Classification of Three Anesthesia Stages Based on Near-Infrared Spectroscopy Signals

Area of Science:

• Biomedical Engineering
• Neuroscience
• Anesthesiology

Background:

• Anesthesia stage monitoring is crucial for surgical safety.
• Current monitoring methods may have limitations.
• Developing advanced, non-invasive monitoring techniques is essential.

Purpose of the Study:

• To classify different anesthesia stages using near-infrared spectroscopy (NIRS) signals and machine learning.
• To investigate the relationship between NIRS-derived cerebral hemodynamic variables and anesthesia depth.
• To assess the feasibility of NIRS for real-time anesthesia monitoring.

Main Methods:

• Collected NIRS signals, specifically right proximal oxyhemoglobin (HbO2), during maintenance (MNT), emergence (EM), and consciousness (CON) anesthesia stages.
• Analyzed phase-amplitude coupling (PAC) to compare differences between stages.
• Extracted time-domain (linear and nonlinear) and frequency-domain features.
• Utilized a support vector machine (SVM) classifier for three-stage classification.

Main Results:

• Phase-amplitude coupling (PAC) of NIRS signals demonstrated a gradual enhancement with increasing anesthesia depth.
• Achieved a three-classification accuracy of 69.27%, outperforming single-feature classification.
• The combined feature set improved classification performance.

Conclusions:

• NIRS signals, analyzed with machine learning, show feasibility for classifying anesthesia stages.
• PAC is a promising indicator of anesthesia depth.
• This approach offers potential for developing novel anesthesia monitoring systems to enhance patient safety.