Fully unsupervised online spike sorting based on an artificial spiking neural network

Spike sorting is a crucial step of neural data processing widely used in neuroscience and neuroprosthetics. However, current methods remain not fully automatic and require heavy computations making them not embeddable in implantable devices. To overcome these limitations, we propose a novel method based on an artificial spiking neural network designed to process neural data online and completely automatically. An input layer continuously encodes the data stream into artificial spike trains, which are then processed by two further layers to output artificial trains of spikes reproducing the real spiking activity present in the input signal. The proposed method can be adapted to process several channels simultaneously in the case of tetrode recordings. It outperforms two existing algorithms at low SNR and has the advantage to be compatible with neuromorphic computing and the perspective of being embedded in very low-power analog systems for future implantable devices serving neurorehabilitation applications.