Spike sorting: a novel shift and amplitude invariant technique

Abstract

This paper deals with the spike classification problem encountered in multi-unit recordings of neural activity in the brain. We recently developed a new methodology for estimating and classifying multi-units recorded by means of multichannel silicon probes from the observed spike trains (Proceedings of the ICASSP’01, May 2001, pp. 2813–2816; Proceedings of the IEEE 35th Asilomar Conference on Signals, Systems and Computers, Pacific Grove, CA, November 2001). In this work, we demonstrate the robustness of the technique to single unit spike amplitude variation often encountered in burst activity or long term chronic recordings. In low signal-to-noise ratio scenarios where variability in spike threshold crossings during classical detection is always a problem, we show that the technique is extremely robust to shifts in spike event times. Results showing the efficiency of the algorithm from simulated and experimental data are presented.