Circuit Dynamics and Computational Neuroscience I.1.l Deep and machine learning Monday AM + Wednesday AM

2474 - A novel software for electrophysiology signal analysis and labelling


Abstract Body

Electrophysiology experiments generate a large amount of data that needs to be pre-processed and labelled to allow understanding of the underlying phenomena. Although historically the labelling of signals has been done manually by electrophysiologists nowadays different software solutions have emerged to facilitate this task. However, to date no software package includes pre-processing, feature analysis and labelling in a single tool. This paper presents a new signal analysis software specialized for MEA but adaptable to EEG that allows to perform all phases of signal analysis.

The signals obtained from the MEA experiments (or EEG) are filtered using a configurable filter, being the default MEA filter a lowpass filter with a passband frequency of 50 Hz to remove the power-line interference.

After that, the signal´s features are extracted using GPU computing. These features include (but are not limited to) Power spectral density, spectral entropy, Fourier transform and statistical features (as mean, skewness or kurtosis).

Once the signal has been analysed, event labelling is carried out. In this case, the software validation has been done with an epilepsy model, so the labelling is focused on searching ictal events. For this purpose, a floating threshold is established by means of a recursive search for peaks, which allows only the most prominent ones to be detected. A clustering of these peaks is then performed using a Kernel density estimation.

This labelling process can be adapted for other purposes like spike sorting. The software allows fast signal analysis in an efficient and reliable way.