Selection of the Best Electroencephalogram Channel to Predict the Depth of Anesthesia.

Clement Dubost, Pierre Humbert, Arno Benizri, Jean-Pierre Tourtier, Nicolas Vayatis, Pierre-Paul VidalFront. Comput. Neurosci 2019 · OpenReview

TL;DR

This study tackles the challenge of accurately monitoring the depth of anesthesia (DoA) by identifying the single most informative electroencephalogram (EEG) channel. Researchers found that the frontal channel F8, when analyzed with Gaussian Naive Bayes, can effectively distinguish between awake and anesthetized states with high accuracy, paving the way for simpler and potentially more robust DoA monitoring systems.

What Problem Does This Paper Address?

Monitoring the depth of anesthesia (DoA) is crucial for patient safety during surgery, aiming to prevent awareness while avoiding over-sedation. However, there's no universally accepted "gold standard" for DoA monitoring. Current methods often rely on a combination of physiological parameters and electroencephalogram (EEG) data, with the latter sometimes using a large number of channels (2 to 128). This complexity and variability make it difficult to establish a consistent and optimal monitoring approach. This paper addresses the gap in understanding which specific EEG channels provide the most valuable information for predicting DoA, aiming to simplify monitoring by identifying a single, optimal channel.

Methods

The researchers employed a machine learning approach to analyze EEG data from patients undergoing general anesthesia (GA).

  • Data Acquisition: 32-channel EEG data was recorded from 30 patients during elective surgeries under GA. Physiological parameters and the Bispectral Index (BIS) monitor were also recorded.
  • Preprocessing: Raw EEG signals were filtered (1-30 Hz bandpass), segmented into 2-second epochs, and artifacts (electrosurgical interference, head movements, EOG) were removed using Independent Component Analysis (ICA) and amplitude thresholding.
  • Feature Extraction: Ten features were extracted from each EEG channel, focusing on the power distribution of different frequency bands (delta, theta, alpha, beta) and their ratios.
  • Channel Selection: Two classification algorithms, Quadratic Discriminant Analysis (QDA) and Gaussian Naive Bayes (GNB), were used to evaluate the discriminative power of each individual EEG channel. A leave-p-out cross-validation strategy (p=2) was employed to assess channel performance.
  • Feature Reduction & Model Validation: After identifying the best channel (F8), further analysis focused on minimizing the number of features required for classification using algorithms like K-Nearest Neighbors (KNN), Decision Tree Learning (DTL), QDA, and GNB.

Datasets

The study utilized EEG data collected from 30 patients undergoing general anesthesia at the Begin Military Hospital. Due to strict quality control and artifact removal, only 8 patients' recordings were ultimately used for the final analysis. No new benchmark datasets were introduced; the study used its own collected data.

Key Insights and Ideas

  • Frontal Lobe Dominance: The frontal brain region, specifically the F8 channel, emerged as the most informative for predicting the depth of anesthesia. This suggests that higher-level cognitive processing, which is heavily influenced by anesthesia, is most prominently reflected in frontal EEG activity.
  • Simplicity is Key: The study demonstrates that a single EEG channel, when analyzed with appropriate machine learning techniques, can achieve a high level of accuracy in distinguishing between awake and anesthetized states. This challenges the notion that a multitude of channels is always necessary.
  • Gaussian Naive Bayes Effectiveness: The Gaussian Naive Bayes (GNB) algorithm proved to be highly effective, particularly when combined with a reduced set of features, achieving excellent classification performance. Its simplicity and ability to handle feature independence made it suitable for this task.
  • Potential for Clinical Translation: The findings suggest a pathway towards developing simpler, more cost-effective, and potentially more reliable DoA monitoring devices that rely on a minimal number of EEG channels.

Key Results

  • The frontal channel F8 was identified as the best single channel for discriminating between awake and asleep states, achieving an Area Under the ROC Curve (AUC) of 0.92 ± 0.04 with the Gaussian Naive Bayes (GNB) algorithm.
  • The temporal channel T7 also showed good performance, with an AUC of 0.91 ± 0.04.
  • Using only data from the F8 channel, the GNB algorithm with just 5 features achieved an AUC of 0.93 ± 0.04.
  • The study found that the frontal and temporal regions generally performed better than central, parietal, and posterior EEG channels.

Limitations and Future Work

  • Small Sample Size: The study acknowledges that the final analysis was based on only 8 patients due to strict quality criteria and artifact removal. This small sample size is a significant limitation, and larger datasets are needed to confirm the findings.
  • Anesthesia Protocol Specificity: The study used a specific anesthesia protocol (propofol/sufentanil induction and sevoflurane maintenance). The results may not generalize to other anesthetic agents or combinations.
  • Binary Classification: The study focused on a binary awake/asleep classification. Future work could explore finer gradations of anesthesia depth and the transition periods (LOC/ROC) more thoroughly.
  • Feature and Algorithm Exploration: While GNB performed well, the authors suggest that other algorithms and more complex features (including physiological parameters) could potentially improve performance further.
  • Clinical Depth of Sedation: The authors propose that these findings could pave the way for monitoring the depth of sedation in intensive care units (ICUs), where current monitoring is challenging.

Summary updated

Abstract

Precise cerebral dynamics of action of the anesthetics are a challenge for neuroscientists. This explains why there is no gold standard for monitoring the Depth of Anesthesia (DoA) and why experimental studies may use several electroencephalogram (EEG) channels, ranging from 2 to 128 EEG-channels. Our study aimed at finding the scalp area providing valuable information about brain activity under general anesthesia (GA) to select the more optimal EEG channel to characterized the DoA. We included 30 patients undergoing elective, minor surgery under GA and used a 32-channel EEG to record their electrical brain activity. In addition, we recorded their physiological parameters and the BIS monitor. Each individual EEG channel data were processed to test their ability to differentiate awake from asleep states. Due to strict quality criteria adopted for the EEG data and the difficulties of the real-life setting of the study, only 8 patients recordings were taken into consideration in the final analysis. Using 2 classification algorithms, we identified the optimal channels to discriminate between asleep and awake states: the frontal and temporal F8 and T7 were retrieved as being the two bests channels to monitor DoA. Then, using only data from the F8 channel, we tried to minimize the number of features required to discriminate between the awake and asleep state. The best algorithm turned out to be the Gaussian Naïve Bayes (GNB) requiring only 5 features (Area Under the ROC Curve - ...
BibTeX
@inproceedings{dubost2019selection,
  author = {Clement Dubost and Pierre Humbert and Arno Benizri and Jean-Pierre Tourtier and Nicolas Vayatis and Pierre-Paul Vidal},
  title = {{Selection of the Best Electroencephalogram Channel to Predict the Depth of Anesthesia.}},
  booktitle = {Front. Comput. Neurosci},
  year = {2019},
  url = {https://openreview.net/forum?id=Z2MYC9i-cE9},
}
Figure