Template generation for neural waveform analysis in extracellular recordings

Abstract

A fully automated system for sorting the action potential waveforms of multiple neurons in extracellular recordings has to generate a template for each neuron by averaging waveforms produced by that neuron. This requires a multidimensional clustering approach that can also segment waveforms from the continual recording, and average them with appropriate alignment. The performance of the sorting process is dictated by the quality of the templates. Confusion errors between classes and misalignment errors within classes deteriorate the templates generated by available algorithms. An algorithm that the authors developed was evaluated with simulations and physiological data.