On the Application of Parsimonious Models for Surgical Anesthesia Depth Prediction using EEG Recordings

Anesthetics are commonly employed during surgical processes to dull out sensations. Despite their widespread adoption and effectiveness, there continue to be shortcomings associated with their use that span cognitive and neurological side effects. An avenue towards minimizing these side effects involves better means of monitoring the depth of anesthesia (DoA) administered during the sedation process. As part of strides towards investigating this problem, this work involved using EEG brain waves acquired from the frontal cortex region during surgery to predict DoA. Specifically, we employed a signal decomposition pre-processing technique, followed by a feature selection process to obtain an optimal feature set used for modelling linear and nonlinear regression-based algorithms for prediction tasks. The obtained results showed a range of prediction accuracies across the various patients, and the proposed approach produced the highest classification accuracy when predicting the awake state-a feat which is also echoed across the literature.