Preliminary Study on the Detection of Apnea Episodes Through the Use of Dictionaries
Silvia González, José Ramón Villar, Javier Sedano, Joaquín Terán, María Luz Alonso Álvarez and Jerónimo González
Sleep apnea is a respiratory disorder that affects a very significant number of patients of different ages. One of the main consequences of suffering apnea is a higher likelihood of stroke onset. Sleep apnea is identified as alterations to the breathing rate of the individual while asleep, due to obstruction of the respiratory tract. These alterations produce abnormal chest movements that can be measured by using accelerometers. In recent studies, the use of simple heuristics to determine the time between inhalation and exhalation has been reported using preset thresholds to determine inhalation/exhalation events. In this study, two different approaches based on SAX Time Series representation are proposed. The first approach identifies the time intervals between respiratory events, while the second approach compiles a dictionary of normal movements while asleep. A dictionary is created for each independent posture. Abnormal movements are detected by means of the SAX distances between the shortened words. Experiments are based on the same realistic data set taken from previous studies in the literature. The results show that the small windows-based detection algorithm could be a suitable approach for simplifying the implementation and the scalability of apnea problems. Further work is needed for accurate automated setting of the thresholds.
Keywords: Apnea episode detection, pattern recognition, SAX, bag of words, accelerometers, respiratory rate