Forehead and In-Ear EEG Acquisition and Processing: Biomarker Analysis and Memory-Efficient Deep Learning Algorithm...

Area of Science:

• Biomedical Engineering
• Neuroscience
• Signal Processing

Background:

• Wearable electroencephalography (EEG) systems offer non-invasive, continuous sleep monitoring outside clinical settings.
• Advancements in EEG technology and feature extraction enable portable sleep analysis.
• Current methods often require complex setups, limiting home-based applications.

Purpose of the Study:

• To develop and evaluate an EEG-based acquisition system for sleep staging adaptable for wearable applications.
• To identify and validate a robust feature set for sleep stage classification from a single EEG derivation.
• To assess the feasibility of a deep learning model for accurate sleep staging in unobtrusive monitoring systems.

Main Methods:

• Utilized a custom experimental setup with the ADS1299EEG-FE-PDK evaluation board for EEG signal acquisition.
• Extracted time, frequency, and non-linear domain features, reduced using mRMR and PCA.
• Trained a two-step deep learning model (LSTM and dense layers) with attention and augmentation for 5-class sleep stage classification on the BOAS dataset.

Main Results:

• Achieved high overall accuracies of 93.5% and 94.7% with reduced feature sets (94% and 98% cumulative explained variance).
• Attained an accuracy of 97.9% using the complete feature set.
• Demonstrated reliable sleep stage classification using a single frontal EEG derivation (F4-F3).

Conclusions:

• A single EEG frontal derivation is sufficient for reliable sleep stage classification.
• The developed system is feasible for unobtrusive, home-based sleep monitoring.
• Wearable EEG systems can significantly advance sleep disorder diagnosis and management.