A generalized activating function for rapid and accurate neural response prediction in spinal cord stimulation

We present the Generalized Activating Function (GAF), a biophysics-grounded framework that accelerates neural activation predictions by up to three orders of magnitude relative to NEURON simulations of multi-compartment cable models while maintaining high predictive accuracy (R2 = 0.99). By overcoming computational bottlenecks in patient-specific modeling, the GAF enables practical translation of computational optimization into neuromodulation therapies such as spinal cord stimulation (SCS). We verify the GAF’s physiological validity by demonstrating near-perfect agreement with multi-compartment simulations for activation thresholds and spike initiation location/timing across multiple pulse widths in an anatomically realistic spine model.

We validate clinical relevance by reproducing a published SCS optimization study for lower-limb motor recovery, matching NEURON-based prediction accuracy while reducing computation time from hours to seconds. Finally, we exploit the GAF’s speed to explore previously intractable parameter spaces: Pareto optimization over pulse waveforms uncovers solutions with approximately seven-fold increased energy efficiency at matched activation, while optimization of multielectrode stimulation (16 currents) increases the functional selectivity index for right hip flexion from 52% (clinical benchmark) to 82% in under one minute. By eliminating computational bottlenecks and enabling gains in efficiency and selectivity, the GAF transforms model-driven optimization into a clinically viable tool for patient-specific neuromodulation.

This study was supported by the Personalized Health and Related Technologies (PHRT) 2022-279 and EuroStars E!835 (OptiStim) grants. A.A., V.G., and A.R. were funded by the Federal Ministry of Research, Technology and Space of Germany (BMFTR), project no. 01ZZ2016. Furthermore, J.G.O. was supported by the La Caixa Fellowship for Postgraduate studies in Europe throughout their master studies, during which this study was initiated.

These authors contributed equally: Taylor H. Newton, Javier García Ordóñez. Foundation for Research on Information Technologies in Society (IT’IS), Zürich, Switzerland Taylor H.

Newton, Javier García Ordóñez, Niels Kuster & Esra Neufeld Department of Information Technology and Electrical Engineering, Swiss Federal Institute of Technology (ETHZ), Zürich, Switzerland ZMT Zurich MedTech AG, Zürich, Switzerland Department of Medical Informatics, Biometry and Epidemiology, Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany Abdallah Alashqar, Vincent Gemar & Andreas Rowald Department of Artificial Intelligence in Biomedical Engineering, Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany NeuroRestore, Swiss Federal Institute of Technology (EPFL), Lausanne University Hospital (CHUV) and University of Lausanne (UNIL), Lausanne, Switzerland Neuro-X Institute, École Polytechnique Fédérale de Lausanne (EPFL), Geneva, Switzerland Department of Clinical Neuroscience, Lausanne University Hospital (CHUV) and University of Lausanne (UNIL), Lausanne, Switzerland Department of Neurosurgery, Lausanne University Hospital (CHUV) and University of Lausanne (UNIL), Lausanne, Switzerland Faculty of Biology and Medicine, University of Lausanne, Lausanne, Switzerland Correspondence to Taylor H. Newton or Javier García Ordóñez. The authors declare the following competing interests: E.N. and N.K. are shareholders of, and J.G.O. is employed by, ZMT Zurich MedTech AG, which produces the Sim4Life software for computational life sciences.

A.R., J.B., and G.C. hold various patents in relation to spinal cord stimulation therapies. J.B. and G.C. are consultants and shareholders of ONWARD Medical, a company with direct relationships with the present work. The remaining authors declare no competing interests.