Sleep Stage Classification of Pediatric Patients with Sleep-Disordered Breathing using Airflow Signals

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Machine Learning

Background:

• Sleep staging is crucial for assessing sleep health.
• Breathing variability during sleep may indicate specific sleep stages.
• Current methods often rely on electroencephalogram (EEG) signals, which are not always available.

Purpose of the Study:

• To evaluate the efficacy of machine learning models in classifying sleep stages using only airflow signals in pediatric patients.
• To determine how well airflow signals can predict sleep stages, particularly in children with varying degrees of sleep-disordered breathing.
• To explore the potential of airflow signals as a surrogate for EEG in sleep staging.

Main Methods:

• Trained machine learning models on sleep airflow signals from pediatric subjects.
• Performed a 3-class classification task: Wake, NREM, and REM sleep.
• Evaluated model performance across different severity levels of sleep-disordered breathing, defined by Apnea-Hypopnea Index (AHI).

Main Results:

• Achieved 66.9% balanced accuracy (BA) for healthy subjects.
• BA decreased with increasing severity of sleep-disordered breathing (61.4% for mild, 58.0% for moderate, 48.9% for high AHI).
• The model achieved >60% BA for classifying sleep stages during apnea/hypopnea events in subjects with AHI < 5, indicating robustness.

Conclusions:

• Airflow signals show potential as a viable alternative for sleep staging when EEG is unavailable.
• The machine learning approach is affected by, but remains robust to, apnea and hypopnea episodes.
• This study supports the clinical relevance of airflow signals for sleep staging in both healthy children and those with sleep-disordered breathing.