SleepHybridNet: A Lightweight Hybrid CNN-Transformer Model for Enhanced N1 Sleep Staging From Single-Channel EEG.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Accurate identification of non-rapid eye movement stage 1 (N1) sleep is crucial for sleep neuroscience and clinical practice.
• Current deep learning models face challenges in N1 sleep classification due to ambiguous signal features.

Purpose of the Study:

• Introduce SleepHybridNet, a lightweight hybrid Convolutional Neural Network (CNN)-Transformer model for enhanced N1 sleep stage classification.
• Improve the accuracy and generalization capability of sleep stage classification using single-channel electroencephalogram (EEG) signals.

Main Methods:

• Developed SleepHybridNet, integrating a Multi-Scale Convolutional Neural Network (MSCNN) module and a Transformer encoder.
• Incorporated spectral feature extraction and a multi-task classifier within the novel architecture.
• Utilized the publicly available Sleep-EDF Expanded dataset for model training and validation.

Main Results:

• SleepHybridNet achieved an overall accuracy of 88.2% and an F1-score of 0.633 for N1 sleep stage classification.
• Demonstrated superior performance, especially for underrepresented N1 and N3 sleep stages, outperforming existing methods.
• The model's lightweight design (5.1M parameters) facilitates practical deployment in clinical settings.

Conclusions:

• SleepHybridNet offers a promising solution for accurate and efficient N1 sleep stage classification from single-channel EEG.
• The model's performance and lightweight nature bridge the gap between advanced deep learning and clinical applicability in sleep medicine.
• Future research may involve integrating multimodal data from wearable sensors for broader applications.