A modified K-means clustering algorithm for classification of neurophysiological waveforms (slow waves)

Two types of slow waves with complex shapes, reflecting sympathetic nervous system activity of the cat, were analyzed in order to find homogeneous waveform classes: compound action potentials of postganglionic sympathetic nerves and excitatory postsynaptic potentials of single preganglionic sympathetic neurons. A modification of the K -means clustering algorithm was developed in which the cluster means do not explicity appear in the calculations. Rather, the input for the clustering is the Euclidean distance matrix computed in a separate step. The advantages of the modified algorithm are (1) fast repetition of clustering with different K´s and (2) more accurate alignment of spontaneously occurring waveforms