Optimal Recognition Of Neural Waveforms
Isaac N. Bankman, Kenneth O. Johnson, W. Schneider · 2005
The investigation of biological neural networks requires reliable classification of neural action potentials in extracellular recordings. When the signal-to-noise ratio is bw, when the spike waveforms gradually change shape and amplitude, or when spikes overlap, the current sorting techniques cannot provide robust on-line operation due to their noise sensitivity and the extent of required human intervention. We present an optimal approach for the detection and classification of neural spikes as well as resolution of their superpositions. The optimal methods which are based on Bayesian classification, perform at the theoretically expected level in simulations with physiobgical neural spikes and neural noise. The optimal performance is obtained by using a whitening filter that eliminates the autocorrelation in neural noise.