Detecting sleep episodes in rats ECoG using simple artificial neural network

Abstract

Most of electroencephalographic-based and electrocorticographic-based (ECoG) sleep detection and sleep staging methods are based on analysis of time–frequency domains, amplitude, nonlinearity and other features. Here we study the interplay of mean and standard deviation of ECoG signals from two channels, recorded in frontal cortexes and occipital. For this purpose, a simple artificial neural network (perceptron) was developed and trained. The network consists of four linear and one nonlinear neurons. The corresponding coefficients for mena and standard deviation were found out after training the network. After training we demonstrate that mean value and standard deviation of at least two-channel ECoG signal can be applied to recognition of sleep episodes. Using the obtained training coefficients, we propose an original algorithm for automatic classification of sleep episodes on ECoG signals in rats. The method is based on the calculation of simple statistical characteristics and further classification based on a nonlinear function taken from their linear combination.
Our algorithms allows to develop a computationally simple approach for real-time sleep recognition (delay is equal to averaging window). Such approach can be very useful, for example, for development of laboratory devices for sleep deprivation.

This work was supported by the Russian Science Foundation, project no. 075-15-2022-1094.

Speaker

Nadezhda Semenova
Saratov State University
Russia

Discussion

Ask question