Favorable Recording Criteria for Spike Sorting

Spike sorting is the generic term used to describe the procedure for identifying spikes in multi-neuron recordings and categorizing them according to waveform and amplitude differences. Correctly relating each spike to a category and accurately estimating its time of occurrence is prerequisite to processing recordings to determine individual and joint response properties of neurons. Several phenomena complicate the detection, classification and time-of-occurrence estimation steps of the sorting procedure. We expect that all spikes will have very similar waveforms since they are governed by the same biophysical laws of Hodgkin-Huxley. Knowing the waveforms of recorded action potentials enhances detection—identifying that a spike occurred in the presence of noise— but their similarity means that classification—relating a spike to a neuron—becomes a difficult problem. Luckily, slight variations in spike shape and amplitude due to differing geometries of the neurons producing the spikes and their differing distance from the recording electrode simplifies spike sorting issues. When spikes from different neurons overlap in time, both classification and detection become more difficult. For example, spikes from different units can negate each other, partially obliterating the spike and thus making detection more difficult, or they can enhance each other, resulting in a waveform that resembles neither, particularly in amplitude [1]. This paper characterizes the theoretical limits to which spikes can be correctly categorized, overlapping or not.