Automatic Sleep Stage Classification Using Nasal Pressure Decoding Based on a Multi-Kernel Convolutional BiLSTM...

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Artificial Intelligence

Background:

• Polysomnography is the gold standard for sleep studies but is burdensome due to multiple sensors.
• Sleep disorders are prevalent and impact overall health, necessitating accessible diagnostic tools.
• Accurate sleep stage classification is crucial for understanding sleep quality and diagnosing disorders.

Purpose of the Study:

• To develop a simplified sleep stage classification method using only nasal pressure data.
• To investigate the efficacy of a deep learning model for classifying sleep stages.
• To enhance the clinical applicability of sleep analysis by reducing complexity.

Main Methods:

• A deep learning model combining multi-kernel convolutional neural networks and bidirectional long short-term memory was proposed.
• Sleep stages (3-class and 4-class) were classified from nasal pressure recordings of 25 healthy subjects.
• A leave-one-subject-out cross-validation strategy was employed for model evaluation.

Main Results:

• The model achieved 70.4% accuracy and a 0.490 F1-score for 3-class classification (wake, REM, non-REM).
• For 4-class classification (wake, REM, light, deep sleep), accuracy was 60.4% and F1-score was 0.349.
• Performance metrics surpassed those of four comparative models, demonstrating the viability of nasal pressure-based classification.

Conclusions:

• Sleep stage classification is feasible using solely nasal pressure recordings and a deep learning approach.
• This simplified method offers high clinical potential for widespread use in sleep disorder assessment.
• The findings suggest a practical tool for interventions targeting sleep-related diseases.