Title :
Automated detection of apnea and hypopnea events
Author :
Koley, Bikash ; Dey, Debabrata
Author_Institution :
Appl. Electron. & Instrum. Eng. Dept, Dr. B. C. Roy Eng. Coll., Durgapur, India
fDate :
Nov. 30 2012-Dec. 1 2012
Abstract :
This paper presents an automatic method for detection of apnea and hypopnea events occurred during sleep from the single channel recording of oronasal airflow signal. For the identification of events, three time domain measures were extracted from each of the overlapping short segment windows of respiration signal. The feature set includes area, upper 90th percentile and variance, which were used to characterize changes in the airflow signal during normal and abnormal breathing events (i.e., apnea, hypopnea). An ensemble of three binary Support Vector Machine (SVM) based classifiers arranged in one-against-all strategy, were used to classify the feature vector among three categories, according to its origin from some breathing events like normal, apnea and hypopnea. The consecutive decisions of classifier model on time sequenced consecutive overlapped windows were combined by some heuristic rules to identify abnormal breathing events from normal breathings. In this study, 14 polysomnography (PSG) recordings diagnosed as obstructive sleep apnea syndrome were analyzed. Independent test was performed on 6 recordings. The cross-validation and independent test accuracies of apneic event detection were found to be 93.3% and 92.8%, respectively. For hypopnea event these two accuracies were 90.1% and 89.6%. The proposed system can be used for home based monitoring of suspected apneic subject, and can count total number of apnea and hypopnea events occurred during sleep.
Keywords :
bioelectric phenomena; feature extraction; medical diagnostic computing; medical disorders; medical signal processing; pneumodynamics; signal classification; support vector machines; PSG; SVM; abnormal breathing events; airflow signal; area feature set; automated apnea event detection; automated hypoapnea event detection; binary support vector machine based classifiers; electrophysiological signals; event identification; home based monitoring; normal breathing events; obstructive sleep apnea syndrome; overlapping short segment windows; polysomnography recordings; respiration signal; single channel oronasal airflow signal recording; sleep apnea; sleep disorder; suspected apneic subject; time domain measures; time sequenced consecutive overlapped windows; upper 90th percentile feature set; variance feature set; Accuracy; Event detection; Sleep apnea; Support vector machine classification; Training; Apnea-hypopnea event detection; respiration signal; support vector machine;
Conference_Titel :
Emerging Applications of Information Technology (EAIT), 2012 Third International Conference on
Conference_Location :
Kolkata
Print_ISBN :
978-1-4673-1828-0
DOI :
10.1109/EAIT.2012.6407868