Multiscale depth of anaesthesia prediction for surgery using frontal cortex electroencephalography

Area of Science:

• Anesthesiology
• Neuroscience
• Biomedical Engineering

Background:

• Anesthetic agents are vital for preventing awareness during medical procedures.
• Precise dosing is essential to mitigate adverse effects.
• Depth of Anesthesia Monitoring can enhance anesthetic agent titration accuracy.

Purpose of the Study:

• To investigate patient-specific anesthesia depth prediction using electroencephalography (EEG) neural oscillations.
• To compare the accuracy of five post-processing methods for anesthesia depth prediction.
• To identify optimal methods for improving the accuracy of anesthesia depth monitoring.

Main Methods:

• EEG data was recorded from the frontal lobe of 10 sedated patients.
• Five post-processing approaches were evaluated: Noise Assisted-Empirical Mode Decomposition, Raw Signal, Linear Series Decomposition Learner, Deep Wavelet Scattering, and Deep Learning features.
• The Bispectral Index (BIS) was used as the ground truth for comparison.

Main Results:

• The Raw Signal, enhanced feature set, and Linear Discriminant Analysis (LDA) model demonstrated the highest classification accuracy.
• An average accuracy of 85.65% ±10.23% was achieved across the 10 subjects.
• Simpler models utilizing raw or enhanced features performed comparably or better than complex deep learning approaches.

Conclusions:

• Raw EEG signals combined with a low-complexity classifier offer a promising approach for accurate anesthesia depth prediction.
• Further validation on larger patient cohorts and exploration of continuous estimation methods are warranted.
• Optimizing features can further enhance model efficiency and reduce computational complexity for real-time monitoring.