An Adaptive Detection for Automatic Spike Sorting Based on Mixture of Skew-t distributions

Abstract

Developing new techniques of simultaneous recoding using thousand electrodes, make the wide variety of spike waveforms across multiple channels. This problem causes spike loss and raise the crucial issue of spike sorting with unstable clusters. While there exist many automatic spike sorting methods, there has been a lack of studies developing robust and adaptive spike detection algorithm. Here, an adaptive procedure is introduced to improve the detection of spikes in different scenarios. This procedure includes a new algorithm which aligns the spike waveforms at the point of extremums. The other part is statistical filtering, which seeks to remove noises that mistakenly detected as true spike. To deal with non-symmetrical clusters, we proposed a new clustering algorithm based on the mixture of skew-t distributions. The proposed method could overcome the spike loss and skewed cells challenges by offering an improvement over automatic detection, alignment, and clustering of spikes. Investigating the sorted spikes, reveals that proposed adaptive algorithm improves the performance of the spike detection in both terms of precision and recall. The adaptive algorithm has been validated on different datasets and demonstrates a general solution to precise spike sorting, in vitro and in vivo.