WaveLimit: An optimal spike sorting inclusion boundary
Spike sorting is the process of assigning neuronal action potentials to individual putative neurons based on extracellular recordings. Spike sorting may be partitioned into five major components: i) raw neural data is filtered, ii) spiking events are extracted as waveforms, iii) features are extracted from the waveforms, iv) clusters of waveforms are defined, and v) individual waveforms are assigned to their respective clusters. Here, we focus on the often underappreciated fifth component, deriving a useful principle to define a cluster boundary to maximize the theoretical information available from a single neuron. We implemented this boundary, along with an automatic cluster identifier, in a novel spike sorting algorithm, WaveLimit. We then compared WaveLimit to three state-of-the-art spike sorters. WaveLimit identified either the same or more units and included more spiking events per unit than the other sorters. WaveLimit also found units with fewer inter-spike interval violations and higher signal-to-noise ratios. Thus, better defining the cluster boundary improved spike sorting.