Comparison of Adaptive and Fixed Segmentation in Different Calculation Methods of Electroencephalogram Time-series Entropy for Estimating Depth of Anesthesia

This paper proposes a combined method including adaptive segmentation and time-series Shannon entropy of electroencephalograms (EEG) to monitor depth of anesthesia (DOA). The entropy of a single channel EEG was computed through various methods of quantization. These methods are different in number of bins associated to the whole range of amplitude. The EEG data was captured in both ICU and operating room and different anesthetic drugs, including propofol and isoflurane were used. Due to the non-stationary nature of EEG signal, adaptive segmentation methods seem to have better results. Our adaptive windowing methods consist of variance and auto correlation (ACF) based methods. We have compared the results of fixed and adaptive windowing in different methods of calculating entropy in order to estimate DOA. Coefficient of determination (R ) was used as a measure of correlation between the predictors and BIS index to evaluate our proposed methods. The results show that entropy decreases with decreasing DOA. In ICU, our proposed method reveals better performance than previous works. In both ICU and operating room, the results indicate the superiority of our method, especially applying adaptive segmentation. The mixture of adaptive windowing methods with different methods of calculating entropy would result in an outstanding performance.