Inference of Brain States Under Anesthesia With Meta Learning Based Deep Learning Models

Area of Science:

• Neuroscience
• Anesthesiology
• Machine Learning

Background:

• Monitoring anesthetic-induced unconsciousness is crucial for clinical care and neuroscience.
• Electroencephalogram (EEG) provides real-time brain activity data but faces challenges with low signal-to-noise ratio (SNR) in anesthesia settings.
• Conventional machine learning models struggle with EEG data variability and noise.

Purpose of the Study:

• To develop and evaluate a novel deep learning framework, Anes-MetaNet, for classifying brain states under anesthesia.
• To address limitations of existing methods in handling low SNR and cross-subject variability in EEG data.
• To improve the accuracy and reliability of real-time consciousness monitoring during anesthesia.

Main Methods:

• Proposed Anes-MetaNet framework integrating Convolutional Neural Networks (CNNs) for feature extraction and Long Short-Term Memory (LSTM) networks for temporal analysis.
• Employed a meta-learning approach to manage significant cross-subject variability in EEG data.
• Utilized a multi-stage training paradigm and visualized high-level feature mapping to enhance model performance.

Main Results:

• Anes-MetaNet demonstrated superior performance in classifying brain states under anesthesia compared to existing methods.
• The framework effectively extracted power spectrum features and captured temporal dependencies in EEG signals.
• Experiments on an office-based anesthesia EEG dataset validated the model's effectiveness.

Conclusions:

• Anes-MetaNet offers a robust and effective deep learning solution for classifying brain states during anesthesia.
• The meta-learning approach significantly improves the handling of cross-subject variability in EEG data.
• This framework holds promise for advancing real-time, objective monitoring of consciousness depth during anesthesia.