A comparison of methods for clustering electrophysiological multineuron recordings

Techniques for the automatic clustering of extracellular multineuron recordings from the nervous system are compared for efficiency and accuracy. Selected waveforms were combined with noise to form test data with known classifications. After identical preprocessing using a Schmitt trigger threshold detector, the K-means, template matching and ART2 algorithms were applied to the same data. Measurements of the efficiency and utility of the three algorithms are presented using both the raw waveforms and the weightings of the first two principal components. Additionally, all three algorithms were tested with data obtained from electrophysiological experiments.