Paradoxical sleep stages detection using somnographic EOG signal for obese and no-obese patients

摘要

The electroencephalographic (EEG) signal is crucial for the classification of sleep stages. Despite it is widely used for the analysis of sleep, several drawbacks were noted. If the EEG is contaminated by artefacts, no classification can be performed. In this context, we propose an alternative method based on using the electrooculographic (EOG) signal for automatic classification of paradoxical sleep stages. Our classification strategy is composed of three phases: a pre-processing phase to remove the different types of artefacts which contaminate the somnographic EOG signal, a descriptors extraction phase and an automatic detection of REM sleep stages phase. We tested our approach on the somnographic EOG signals from PHYSIOBANK database. Our experimental results present the interest of this method. In fact, we reached a total classification accuracy of 93.28% compared to expert's results that have used the four polysomnographic signals. Our classification results are related to BMI (Body Mass Index) of patients: when BMI augments the classification accuracy decrease.