Support-vector classification of low-dose nitrous oxide administration with multi-channel EEG power spectra

Area of Science:

• Neuroscience
• Anesthesiology
• Machine Learning

Background:

• Nitrous oxide is commonly used in anesthesia.
• Monitoring patient response to anesthetic agents is crucial for safety.
• Electroencephalography (EEG) reflects brain activity and can be affected by anesthetic agents.

Purpose of the Study:

• To investigate the use of support-vector machines (SVMs) for classifying EEG patterns during nitrous oxide anesthesia.
• To determine the generalizability of EEG changes across participants.
• To identify key EEG features indicative of nitrous oxide dosage.

Main Methods:

• A single-blind, cross-over study involving 12 healthy participants.
• 32-channel EEG recordings were obtained during exposure to 0%, 20%, 30%, and 40% end-tidal nitrous oxide.
• Two SVMs were trained and tested: binary (0% vs. 40%) and multi-class (0%, 20%, 30%, 40%), using delta, theta, alpha, and beta band power features.

Main Results:

• The binary SVM achieved 92% accuracy, and the multi-class SVM achieved 52% accuracy, both significantly better than chance.
• Feature importance analysis revealed that decreased delta power in frontal regions was the most significant indicator of nitrous oxide effects.
• These findings were consistent across participants.

Conclusions:

• SVM classification of EEG data is a viable method for detecting nitrous oxide levels during anesthesia.
• The study highlights the importance of delta band activity in frontal EEG for monitoring nitrous oxide's effects.
• This approach holds potential for improving patient safety and monitoring in clinical anesthesia settings.