Spike separation in multiunit records: a multivariate analysis of spike descriptive parameters.

Variables describing unit discharges from multiunit extracellular records were treated, using multivariate statistical analysis for further automatic spike separation. A spike was described by 8 parameters (3 amplitude values, 3 time values and 2 slopes). These 8 parameters were summarized by 2 new independent factors, the principal components (PC). Examination of the 2 PCs from each spike allowed the observer to determine PC families corresponding to spike populations. From these the spikes were automatically classified. PC analysis showed groups of redundant parameters. Elimination of redundancy decreased the computing time without lowering the separation efficiency of the method. Three parameters turned out to be sufficient for spike discrimination: the maximum and the minimum amplitudes and the time between these. This agrees well with the physical basis of current propagation in the conducting medium, from the firing neurone to the recording electrode.