Computationally Efficient Unsupervised Online Spatial Spike Sorting Algorithm

Online spike sorting provides on-the-fly single-neuron spiking activity that can be critical to real-time and implantable closed-loop neuromodulation devices, as well as advancing neuroscience. Neural recording electrodes can be manufactured with various configurations and sensing contact densities. Offline algorithms have improved rapidly by adapting to progressive electrode manufacturing technology. This paper proposes an unsupervised online spike-sorting algorithm that utilizes the tetrode’s 4-contact information for improved performance. This was achieved by incorporating the unique spatial signatures of the observed firing neurons extracted from the multi-contact electrode. The algorithm was evaluated on a realistic simulated neural spiking activity to assess the performance for varying numbers of ground-truth neurons and SNR values. For comparison, we extended Osort, the online sorting algorithm, to process tetrodes by concatenating the 4-contact waveforms. The results show improvements in accuracy and neuron yield, and a reduction in computational complexity, which is feasible for implantable devices. As expected, comparisons with offline methods show noticeable performance degradation; nevertheless, this work narrows the gap between online and offline tradeoffs, particularly for interactive neuroscience experiments.