Breathing Segmentation for Per Breath Sleep Apnea Detection.

Sleep apnea is a common disorder characterized by episodes of complete (apnea) or partial (hypopnea) cessation of breathing during sleep. Apneas may result from upper airway obstruction (obstructive apnea) or reduced respiratory effort (central apnea). Although expert scoring is the clinical standard, significant inter-and intra-rater variability has spurred the development of automated detection methods. In this work, we propose a method to segment respiratory modalities from nocturnal polysomnography into individual breaths and to classify each breath as obstructive apnea, central apnea, hy-popnea, or normal. Using recordings from 1,315 subjects in the MESA sleep study, we developed a Hilbert transform-based segmentation algorithm to segment individual breaths from polysomnography data, then we trained a model for per-breath apnea classification. Our results demonstrate that individual breaths can be reliably segmented and that apneic events are detected with an F1 score of 0.47, while normal breathing is classified with an F1 score of 0.95.Clinical relevance: By using breath-specific windowing and thus providing apnea features on a per-breath basis, this approach enables interesting physiological modeling of sleep-disordered breathing. Such detailed analysis may yield novel insights into the underlying pathophysiology of sleep apnea, potentially leading to improved, personalized treatment strategies and better prediction of clinical outcomes.