A neural network for online spike classification that improves decoding accuracy

Objective: Separating neural signals from noise can help to improve brain-computer interface performance and stability. However, "spike-sorting" suffers from a lack of a ground truth for classifying waveforms, and most algorithms for separating neural action potentials from noise are not suitable for use in real time and have shown mixed effects on decoding performance. Our goals were to automate the intuition of human spike-sorters to operate in real time with an easily tunable parameter governing the stringency with which spike waveforms are classified. Approach: We trained an artificial neural network with one hidden layer on neural waveforms that were hand-labeled as either spikes or noise. The network output was a likelihood metric for each waveform it classified, and we tuned the network9s stringency by varying the minimum likelihood value for a waveform to be considered a spike. Using the network9s labels to exclude noise waveforms, we decoded remembered target location from electrode arrays implanted in prefrontal cortex (PFC) during a memory-guided saccade task in two rhesus macaque monkeys. We assessed our network9s performance by measuring how it influenced decoding accuracy. Main results: The network classified waveforms in real time, and its classifications were qualitatively similar to those of a human spike-sorter. Compared to decoding with all waveforms (i.e. threshold crossings), we found that we could improve decoding performance, or at worst cause only a small change, if we removed waveforms with low spike likelihood values. Furthermore, decoding with our network9s classifications became more beneficial as time since the array was implanted increased. Significance: Our classifier serves as a feasible preprocessing step that could be applied to both offline neural data analyses and online decoding. Additionally, we demonstrated that using our network9s classifications for noise removal could improve decoding performance with little risk of harm.