Depth Analysis of Anesthesia Using EEG Signals via Time Series Feature Extraction and Machine Learning

Abstract

The term “anesthetic depth” refers to the extent to which a general anesthetic agent sedates the central nervous system with specific strength concentration at which it is delivered. The depth level of anesthesia plays a crucial role in determining surgical complications, and it is imperative to keep the depth levels of anesthesia under control to perform a successful surgery. This study used electroencephalography (EEG) signals to predict the depth levels of anesthesia. Traditional preprocessing methods such as signal decomposition and model building using deep learning were used to classify anesthetic depth levels. This paper proposed a novel approach to classify the anesthesia levels based on the concept of time series feature extraction, by finding out the relation between EEG signals and the bi-spectral Index over a period of time. Time series feature extraction on basis of scalable hypothesis tests were performed to extract features by analyzing the relation between the EEG signals and Bi-Spectral Index, and machine learning models such as support vector classifier, XG boost classifier, gradient boost classifier, decision trees and random forest classifier are used to train the features and predict the depth level of anesthesia. The best-trained model was random forest, which gives an accuracy of 83%. This provides a platform to further research and dig into time series-based feature extraction in this area.