Michigan State University - Research Groups

Spike detection is a fundamental step in the analysis of neural data collected during extracellular neural recording. Here, we propose spike detection algorithm that is a transformation of the raw data in a sparse representation space. The proposed method captures wavelet footprint, a compact representation of the transient of the signal, by measuring the power of scale space vectors in wavelet domain, and finds the optimal detection threshold by converting the original signal distribution to more threshold selection favorable distribution. Under the proposed scheme, a compact feature set is obtained at the same time detecting spikes that eliminates further feature extraction process for spike sorting. Our results demonstrate that this detection method yields improved performance, particularly in low SNRs, while keeping the desired separability of clusters for the spike sorting.