Automatic detection of sleep apnea and hypopnea events from single channel measurement of respiration signal employing ensemble binary SVM classifiers

作者：Koley Bijoy Laxmi^*; Dey Debangshu

来源：Measurement, 2013, 46(7): 2082-2092.

DOI：10.1016/j.measurement.2013.03.016

• Event detection
• Recursive feature elimination
• Respiration signal
• Sleep apnea
• Support vector machine

This paper presents a novel method for automatic detection of apnea and hypopnea events as well as mean duration of events from the recording of single channel oronasal airflow signal, moreover the automated algorithm has been implemented with PC based low cost Data Acquisition System (DAS). The method divides the respiration signal into overlapping segments of typical 8 s duration and then categorize the segments with the help of ensemble binary Support Vector Machine (SVM) classifiers, according to the origin of the segments, i.e. %26apos;N%26apos; if the segment originates from normal respiration signal during sleep, %26apos;A%26apos; if it originates from apnea and %26apos;H%26apos; for hypopnea event related breathing signal. Finally, it uses a heuristically derived rule based system to identify the apnea or hypopnea events by combining the time sequenced decisions of the classifiers. Automatic identification of events helps to provide the direct estimation of Apnea Hypopnea Index (AHI) and thus severity. The overall correlation coefficients between the automatic model predicted indexes and the PSG based manual indexes were 0.970, 0.986 and 0.982 for HI, AI, and AHI respectively.

• 出版日期2013-8