A novel automated spike sorting algorithm with adaptable feature extraction

Abstract

To study the electrophysiological properties of neuronal networks, in vitro studies based on microelectrode arrays have become a viable tool for analysis. Although in constant progress, a challenging task still remains in this area: the development of an efficient spike sorting algorithm that allows an accurate signal analysis at the single-cell level. Most sorting algorithms currently available only extract a specific feature type, such as the principal components or Wavelet coefficients of the measured spike signals in order to separate different spike shapes generated by different neurons. However, due to the great variety in the obtained spike shapes, the derivation of an optimal feature set is still a very complex issue that current algorithms struggle with. To address this problem, we propose a novel algorithm that (i) extracts a variety of geometric, Wavelet and principal component-based features and (ii) automatically derives a feature subset, most suitable for sorting an individual set of spike signals. Thus, there is a new approach that evaluates the probability distribution of the obtained spike features and consequently determines the candidates most suitable for the actual spike sorting. These candidates can be formed into an individually adjusted set of spike features, allowing a separation of the various shapes present in the obtained neuronal signal by a subsequent expectation maximisation clustering algorithm. Test results with simulated data files and data obtained from chick embryonic neurons cultured on microelectrode arrays showed an excellent classification result, indicating the superior performance of the described algorithm approach.

Spiking Neural Networks Based on OxRAM Synapses for Real-Time Unsupervised Spike Sorting

“…The most commonly used spike sorting approach consists of spike detection [mainly by thresholding (Tanskanen et al, 2015)], feature extraction (typically Principal Components Analysis, PCA) and clustering (e.g., k-means). Algorithms of this type have been implemented in commercial software (Bestel et al, 2012), however, they present several limitations, as they often need user supervision (manual tuning of the threshold parameters, choice of features to be extracted), they can fail to recognize overlapping spikes and moreover they are computationally expensive. Therefore, most of the neural signal processing is performed via offline software on desktop computers.…”

Wavelet methodology to improve single unit isolation in primary motor cortex cells

“…The shape of the action potentials is affected by the distances, orientation, and impedance of the recording device, as well as other surrounding factors such as proximate cells and overall background activity. A few researchers (Bestel et al, 2012; Lai et al, 2011) have used Haar wavelet, the oldest mother wavelet, in their analysis of spike sorting. This wavelet was included in this research in the form of the first order Daubachies (db1 = haar) even though it is highly dissimilar to a spike and displays poor performance in other signal processing applications (Cao et al, 2003).…”

“…PCA is used for data reduction and feature extraction (Bestel et al, 2012; Ghosh-Dastidar et al, 2008). PCA operates by deconstructing a data set into linearly uncorrelated variables called principal components.…”

Recent progress in multi-electrode spike sorting methods

“…For a comparison between PCA and wavelet based analysis, see Pavlov et al (2007). Note that the two can be combined (Bestel et al, 2012). …”