OSA-weigher: an automated computational framework for identifying obstructive sleep apnea based on event phase segmentation

Abstract

Obstructive sleep apnea (OSA) is one of the most common sleep-related breathing disorders, which causes various diseases and reduces life quality severely. In this paper, we propose OSA-Weigher, an automated computational framework that can improve the performance of identifying OSA events. Particularly, the key idea of OSA-Weigher is to subdivide each potential event segment (PES, i.e., a data segment that may or may not contain an OSA event) and to explore more information of respiratory pattern, so as to improve OSA events identification performance. Concretely, we utilize a micro-movement sensitive mattress (MSM) to get ballistocardiography (BCG) signal during sleep, and locate PESs by identifying the occurrence of arousals (i.e., a mechanism that makes patients recover from being apneic). Afterwards, we divide each PES into three phases (i.e., Apnea Phase, Respiratory Effort Phase and Arousal Phase) using a sliding window-based adaptive method. Based on these phases, we further extract and select efficient fine-grained features to characterize respiratory pattern from multiple aspects. Finally, these PESs are classified into OSA events or non-OSA events by employing an optimized ensemble classifier. Experimental results based on a real BCG dataset of 116 subjects show that OSA-Weigher outperforms the baseline method by 12.7% in terms of Precision, 14.8% in terms of Recall and 0.152 in terms of AUC (area under ROC curve).