SleepEEGNet

SleepEEGNet performs automated sleep stage scoring from single-channel EEG using deep convolutional neural networks and sequence-to-sequence modeling to provide objective sleep staging for clinical and research use.

Key Features:

• Single-Channel EEG Processing: Processes single-channel EEG signals including Fpz-Cz and Pz-Oz from the Physionet Sleep-EDF datasets.
• Deep Convolutional Neural Networks (CNNs): Uses deep CNN architectures to extract time-invariant features and frequency information from EEG signals.
• Sequence-to-Sequence Modeling: Incorporates sequence-to-sequence modeling to capture long-term dependencies across sequential sleep epochs.
• Class Imbalance Mitigation: Employs novel loss functions during training to mitigate class imbalance and equalize misclassification errors across sleep stages.

Scientific Applications:

• Clinical Sleep Staging: Automates sleep stage scoring to support diagnosis and assessment of sleep disorders in clinical settings.
• Sleep Research: Provides standardized EEG-based sleep stage annotations for research into sleep patterns and physiology.
• Benchmarking: Supplies performance metrics for method comparison, reporting overall accuracy 84.26%, macro F1-score 79.66%, and Cohen’s kappa (κ) 0.79.

Methodology:

Training on single-channel EEG from the Physionet Sleep-EDF datasets using deep CNNs for feature extraction, sequence-to-sequence modeling for temporal dependencies, and novel loss functions to address class imbalance.