Deep Learning Techniques to Improve Intraoperative Awareness Detection from Electroencephalographic Signals

Area of Science:

• Anesthesiology
• Neuroscience
• Medical Technology

Background:

• Millions of patients experience intraoperative awareness annually, risking post-traumatic stress.
• Detecting patient movement during surgery is crucial for preventing adverse outcomes.
• Electroencephalographic (EEG) signals offer a potential method for monitoring awareness.

Purpose of the Study:

• To evaluate deep learning models for detecting motor imagery from EEG signals during general anesthesia.
• To compare the performance of EEGNet and other deep learning models against traditional machine learning approaches.
• To assess the accuracy and false positive rates of these models in identifying intraoperative awareness.

Main Methods:

• Filtered EEG data was used to train and test several deep learning models, including EEGNet, deep convolutional networks, and shallow convolutional networks.
• Performance was compared against non-deep learning methods: linear discriminant analysis with common spatial patterns, minimum distance to Riemannian mean, and tangent space projection with logistic regression (TS+LR).
• Key metrics included classification accuracy and false positive rates.

Main Results:

• EEGNet demonstrated a statistically significant improvement in classification performance (p < 0.01) compared to other classifiers.
• EEGNet achieved 7.2% higher accuracy than the best-performing non-deep learning classifier (TS+LR).
• The study successfully measured accuracy and false positive rates for all evaluated models.

Conclusions:

• Deep learning models, particularly EEGNet, show significant promise for enhancing intraoperative awareness detection.
• Accurate detection of motor activity via EEG can alert medical staff, potentially mitigating risks associated with surgical awareness.
• This technology offers a pathway to improved patient safety during general anesthesia.