SSC-SleepNet: A Siamese-Based Automatic Sleep Staging Model With Improved N1 Sleep Detection.

Area of Science:

• Artificial Intelligence
• Neuroscience
• Biomedical Engineering

Background:

• Automatic sleep staging from electroencephalography (EEG) is an AI-driven alternative to manual polysomnography.
• Current AI models struggle with accurate N1 sleep stage detection due to its rarity and ambiguity.

Purpose of the Study:

• To develop an AI algorithm, SSC-SleepNet, to enhance N1 sleep stage detection in automatic sleep staging.
• To improve the learning capability for minority sleep stages using a novel adaptive loss function.

Main Methods:

• Proposed SSC-SleepNet, a pseudo-Siamese neural network with two branches (Squeeze-and-Excitation residual network and CNN-LSTM).
• Employed an adaptive loss function combining weighted cross-entropy and focal loss to address class imbalance.
• Validated on four public datasets: Sleep-EDF-SC, Sleep-EDF-X, SHHS, and HMC.

Main Results:

• SSC-SleepNet achieved high overall macro F1-scores across datasets (84.5%-89.6%).
• Significantly improved N1 sleep stage F1-scores (55.2%-60.2%) compared to state-of-the-art models.
• Demonstrated superior performance in automatic sleep staging using single-channel EEG signals.

Conclusions:

• SSC-SleepNet effectively improves automatic sleep staging, particularly for the challenging N1 sleep stage.
• The proposed deep learning architecture and adaptive loss function offer a promising approach for single-channel EEG sleep analysis.
• This advancement has implications for more accessible and efficient sleep disorder diagnosis.