Akademik Veri Yönetim Sistemi
CHAOS SOLITONS & FRACTALS, cilt.144, 2021 (SCI-Expanded)
Sleep scoring in psychiatry is one of the important issues for classification of diseases, investigation of sleep patterns and also appropriate treatment practices. Sleep specialists usually perform manual sleep stage scoring by visually examining the patient's neurophysiological signals collected in sleep laborato-ries. This is not only very difficult but also very time consuming and easing to make mistake task. Sleep spindles are characteristic electroencephalogram (EEG) signatures of the sleep of stage-2 non-rapid eye movement. Spindles taking place in sleep regulation and cognitive functions may represent biomarkers of a neuropsychiatric disease. Sleep spindles, which are considered to have an important role in enhancing sleep duration, are characteristic waveforms seen in the non-REM sleep phase. The detection and analysis of the waveforms of the spindles, as well as determining the time and duration of the spindles, are very important to fully understand the process of sleep. However, as the spindles have transient properties compared to background EEG signals, it becomes difficult to analyze and distinguish them visually. There are automatic spindle detection methods (ASDM). But there is no consensus on the usage of these au-tomated methods since they are not robust especially in different characteristics of the spindles. In this study, raw sleep EEG data is taken as a time-series data and it is used without any pre-processing (with-out filtering). Adaptive extended Kalman filter (AEKF), a recursive estimation method, is used to separate this time series data into trend, cycle and drift components. In the zones where the psychiatrist or an expert in sleep scoring specifies the occurrence of the spindles, the trend and cycle values obtained by using AEKF are also indicating the occurrences of the spindles very clearly in the same zones. Also, the estimation of the cycle gives some information about the starting point of the spindle. Just by viewing both raw data and trend and cycle components, the expert can easily recognize whether there is a spindle or not. Therefore, the main contribution of this article is to suggest a more effective methodology for de-tecting spindles. The estimation results support the methodology used in this article. Such a finding also shows that the nonlinear state-space model and the AEKF can be properly applied in similar frameworks. In particular, this study separates EEG signals into trend and cycle components, where the methodology allows for system parameters and the components to be estimated simultaneously. Since the methodol-ogy allows for the parameters to vary over time, observing the time patterns of the estimated parameters have the potential to reveal further information about the sleep process.