A COMPUTATIONAL FRAMEWORK TO DISCRIMINATE DIFFERENT ANESTHESIA STATES FROM EEG SIGNAL

This paper develops a computational framework to classify different anesthesia states, including awake, moderate anesthesia, and general anesthesia, using electroencephalography (EEG) signal. The proposed framework presents data gathering; preprocessing; appropriate selection of window length by genetic algorithm (GA); feature extraction by approximate entropy (ApEn), Petrosian fractal dimension (PFD), Hurst exponent (HE), largest Lyapunov exponent (LLE), Lempel-Ziv complexity (LZC), correlation dimension (CD), and Daubechies wavelet coefficients; feature normalization; feature selection by non-negative sparse principal component analysis (NSPCA); and classification by radial basis function (RBF) neural network. Because of the small number of samples, a five-fold cross-validation approach is used to validate the results. A GA is used to select that by observing an interval of 2.7 s for further assessment. This paper assessed superior features, such as LZC, ApEn, PFD, HE, the mean value of wavelet coefficients for the beta band, and LLE. The results indicate that the proposed framework can classify different anesthesia states, including awake, moderate anesthesia, and general anesthesia, with an accuracy of 92.07%, 96.18%, and 93.42%, respectively. Therefore, the proposed framework can discriminate different anesthesia states with an average accuracy of 93.89%. Finally, the proposed framework provided a facilitative representation of the brain's behavior in different states of anesthesia.