Loading...
Decoding Anesthesia Depth: Quantifying EEG Signals for Accurate Measurement
Mustafa, Lama Al-Amoudi, Nora Baqais, Fatima
Mustafa, Lama
Al-Amoudi, Nora
Baqais, Fatima
Citations
Altmetric:
Type
Supervisor
Subject
Date
Collections
Research Projects
Organizational Units
Journal Issue
Abstract
The analysis of electroencephalogram (EEG) signals during anesthesia is essential for
enhancing patient safety, reducing risks, and improving anesthesia management. This research
focuses on developing a systematic pipeline for anesthesia monitoring which includes applying
various preprocessing, feature extraction, feature reduction, and machine learning
classification techniques for EEG signals to improve diagnostic accuracy and noise reduction
during anesthesia. EEG preprocessing employs three phases which are detrending, variational
mode decomposition (VMD), and denoising including Gaussian filtering, moving averages,
Independent Component Analysis (ICA), and Discrete Wavelet Transform (DWT).
Additionally, feature extraction was performed using time-domain features (statistical and
entropy-based) and frequency-domain features. After that, the features were reduced using
principal component analysis, correlation matrices, and variance methods. Furthermore, the
feature classification was applied using machine learning models, including Artificial Neural
Networks (ANN), Support Vector Machines (SVM), and Convolutional Neural Networks
(CNN). The results revealed that DWT was the most effective noise reduction method,
achieving a signal-to-noise ratio (SNR) of 43.23 dB. Feature extraction and reduction processes
reduced 20 extracted features (13 from the time domain and seven from the frequency domain)
into nine critical features using Principal Component Analysis (PCA), correlation matrices, and
variance-based methods. In the classification phase, SVM outperformed other machine
learning models, achieving an accuracy of 98.7%, compared to 89.2% for ANN and 85.6% for
CNN. These findings underscore the importance of, DWT for denoising, feature extraction and
reduction, and SVM for accurate EEG classification in anesthesia monitoring.