Improving spike separation using waveform derivatives

This paper presents spike derivatives as a tool for spike feature extraction to improve the separation of similar neurons. The theoretical framework of neuronal geometry signatures and noise shaping to perform the spike derivative is formulated first, and based on the derivations we show that the first derivative of the spikes manifests the waveform difference contributed by the geometry signatures and also reduces the associated low-frequency noise. Quantitative comparisons of sorting neurons using spikes and their derivatives are performed on spike sequences from a public database, and improved results are observed when using spike derivatives.