Artificial neural networks facilitate bispectral analysis of electroencephalographic data

The brain is the target organ of anaesthesia yet the electroencephalogram (EEG) is not routinely monitored during anaesthetic procedures. This is partly due to the difficulty of interpreting complex changes in the EEG waveform with respect to anaesthetic conditions. Most attempts at developing EEG derived variables and display techniques have been based on spectral analysis. Bispectral analysis is a signal processing technique capable of detecting phase-coupling within a signal (which is lost using conventional power spectral analysis). Artificial neural networks (ANN) which excel at pattern classification were used in this study to interpret results of bispectral analysis. Six human subjects were studied at three anaesthetic levels (light, nominal, and deep anaesthesia). ANNs are shown to provide an efficient approach for extracting and using the additional signal information provided by bispectral analysis.