Developing a robust model to predict depth of anesthesia from single channel EEG signal

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Monitoring depth of anesthesia (DoA) using electroencephalograph (EEG) signals presents ongoing challenges for anesthesiologists.
• Accurate DoA monitoring is crucial for patient safety and optimizing anesthetic administration.
• Existing methods may face limitations, particularly in scenarios with compromised signal quality.

Purpose of the Study:

• To propose and evaluate an intelligent model for predicting DoA from single-channel EEG signals.
• To assess the model's performance using hierarchical dispersion entropy (HDE) and community graph detection approach (CGDA).
• To compare the model's efficacy against the Bispectral Index (BIS) especially under poor signal conditions.

Main Methods:

• EEG signal segmentation using a sliding window approach.
• Feature extraction from EEG segments and application of Hierarchical Dispersion Entropy (HDE).
• Feature selection using a Community Graph Detection Approach (CGDA) to identify relevant predictors of DoA.

Main Results:

• The proposed HDE coupled with CGDA model demonstrated effective DoA prediction.
• Evaluation using statistical metrics including Q-Q plots, regression, and correlation coefficients.
• The model exhibited an earlier reaction than the BIS index during transitions from deep to moderate anesthesia in cases of poor signal quality, achieving a highest Pearson correlation coefficient of 0.96.

Conclusions:

• The developed intelligent model offers a viable approach for DoA monitoring from single-channel EEG.
• The combination of HDE and CGDA shows potential for robust feature extraction and selection in noisy EEG data.
• The model's improved performance in poor signal quality warrants further investigation for clinical application.