An empirical centre assignment in RBF network for quantification of anaesthesia using wavelet-domain features

The assessment of the hypnotic state of the brain is crucial to the process of an operation under general anaesthesia. A noninvasive method of quantifying depth of anaesthesia is through analysis of electroencephalogram (EEG). Among number of works done in the field, no single algorithm has been found exhibiting a precise measure in all of the hypnotic states. One can categorise algorithms as either a state-quantifier or a trend measure. State-quantifier algorithms can discriminate between different hypnotic states such as awake, light sedation, deep anaesthesia, etc. On the other hand, trend measure algorithms are employed to specify the short-term changes in hypnotic brain conditions, whether stable or not. In this paper, a trend measure algorithm is proposed, best describing changes in deep sedation and general anaesthesia. Wavelet analysis is used to extract features of the EEG signal and then a radial-basis function (RBF) network is trained to calculate the index. The training procedure is optimised according to meaningful selection of centres and spreads of the RBF. The proposed index is calculated for each five seconds of a signal, and an FIR filter smoothes the index. Results of comparison with BIS index show the proposed index ability to follow trends of hypnotic state well in the deep sedation and general anaesthesia regions.