High Performance Decoding of Behavioral Information from Background Activity in Extracellular Neural Recordings

Abstract

The usual approach to extracting information from extracellular neural recordings is to set an amplitude threshold and extract the information contained in suprathreshold data. Subthreshold data, on the other hand, contain superimposed low-amplitude spike waveforms from distant neurons and are usually considered background noise. Here we show that the estimated standard deviation of this background activity strongly covaries with behavior and can be used to decode behavioral information with very high accuracy. Our method for extracting behavioral information from subthreshold data can be used with traditional methods in brain-machine interfaces to further improve decoding accuracy.