L.A. Manilo, S.S. Volkova

Entropy as an information measure reflects the degree of irregularity, complexity and non-predictability of the signal. Electroencephalogram regularity degree is strongly correlated with the current level of the cerebral activity. So entropy parameters calculation can prove to be an effective way for anaesthesia depth estimation. In current research we introduce a few indexes representing different methods of approximate entropy estimation. This research is made with the outlook for application one of this parameters in an anaesthesia monitor. This means that resulted algorithm should be fast enough to work in real-time mode, also it should increase smoothly with anaesthesia depth decreasing, be noise-immune, and so on. Since approximate entropy calculation needs big computer computational capabilities, it was decided to estimate the time it computation had taken. This was carried out on the machine working in mode of 1.2 GHz processor frequency. Processor frequency was decreased for low-price processor emulation, which are used in medical devices. Reliable entropy estimation needs to analyze signal fragments of 5 sec length. The conclusion is that these signals should not have sampling frequency over 250 Hz. All further tests were made on signals sampled with 250 Hz. Entropy parameters comparison is based on estimation of their classification quality of two conditions "wake" and "anaesthesia". It was decided to use Fisher test J as a measure. The best results of recognition of two conditions were obtained ar R = 2 µV and R = 0.15*STD, where R is a threshold value determining the size of phase space cells (and the main parameter determining the entropy calculation), and STD is standard deviation of a signal. In this case signal components with amplitude of less than 2 µV are excluded from the analysis. Maximum obtained J value is 7.4, which tells about high classification quality.