ENG Signal Classification via Parallel Spiking Neurons for Implantable Devices

Neural Decoding and Stimulation (ND&S) systems offer a promising alternative to conventional treatments for peripheral nerve injuries by decoding neural signals and delivering targeted stimulation via implantable devices. A primary challenge in ND&S development is the accurate classification of electroneurography (ENG) signals under strict constraints on computational resources, processing time and power constraints. To address this, we propose a Parallel Spiking Neural Network (PSNN) architecture based on a novel spiking neuron model called Parallel Spiking Neurons (PSNs), optimized for ENG classification. The model combines event-driven processing with low computational complexity, making it well-suited for implantable applications. Compared to the state-of-the-art ESCAPE-Net, the PSNN achieves higher test accuracy (87.21%±4.9%) and macro F1-score (84.98%±9.39%), while reducing the number of the required model parameters by 99.97%. These results underscore the effectiveness of PSNNs in achieving high classification accuracy with minimal computational overhead, aligning with the stringent requirements of ND&S systems.