Closing the Access Gap: A Sample-Efficient, Digital-Twin-Guided Framework for Scaling Epidural Stimulation Therapy to the Global Spinal Cord Injury Population

Epidural Electrical Stimulation (EES) can help some people with chronic Spinal Cord Injury (SCI) stand and step again, but specialists need days to weeks of repeated testing to find the right settings for each patient. To address this, we propose a two-stage framework in which a digital twin, a Gaussian process trained on 600 observations from 30 simulated patients, warm-starts Bayesian optimization for a new patient. We simulate patients using a Matsuoka half-center oscillator model of the spinal circuit. In simulation, the twin reached a near-best setting in a median of 1 trial, compared with 6 for random search and 4 for tuned Bayesian optimization (90 runs per method). In our additional analyses, we found that the benefit comes mainly from the pretrained prior and not from tailoring to injury severity, that it is robust to parameter changes but weaker when the response landscape changes, and that the model’s gait rhythm is 16.2 percent slower than that of ten healthy adults from a public database. This is a simulation study, and we propose calibration with real EES patient data and a small clinician pilot as the next step.