RF Interactions with Graphene Microtransistor-Based Neural Probes for Electrophysiology-fMRI Integration

Implementing electrophysiological recordings within an MRI environment is challenging due to complex interactions between recording probes and MRI-generated fields, which can affect both safety and data quality. This study aims to develop and evaluate a hybrid electromagnetic (EM) simulation framework for efficient and accurate assessment of such interactions. Methods: A hybrid EM strategy integrating the Huygens’ Box (HB) method with sub-gridding was implemented in an FDTD solver (Sim4Life). RF coil models for mouse and rat head were simulated with and without intracortical (IC) and epicortical (EC) graphene-based micro-transistor arrays. Three-dimensional multi-layered probe models were reconstructed from two-dimensional layouts, and transmit field ($B_{1}^{+}$), electric field ($E$), and specific absorption rate (SAR) distributions were evaluated. Performance was benchmarked against conventional full-wave multi-port (MP) simulations using Bland-Altman analysis and voxel-wise percentage differences. Results: HB simulations reduced computational time by approximately 70-80%, while preserving spatial patterns of $|B_{1}^{+}|$, $|E|$, and SAR, including transmit-field symmetry and localized high-field regions. Deviations from MP were minimal for $|B_{1}^{+}|$ (median $Δ$% 0.02-0.07% in mice, -3.7% to -1.7% in rats) and modest for $|E|$ and SAR, with absolute SAR values remaining well below human safety limits. Graphene-based arrays produced negligible effects on RF transmission and SAR deposition. Conclusion: The HB approach enables computationally efficient, high-resolution evaluation of EM interactions involving microscopic probes in MRI environments, supporting simulations that are otherwise impractical with full-wave MP modeling.

💡 Research Summary

The paper addresses a critical bottleneck in the simultaneous use of electrophysiology and functional magnetic resonance imaging (fMRI) in pre‑clinical studies: the computationally intensive electromagnetic (EM) modeling required to assess safety and performance of ultra‑small neural probes within the high‑frequency RF field of an MRI scanner. The authors focus on graphene‑solution‑gated field‑effect transistor (gSGFET) arrays, which combine sub‑micron thickness, minimal metal content, and high‑fidelity neural recording capabilities. Because these devices are orders of magnitude smaller than the RF coil and the animal head, conventional full‑wave finite‑difference time‑domain (FDTD) simulations become impractical due to excessive memory consumption and runtime, especially when high spatial resolution is needed to resolve the probe geometry and the resulting electric field (E) and specific absorption rate (SAR) distributions.

To overcome this, the authors develop a hybrid EM simulation framework that couples the Huygens’ Box (HB) method with a sub‑gridding technique, implemented in the commercial FDTD solver Sim4Life. The HB approach treats the RF coil as a source that generates a “virtual” Huygens’ surface enclosing a region of interest (the probe and surrounding tissue). By exciting the coil once in circularly polarized mode, the electromagnetic fields on this surface are recorded and later re‑used as boundary conditions for any configuration placed inside the box, eliminating the need to re‑solve the entire coil‑head system for each probe layout. Sub‑gridding then refines the mesh only within the box, allowing cell sizes on the order of 50 µm around the probe while keeping a coarser grid elsewhere. This combination yields a dramatic reduction in computational load while preserving the high‑resolution detail required for accurate SAR assessment.

The framework is validated against conventional multi‑port (MP) FDTD simulations, which treat each coil port separately and solve the full geometry each time. Simulations are performed on anatomically realistic mouse (B6C3F1) and rat (Sprague‑Dawley) models, each containing 68–77 tissue types, with a standard 8‑rung high‑pass birdcage coil (72 mm diameter). Two probe configurations are examined: an intracortical (IC) array inserted into brain tissue and an epicortical (EC) array placed on the cortical surface. The gSGFET arrays consist of up to seven material layers (polyimide, Ti/Ni/Au, graphene monolayer, Ni/Au, SU‑8) with total thicknesses of ~10 µm. Graphene’s conductivity is modeled using the Drude formulation.

Quantitative comparison focuses on three key EM quantities: the transmit field B1⁺, the electric field magnitude |E|, and the SAR (both mass‑averaged and 0.1 g spatially averaged). Voxel‑wise Bland‑Altman analyses and percentage‑difference maps reveal that HB reproduces B1⁺ with median deviations of 0.02–0.07 % in mice and –3.7 % to –1.7 % in rats—well within acceptable limits for coil tuning and image uniformity. |E| and SAR show modest differences (typically 5–10 %); however, absolute SAR values remain far below IEC safety thresholds (peak SAR <0.1 W/kg). Importantly, the presence of the graphene arrays induces negligible perturbations in B1⁺ and SAR, confirming their intrinsic MRI compatibility compared with conventional metallic electrodes that often cause significant image distortion and heating.

From a performance standpoint, the HB‑sub‑gridding workflow reduces simulation time by roughly 70–80 % (e.g., mouse simulations drop from ~12 h to ~2.5 h, rat simulations from ~18 h to ~3.5 h) on a single NVIDIA RTX 4090 GPU, and memory usage is similarly curtailed, enabling high‑resolution runs that would otherwise exceed GPU capacity. The authors also conduct uncertainty analyses varying grid resolution and box size, demonstrating robustness of the HB results across reasonable parameter ranges.

The study’s implications are multifold. First, it provides a practical, high‑fidelity tool for pre‑clinical safety assessment of ultra‑small neural implants, allowing designers to iterate probe geometries and materials without prohibitive computational cost. Second, the methodology is readily extensible to human‑scale models and higher field strengths (e.g., 7 T), facilitating translation of graphene‑based neural interfaces to clinical neuroimaging environments. Third, the negligible RF interaction of gSGFET arrays supports their use in simultaneous electrophysiology‑fMRI experiments, potentially unlocking new insights into neurovascular coupling and disease mechanisms. Future work could integrate experimental SAR measurements, explore other coil designs, and implement real‑time GPU‑accelerated HB simulations for adaptive safety monitoring during in‑vivo studies. Overall, the hybrid Huygens‑Box with sub‑gridding represents a significant advance in MRI‑compatible neural probe modeling, balancing accuracy, speed, and scalability.