Neuromorphic Neuromodulation: A Low-Power Edge-Training Framework for the Future of Personalized and Closed-Loop Neurostimulation

Epilepsy affects approximately 1% of the global population, with 30-40% of cases resistant to conventional pharmacological treatments. Current neurostimulation offers a different perspective by sending electrical stimulation to the brain. However, these devices rely on cloud-dependent AI, which introduces significant challenges, including high false positive rates, latency issues, and privacy concerns. Here we present a neuromorphic framework for real-time seizure detection and prediction, directly implemented on a neuromorphic SOC with on-chip learning capabilities. By leveraging spiking neural networks and few-shot edge learning, our system enables continual, patient-specific adaptation without requiring data transmission or cloud connectivity. For a detection framework, we pre-train a model using the TUH dataset and deploy it on the BrainChip Akida neuromorphic processor, enabling few-shot personalization on long-term recordings from the EPILEPSIAE dataset. For seizure prediction, we demonstrate robust performance on both scalp and intracranial EEG from Children’s Hospital Boston (CHB-MIT) and Freiburg (FB) datasets using a leave-one-seizure-out approach for training at the edge. For all the datasets, we achieved commendable and superior performance to state-of-the-art neural networks in metrics as AUROC, Sensitivity, and False Positive Rates (FPR), with a memory footprint drastically reduced through quantization and energy consumption orders of magnitude below conventional processors. This work represents a foundational advance toward autonomous, closed-loop neurostimulation systems capable of learning and adapting at the edge for newer generation of precision neurotechnology.