Optimization of Electrode and Stimulus Parameters for Translatable Preclinical Models of Dorsal Root Ganglion Stimulation

Introduction: Preclinical models are indispensable for defining neuromodulation mechanisms, but their interpretive value hinges on reproducing direct neural effects1. These effects are governed by spatiotemporal potential fields and gradients set by electrode design, stimulus dosing, and biophysical constraints of the model2. Simple scaling across species/devices rarely yields equivalent neural recruitment and can mislead mechanistic inference3. A strategy to address scaling problems is to match an excitability metric of neural recruitment within a region of interest across settings. We tested this approach with dorsal root ganglion stimulation (DRGS). Using parametric finite-element method (FEM) models coupled with Bayesian optimization (BO), we evaluated multiple excitability measures to guide the design of surface electrode arrays (SEAs) and stimulus scaling in rodents that approximate the direct neural effects of clinical DRGS.

Methods: We built three DRGS FEM models: a clinical human-scale model with a percutaneous array (reference), an in vivo rat preparation with a two-contact SEA, and an ex vivo mouse preparation with a two-contact SEA. SEA designs were parameterized by contact width/length and spacing. Using BO, we sequentially proposed SEA parameters; for each parameter set, we solved the corresponding FEM model and computed an excitability measure in the region of interest (i.e., dorsal root ganglion). Designs were scored by the Wasserstein distance between the spatial distribution of that measure and the clinical reference; lower values indicated a closer distributional match. We ran separate optimizations for three measures: electric field magnitude1, activating function (second spatial derivative of the extracellular potential)4, and mirror estimate (negative extracellular potential)5. Finally, a second BO stage optimized pulse width by simulating Aβ-fibers (putative target6) across multiple pulse widths and selecting the pulse width that minimized differences between clinical and preclinical neural recruitment.

Results: Neuron simulations indicated that optimization-derived electrodes achieved threshold distributions more like the clinical reference than arrays scaled simply based on size; pulse-width tuning further reduced mismatch. The current scaling required to minimize differences between the preclinical and clinical models varied with targeted recruitment level across electrodes, underscoring the need to specify target populations when translating stimulation paradigms.

Conclusion: We developed an approach to translate clinical DRGS to preclinical models by determining electrode and stimulus parameters via computational modeling and BO. This approach could improve the mechanistic fidelity of preclinical models and can be generalized to other neuromodulation therapies where translation hinges on recapitulating field-driven neural recruitment. Experimental fabrication and validation are forthcoming.