Kilosort: realtime spike-sorting for extracellular electrophysiology with hundreds of channels

Abstract

Advances in silicon probe technology mean that in vivo electrophysiological recordings from hundreds of channels will soon become commonplace. To interpret these recordings we need fast, scalable and accurate methods for spike sorting, whose output requires minimal time for manual curation. Here we introduce Kilosort, a spike sorting framework that meets these criteria, and show that it allows rapid and accurate sorting of large-scale in vivo data. Kilosort models the recorded voltage as a sum of template waveforms triggered on the spike times, allowing overlapping spikes to be identified and resolved. Rapid processing is achieved thanks to a novel low-dimensional approximation for the spatiotemporal distribution of each template, and to batch-based optimization on GPUs. A novel post-clustering merging step based on the continuity of the templates substantially reduces the requirement for subsequent manual curation operations. We compare Kilosort to an established algorithm on data obtained from 384-channel electrodes, and show superior performance, at much reduced processing times. Data from 384-channel electrode arrays can be processed in approximately realtime. Kilosort is an important step towards fully automated spike sorting of multichannel electrode recordings, and is freely available (github.com/cortex-lab/Kilosort).

A Fully Automated Approach to Spike Sorting

“…In other situations, clustering is automated, but the user must curate the results by selecting which clusters to reject, merge, or even split (Hill et al, 2011; Kadir et al, 2014; Rossant et al, 2016). There also exist post-processing steps that resolve overlapping spikes (Ekanadham et al, 2014; Franke et al, 2015; Pachitariu et al, 2016; Pillow et al, 2013), and algorithms based on independent component analysis (ICA) that do not explicitly involve clustering (Takahahi et al, 2002). Presently, despite many available packages and proposed algorithms, no generally adopted software packages offer fully automated sorting that can take in the raw time series data and output spike times and identities without the expectation of further curation.…”

A neural decoder for learned vocal behavior

“…Spikes were detected and sorted using Kilosort; details of the procedure can be found in 14 . The number of clusters was set to 32 or 64 (twice the number of channels of the probe), and we did not perform a post-hoc instance of manual curation, splitting or merging after the initial automatic splitting.…”

“…1d-f). The neural activity is (t , t , ) r i i−1 ... t i−M +1 fed as a matrix comprising mean firing rates in each time bin, of each putative single/multi-unit automatically sorted from the recordings 14 (32/64 clusters); the spectral components of the song are represented by the power across 64 log-spaced frequency bands. For each session (day), we separate the 70-110 renditions of a motif the bird sang.…”

Electrical Stimulus Artifact Cancellation and Neural Spike Detection on Large Multi-Electrode Arrays

“…Furthermore, we notice the EI estimation step is essentially spike sorting; therefore, there is room for the use of state-of-the-art [48,49] methods to achieve efficient implementations. This outer loop would be especially helpful to enable the online update of the EI in order to counteract the effect of tissue drift, or to correct possible biases in estimates of the EI provided by visual stimulation [50,51], which could lead to problematic changes in EI shape over the course of an experiment.…”