Breathing Segmentation for Per Breath Sleep Apnea Detection

Area of Science:

• Medical Devices
• Biomedical Engineering
• Sleep Medicine

Background:

• Sleep apnea is a prevalent sleep disorder with breathing cessations.
• Current expert scoring methods for sleep apnea suffer from inter- and intra-rater variability.
• Automated detection methods are needed to improve diagnostic consistency.

Purpose of the Study:

• To develop and validate a novel method for segmenting individual breaths from polysomnography data.
• To classify each segmented breath as obstructive apnea, central apnea, hypopnea, or normal.
• To enhance the physiological modeling of sleep-disordered breathing through per-breath analysis.

Main Methods:

• A Hilbert transform-based algorithm was developed to segment individual breaths from polysomnography recordings.
• A classification model was trained on data from 1,315 subjects in the MESA sleep study.
• The model performed per-breath classification of respiratory events.

Main Results:

• Individual breaths were reliably segmented from polysomnography data.
• The classification model achieved an F1 score of 0.95 for normal breathing.
• Apneic events were detected with an F1 score of 0.47.

Conclusions:

• The proposed method enables reliable breath segmentation and classification for sleep apnea detection.
• Per-breath analysis offers a novel approach for understanding sleep-disordered breathing pathophysiology.
• This detailed analysis may lead to improved personalized treatment strategies and outcome prediction for sleep apnea.