Estimating Depth of Anesthesia for Drug Infusion Using Human EEG Signals

Abstract

General anaesthesia is a medically induced unconscious state or coma resulting from the use of anaesthetic agents on patients. It guarantees a safe and painless surgical operation for patients. However, there are numerous evidences that the patients may suffer from postoperative side effects. A limited number of solutions are currently available to deal with these issues. Therefore, optimal quantity of anaesthesia is a must to reduce the inverse effects of under-dosing or overdosing. EEG signals are, at present, most widely used signals for estimating Depth of Anaesthesia (DoA). The aim is to estimate the DoA based on the Sample Entropy of the EEG signals using Intrinsic Mode Functions (IMFs) which were extracted using Empirical Mode Decomposition (EMD) techniques. These Sample Entropy values were used along with the recorded BIS values in Machine Learning algorithms such as Artificial Neural Network (ANN) and Random Forest Regressor. In this paper, we have proposed and implemented a method for estimation of DoA using human EEG signals. We have filtered the raw EEG data using EMD and MEMD methods. And then calculated the SampEn value of 5s segments of that filtered data. Then applied different ML models to predict the values based on BIS data as target. We have taken RMSE and MAE as metrics to evaluate the models. The Random Forest Regressor model gives the results with a Root Mean Squared Error (RMSE) of 11.73 and the Mean Absolute Error (MAE) of 5.75 as the lowest error amongst all models.

Keywords: EEG, DoA, IMFs, EMD, MEMD, BIS, Sample Entropy, ANN, Random Forest Regressor.