An automatic sleep classifier for laboratory rodents.

Abstract

Two successive main steps are usually performed in automatic sleep classification: a) a periodic quantification of the EEG signal by a set of indices featuring each polygraphic epoch by a point in a multidimensional space, b) a stage decision to determine the cluster that includes the representative point for a given epoch (discriminant analysis). As solid state logic comparators simulate linear discriminant functions, this article describes an improved classifier for automatic sleep scoring in laboratory rodents. Four indices were found to be discriminant on both rats and mice: F(the energy in the EEG theta band relative to that in the delta band), Z(number of EEG zero-crossings), D(variability of the EEG amplitude) and M(integrated EMG). They are measured by analog techniques for 30 sec epochs from an EEG and an EMG lead per animal. At the end of each epoch, the logic results of their respective comparators serves as an input to a hardwired modifiable truthtable where a sleep stage decision is elaborated. The outputs of 4 multiplexed classifiers are controlled by microprocessor and may be used for chart or digital recordings on various peripherals. The overall agreement with visual scoring is more than 95% for both rats and mice.