OR in the News

Kumar, R., Skowno, J., von Ungern-Sternberg, B. S., et al. | J Clin Monit Comput (2025) 39(5):999-1014 | Quantitative electroencephalogram and machine learning to predict expired sevoflurane concentration in infants

November 29, 2025

Background: Processed electroencephalography (EEG) indices used to guide anesthetic dosing in adults have not been validated in young infants. Quantitative EEG (qEEG) parameters derived from raw EEG, combined with machine learning, may improve anesthetic monitoring in this population. This study evaluated whether machine learning models using qEEG could accurately classify expired sevoflurane (eSevo) concentrations in young infants.

Methods: Frontal EEG recordings were obtained from infants aged ≤3 months and matched into one-minute epochs with expired sevoflurane concentrations. Fifteen qEEG parameters were extracted from each epoch. Eight machine learning models combined these parameters to classify each epoch into one of four eSevo concentration ranges (0.1-1.0%, 1.0-2.1%, 2.1-2.9%, and >2.9%). A post hoc SHAP analysis was performed on 64 epochs to identify the most influential qEEG features. The remaining epochs were randomly divided into training (80%) and testing (20%) datasets over 50 iterations. Model performance was assessed using accuracy and F1-score.

Results: Forty-two infants contributed 4,574 EEG epochs. The best-performing classifiers, including K-nearest neighbors, the default multi-layer perceptron, and support vector machine, achieved accuracies ranging from 67.5% to 68.7%. Burst suppression ratio and beta entropy were the strongest contributors to model performance. Post hoc analysis excluding burst suppression ratio produced similar predictive performance.

Conclusions: Machine learning applied to quantitative EEG moderately predicted expired sevoflurane concentrations in young infants. Burst suppression ratio emerged as the most informative EEG feature, reflecting underlying EEG changes captured by other quantitative parameters. These findings provide insight into EEG feature selection and machine learning approaches for developing infant-specific EEG-based anesthesia monitoring algorithms.

Keywords: machine learning; infant electroencephalogram; EEG; pediatric EEG anesthesia; quantitative EEG anesthesia; SHAP EEG analysis.