EEG Signals Analysis Using non-linear dimension reduction method and Support Vector Machine for Monitoring the Depth of Anesthesia

1. BackgroundEstimating the depth of anaesthesia (DOA) during surgery is a challenge in anaesthesia research to prevent intraoperative awareness and delayed recovery during anaesthesia. Because the anaesthetic drugs act mainly on the central nervous system, Electroencephalogram (EEG) signal analysis is very useful. This study introduces a new approach for measuring the DoA. 2. MethodEEG signals of 17 patients were collected during anaesthesia with sevoflurane and their anaesthetic depth levels were divided into four levels of awake, light, general and deep states. Firstly, 12 features including time (Kurtosis, Skewness), frequency (Alpha, Beta, Delta, Theta index, spectral edge frequency, and medium frequency), entropy (sample entropy, shannon permutation entropy) and non-linear features (Lyapunov, detrendedfluctuation analysis) are extracted from EEG signal. Then, by applying an algorithm according to nonlinear dimension reduction method named, Local Linear Embedding (LLE), the best features is extracted. Finally, we feed these extracted features to Support Vector Machine (SVM) classification algorithm. 3. ResultsThe presented method classifies EEG data into four states in 17 patients with accuracy is 98%, and compared to a commercial monitoring system successfully. By using non-linear dimensional reduction method to 8 and classification, we able to improve the classification accuracy. This structure can successfully model systems with nonlinear relationships between input and output, and also classify overlapped classes accurately.4. Conclusions This method based on best features of spectral, fractal entropy and nonlinear dimension reduction method is potentially applicable to a new real time monitoring system to help the anesthesiologist for continuous assessment of DoA.