SCREENING OF OBSTRUCTIVE SLEEP APNEA BY RR INTERVAL TIME SERIES USING A TIME SERIES NOVELTY DETECTION TECHNIQUE A. P. Lemos, C. J. Tierra-Criollo and W. M. Caminhas Department of Electrical Engineering, Universidade Federal de Minas Gerais, Antonio Carlos Av., 6627, Belo Horizonte, Brazil Keywords: Obstructive sleep apnea, RR interval time series, Time series novelty detection. Abstract: This work proposes a methodology to screen obstructive sleep apnea (OSA) based on RR interval time series using a time series novelty detection technique. Initially, the RR interval is modeled using an autoregressive model. Next, for each data point of the time series, the model output, ˆ x(t ), is compared with the observed value, x t , and the prediction error is generated. The prediction error is then processed in order to detect novelties. Finally, the novelties detected are associated with apnea events. This methodology was applied to the Computers in Cardiology sleep apnea test data and correctly classified 29 out of 30 cases (96.67%) of both OSA and normal subjects, and correctly identified the presence of apnea events in 14078 out of 17268 minutes (81.53%) of the test data set. 1 INTRODUCTION Obstructive sleep apnea (OSA) is a sleep disorder characterized by pauses in breathing during sleep with a reported prevalence in 4% in adult men and 2% in adult women (Young et al., 1993). Obstructive sleep apnea is associated with increased risks of high blood pressure, myocardial infarction, stroke, and with in- creased mortality rates. According to the (AASM, 1999) patients are diag- nosed with OSA if they have 5 or more events of ap- nea per hour of sleep during a full night sleep period. Each event is characterized by a respiratory pause dur- ing 10 seconds. The definitive diagnosis of OSA is made by polysomnography (PSG). PSG is a multi-parametric test based on brain electrical activity (EEG), eye and jaw muscle movement, leg muscle movement, air- flow, respiratory effort (chest and abdominal excur- sion), electrocardiography (ECG) and oxygen satura- tion. This exam is expensive and requires the patient to spend the night in the hospital. In (Guilleminault et al., 1984) is reported that OSA can be characterized by cyclical variations on RR interval time series caused by progressive brady- cardia, followed by abrupt tachycardia on resumption of breathing. This events are highly nonlinear and non stationary. Figure 1 illustrates a RR interval time se- ries in two distinct time intervals, the first one, with no apnea events and, the second one, with these events. If an automatic method is developed to screen the pathology using ECG monitoring instead of PSG, this can be done on basis of a portable and inexpensive device from patient home. This paper proposes a methodology to detect OSA from RR interval time series based on a novelty detec- tion technique. The normal behavior of a system can be characterized by a series of observations through the time. The problem of novelty detection consists in finding time periods where some characteristic of the monitored system has been changed. An autoregressive model is used to model the RR interval time series using a subset without ap- nea events. For each data point of the time series, the model output is compared with the observed value and the prediction error is generated. The prediction error is then processed in order to detect novelties. Fi- nally, the novelties detected are associated with apnea events, since based on information given by (Guillem- inault et al., 1984), this events are nonlinear and non stationary. This paper is divided as follows: in section 2 the RR interval time series is preprocessed in order to be modeled using an autoregressive model. Next, in sec- tion 3 the time series novelty detection technique is presented. In section 4 this technique is applied on Computers in Cardiology sleep apnea dataset (Gold- berger et al., 2000) in order to detect OSA. Finally, section 5 presents conclusions and suggestions for further research. 570 P. Lemos A., J. Tierra-Criollo C. and M. Caminhas W. (2008). SCREENING OF OBSTRUCTIVE SLEEP APNEA BY RR INTERVAL TIME SERIES USING A TIME SERIES NOVELTY DETECTION TECHNIQUE. In Proceedings of the First International Conference on Bio-inspired Systems and Signal Processing, pages 570-575 DOI: 10.5220/0001067505700575 Copyright c  SciTePress