NeuroRehab-Sim: A Multi-Scale Simulation Testbed for Closed-Loop Stroke-Rehabilitation Brain-Computer Interfaces
Year: 2026
Type: conference-paper
Abstract: Stroke is the leading cause of long-term motor disability worldwide, and brain-computer interfaces (BCIs) have emerged as a promising tool for neurorehabilitation — using decoded neural signals to drive functional electrical stimulation, robotic orthoses, or visual feedback that promotes cortical plasticity and motor recovery. Yet the field faces a fundamental bottleneck: developing and validating closed-loop BCI systems for stroke rehabilitation requires iterating on stimulation policies, decoder designs, and neural interface configurations that are expensive, slow, and ethically constrained to test in human patients. There is no public invasive (ECoG or sEEG) recording from stroke patients, and the non-invasive (EEG) stroke datasets that do exist are small and task-limited. NeuroRehab-Sim is a proposed open-source simulation testbed that addresses this bottleneck. It is a computational platform that models the post-stroke brain, generates realistic invasive and non-invasive neural signals from the same underlying brain state, augments scarce public data with mechanistically-grounded synthetic signals, and trains reinforcement-learning (RL) stimulation controllers in a closed-loop BCI architecture — all in silico, before any human data is needed. The testbed integrates four disciplines: · Computational brain modeling: a multi-scale model of post-stroke motor cortex — neural mass models for whole-cortex dynamics coupled with spiking neural networks in motor areas that implement synaptic plasticity (STDP, homeostatic regulation) and produce emergent recovery dynamics. · Neural systems engineering: a forward model that generates realistic ECoG and EEG sensor signals from the same brain state via biophysical lead-field models, plus a modular closed-loop BCI architecture. · Artificial intelligence: (1) a hybrid generative augmentation pipeline (mechanistic skeleton + learned refiner) that synthesizes realistic post-stroke neural signals, and (2) a model-based RL stimulation controller with multi-channel actions (peripheral FES + cortical tDCS) and a multi-objective reward. · Neurosurgery: the testbed models the surgical context — electrode placement geometry, lesion parameters tied to neuroanatomy, and stimulation targets informed by surgical accessibility. The deliverable is a documented open-source platform (dual compute: laptop tutorials + cluster research) plus a three-phase validation study demonstrating the testbed reproduces known stroke-BCI findings: (1) the bistable recovery dynamics and learned-nonuse threshold from Norman, Wolpaw & Reinkensmeyer [1, 2], (2) signal realism validated by decoder transfer to public stroke EEG datasets, and (3) closed-loop efficacy of the RL stimulation policy. The project uses public datasets only and is scoped as a methods/infrastructure contribution serving the whole field.
Author: Amir Aghili, Amir, A
Language: English
Subfield: Cognitive Neuroscience
Field: Neuroscience
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7166348351