Source location as a feature for the classification of multi-sensor extracellular action potentials

A novel framework for feature extraction in multi-sensor action potential sorting

“…4) provide promising results as well. However, unlike clustering with the estimated neuron locations (Chelaru and Jog, 2005;Szymanska et al, 2013), there is no need for a forward model. The PCA based approach outperforms the wavelet based approaches (e.g., MSD-WT) for a given number of features since any AP is comprised more wavelet basis functions than those of PCA.…”

“…Other approaches utilize spatial information in multi-channel measurements to extract features for clustering. Examples include estimating neuron locations (Chelaru and Jog, 2005;Szymanska et al, 2013) or calculating independent components Brown et al, 2001). To localize a neuron with multi-sensor measurements, a "forward model" describing the propagation of APs through extracellular media is typically adopted.…”

Intrinsic dimensionality of extracellular action potentials

“…EAPs were detected in an unsupervised manner using the continuous WT method described in [2]. This resulted in detecting 1030 EAPs (result comparable to a supervised detection method in [19]). The EAPs were aligned [10], and 2 ms of data centered at the EAP peak were extracted.…”

“…To facilitate further analysis, the EAPs were represented in the wavelet domain using bior1.3 wavelet class [2], [20], and classified using the top 15 wavelet coefficients. Our prior study [19] has revealed 6 EAP classes in these data, and so k-means clustering (k = 6) was performed in the 15-D wavelet space. The results of classification are shown in Fig.…”

A supervised multi-sensor matched filter for the detection of extracellular action potentials

“…Traditional template approaches are supervised, using both spike and noise training measurements for various parameter estimation before detection. The template is then usually generated from extracellular action potentials (EAPs) selected by an analyst from the data [3], [6]. Un-supervised algorithms using wavelets as templates, and assuming no prior knowledge about the signal, are also becoming increasingly popular [4], [5], [7], [8].…”