Towards online spike sorting for high-density neural probes using discriminative template matching with suppression of interfering spikes

Objective. The process of grouping neuronal spikes in an extracellular recording according to their neuronal sources, is generally referred to as spike sorting. Currently, the use of spike sorting is mainly limited to an offline usage, where spikes are sorted after the data acquisition has been completed. In this paper, we propose a discriminative template matching algorithm for threshold-based spike sorting on high-density extracellular data. Such threshold-based spike sorting has a low and deterministic algorithmic delay, allowing for fast online spike sorting. Approach. At its core, threshold-based spike sorting is driven by linear filters. The proposed discriminative template matching filter design algorithm optimizes the output signal-to-peak-interference ratio in a data-driven fashion, assuming the template of the target spike is available. The latter allows the filter to suppress the spikes of interfering neurons and to resolve spike overlap. The data-driven filter design algorithm requires only templates of the target neurons of interest, which can be retrieved, e.g. through a prior clustering on an initial recording. Main results. The proposed discriminative template matching filters are validated on in vivo ground truth data and are shown to provide single-unit activity with good accuracy using a simple thresholding operation on the filter outputs. Significance. The low algorithmic complexity allows for computationally cheap and fast spike sorting. Also the proposed filters are guaranteed to be stable and have a deterministic delay. These characteristics make the proposed filter design method a valuable building block for online spike sorting, thereby enabling unit activity-based real-time and closed-loop experiments for high-density neural recordings.