Automatic Sleep Staging in Children with Sleep Apnea using Photoplethysmography and Convolutional Neural Networks

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Pediatric Sleep Medicine

Background:

• Pediatric obstructive sleep apnea (OSA) diagnosis relies heavily on polysomnography (PSG), which is complex and costly.
• Alternative methods for sleep staging are needed, especially for home-based testing.
• Photoplethysmography (PPG) signals contain valuable data on autonomic nervous activity linked to sleep stages.

Purpose of the Study:

• To investigate the use of PPG signals and deep learning for automatic sleep staging in children.
• To develop and validate a convolutional neural network (CNN) model for classifying wake, REM, and non-REM sleep stages.
• To assess the performance of PPG-based CNN sleep staging against PSG in pediatric OSA patients.

Main Methods:

• A database of 366 PPG recordings from pediatric OSA patients was utilized.
• A CNN architecture was trained on 30-second epochs of PPG data for three-stage sleep classification (Wake/NREM/REM).
• The CNN model's performance was evaluated on an independent test set.

Main Results:

• The CNN model achieved 78.2% accuracy and a Cohen's kappa of 0.57 for W/NREM/REM classification.
• The percentage of time spent in the wake stage calculated by the model showed no significant difference compared to manual PSG scoring.
• The results surpassed previous studies analyzing PPG for automated sleep staging in children with OSA.

Conclusions:

• CNNs combined with PPG signals show potential for accurate sleep stage scoring in pediatric home sleep apnea tests.
• This approach offers a more accessible and less intrusive method for sleep analysis in children.
• PPG-based deep learning presents a viable alternative to traditional PSG for certain aspects of pediatric sleep assessment.