A fully automated spike sorting algorithm using t-distributed neighbor embedding and density based clustering

Abstract

In this study, a new spike sorting method was developed based on a combination of two methods, t-Distributed Stochastic Neighbor Embedding (t-SNE) and Density-Based Spatial Clustering of Applications with Noise (DBSCAN). Parameters of both methods were simultaneously optimized using a Genetic Algorithm (GA) using a simulated dataset containing 2 to 20 simultaneously recorded neurons. The performance of this method was evaluated using both the simulated dataset as well as real multichannel electrophysiological data. The results indicated that our fully automated algorithm using t-SNE-DBSCAN outperforms other state-of-the-art algorithms and human experts in spike sorting especially when there are a large number of simultaneously recorded units. Our algorithm also determines the noise waveforms and has an overall high sensitivity, precision and accuracy for correctly classifying waveforms belonging to each neuron (all >90%) without the need for manual corrections afterwards. Our method can be a crucial part of the analysis pipeline in particular when manual sorting of units is becoming prohibitive due to the sheer number of recorded neurons per session.