Automatic selection of neuronal spike detection threshold via smoothed Teager energy histogram

Spike detection is a prerequisite to analyzing neuronal activity. While simple spike detectors are favorable for hardware implementation, manual setting of spike detection threshold can be tedious and time-consuming, especially for extracellular recordings of multiple neuronal activity. This paper, therefore, investigates and proposes an automatic threshold selection using smoothed Teager energy histogram (STEH), with consideration of signal prewhitening, histogram bin width, and histogram equalization. Results from spikes with signal-to-noise ratio = 0.9-2.6 dB reveals that (1) prewhitening of neural signals can enhance true detection rate (TDR) by 10-20% at constant false alarm (FA) ranging 3-12 spikes/s; (2) Freedman-Diaconis choice of STEH bin width delivers higher TDR (1.52 ± 1.41%) and FA (0.48 ± 0.25 spikes/s) than square-root choice; and (3) histogram equalization can raise average TDR by 2.84% and FA by 1.01 spikes/s. Thresholds determined by STEH with signal prewhitening, Freedman-Diaconis choice or square-root choice of bin width, and histogram equalization fall around knee point of receiver operating characteristic curve, yielding average TDR = 87.88% and FA = 1.82 spikes/s for Freedman-Diaconis choice, and TDR = 86.49% and FA = 1.41 spikes/s for square-root choice of STEH bin width.