A Takagi-Sugeno fuzzy neural network-based algorithm with single-channel EEG signal for the discrimination between light and deep sleep stages

Abstract

We propose in this paper an efficient and effective algorithm for the discrimination between light and deep sleep stages of a subject based on the single-channel electroencephalography (EEG) signal. To realize the proposed approach, the acquired EEG signal is first convolved with a set of three filters to extract the corresponding component of δ, θ, and α waves. After that, we calculate the average point energy (A.P.E.) of the extracted δ, θ, and α components to construct a three-dimensional feature vector, which will be used as the input of a four-layer Takagi-Sugeno fuzzy neural network (TS-FNN) for further discrimination process. Finally, a value which indicates the depth degree of sleep will be given in the fourth layer of TS-FNN. In order that a satisfactory accuracy can be obtained with the proposed TS-FNN classifier, EEG signals of four subjects in the Sleep-EDF database will be used as the training data. During the performance evaluation process, the Sleep-EDF database will be used again, but with EEG signal from different subjects. As we will see the proposed algorithm is quite feasible for real-time applications due to moderate computational cost. Moreover, our experiment shows that the classification accuracy up to 81.98% can be obtained with the proposed approach when 3418 epochs are evaluated. Besides, the sensitivity and specificity for deep sleep stages are 74.19% and 84.25% respectively, while that for light sleep stage are 84.25% and 74.19% respectively, which outperforms that by most of the existing state-of-the-art algorithms.