JMIR Publications

ABSTRACT

Background:

General anesthesia induces temporary loss of consciousness, and electroencephalography (EEG)-based monitoring is crucial for tracking this state. However, EEG-based indices that are used to assess the depth of anesthesia can be influenced by various factors, potentially leading to misleading outputs.

Objective:

This study aimed to explore the feasibility of using unsupervised machine learning on processed EEG data to enhance anesthetic depth assessment.

Methods:

Over 16,000 data points were collected from patients who underwent elective lumbar spine surgery. The EEG data were processed using a bandpass filter and Fast Fourier Transform for power spectral density estimation. Unsupervised machine learning with Fuzzy C-means clustering was applied to categorize anesthesia depth into three clusters: slight, proper, and deep.

Results:

Fuzzy C-means clustering identified distinct anesthesia depth groups based on delta, alpha, theta, and beta band power ratios. Visual representations validated the clustering results, which were consistent across individual patient data. The figures demonstrate the application of clustering to EEG data, revealing detailed anesthesia depth estimations.

Conclusions:

This study developed a machine learning-based methodology for anesthesia depth assessment, demonstrating feasibility and providing preliminary insights into classification, visualization, and patient-specific management. By applying Fuzzy C-Means clustering to processed EEG data, this approach enhances anesthesia depth understanding and integrates with existing monitoring modalities.