Artificial apnea classification with quantitative sleep EEG synchronization

Area of Science:

• Neuroscience
• Biomedical Engineering
• Sleep Medicine

Background:

• Sleep apnea, including Central Sleep Apnea (CSA) and Obstructive Sleep Apnea (OSA), significantly impacts sleep quality and overall health.
• Accurate differentiation between CSA and OSA is crucial for effective treatment strategies.
• Electroencephalography (EEG) signals offer potential biomarkers for sleep disorder diagnosis.

Purpose of the Study:

• To evaluate the efficacy of linear (Coherence Function - CF) and nonlinear (Mutual Information - MI) EEG synchronization methods for discriminating CSA and OSA from healthy controls.
• To utilize Feed Forward Neural Network (FFNN) architectures for classifying sleep EEG data based on synchronization patterns.
• To identify reliable EEG signal features for sleep apnea classification.

Main Methods:

• Sleep EEG data from stage 2 sleep were analyzed using CF and MI to quantify synchronization between C3 and C4 EEG recordings.
• Several FFNN architectures with varying numbers of neurons and hidden layers were trained to classify the sleep EEG data.
• The performance of CF and MI as classification features was assessed using FFNN models.

Main Results:

• EEG synchronization patterns, analyzed by CF and MI, are closely related to the presence of sleep disorders like CSA and OSA.
• Both CF and MI provided complementary and meaningful information that supports clinical findings.
• A classification model with two hidden layers achieved very low absolute error, utilizing the average area of CF curves (0-10 Hz) and average MI values as features.

Conclusions:

• EEG synchronization analysis using CF and MI is a valuable tool for differentiating sleep apnea types (CSA, OSA) from controls.
• The identified features (CF curve area, MI values) demonstrate high classification accuracy.
• Future research can focus on integrating these features into a single metric for potentially error-free apnea classification.